Awesome Opencode

  

     

A curated list of plugins, themes, agents, and resources for Opencode.

The AI coding agent for the terminal, built by the team at Anomaly.

Note

Active Hard Fork: Heretek-AI/awesome-opencode is the actively maintained community continuation of awesome-opencode. All original upstream issues, pull requests, and discussions are preserved and imported. Submissions and PRs are actively reviewed!

528 entries  •  271 plugins: 135 V2, 124 V1 only (removed after 2027-01-01), 12 unverified

🔎 Search & filter the catalog  •  ➕ Submit an entry  •  📋 Catalog policy

OFFICIAL • NEW • PLUGINS • THEMES • AGENTS • PROJECTS • RESOURCES

---

⭐️ Official Repositories

Project	Stars	Description
opencode		The official opencode AI coding agent.
opencode-sdk-js		Official JavaScript/TypeScript SDK for opencode.
opencode-sdk-go		Official Go SDK for opencode.
opencode-sdk-python		Official Python SDK for opencode.

🆕 Recently Added

Name	Category	Tagline
handoff	agents	Delegate tasks to OpenCode / DeepSeek V4 right inside your Claude Code / Codex sessions.
Reflection-3	projects	Judge layer that catches premature agent stops
crowdsource-debugger	agents	Debug stubborn errors with community help
opencode-mcp-registry	agents	Register MCP servers in opencode
sensitive-operations	agents	Gatekeeper for privileged operations
uv-skill	agents	Use uv for all Python package management
Prompt to Asset	projects	Generate visual assets across 30+ image models via MCP
Emdash Skills	agents	14-category autonomous product-building OS
MorningAI	agents	AI news tracking skill
opencode-telegram	projects	Control OpenCode sessions from Telegram
MonadicSharp	projects	Railway-Oriented Programming for C#
AgenTopology	projects	Declarative multi-agent topology language
Untether	projects	Telegram bridge for remote task control
PPTX from Layouts	agents	PowerPoint from slide master layouts
Promption	projects	Manage AI agent configs

🧩 PLUGINS (271)

Memory ₍₂₇₎

AccInt V1 only - Local Work Model memory for OpenCode agents

Accreted scored memory for OpenCode sessions with MCP retrieval and action verbs, lifecycle hooks, and cross-project acc.db continuity for agent work.

🔗 View Repository

Agent Memory V2 - Persistent, self-editable memory and a local semantic journal

Maintained OpenCode plugin providing shared global and project memory blocks, dedicated memory tools, and an opt-in append-only journal with local multilingual semantic search.

🔗 View Repository

basemind V2 MCP - Full AI context layer over MCP — code-map, document RAG, agent memory, web crawl, git history

Full AI context layer over MCP — tree-sitter code-map, document RAG, shared agent memory, web crawl, and git history. 300+ languages. Published to npm as basemind-opencode (https://www.npmjs.com/package/basemind-opencode).

🔗 View Repository  •  📦 npm

codemem V2 - Persistent coding memory across sessions, machines, and teammates

Captures coding decisions, discoveries, and fixes, with automatic context injection in OpenCode, Claude Code, and Codex. Includes MCP tools for searching and managing memories, a viewer, optional cross-machine sync, and sharing of selected projects with teammates.

🔗 View Repository  •  📖 Docs

Daily Journal V2 - Auto-journal every conversation into daily/YYYY-MM-DD.md, record-only (never injects)

An OpenCode V2 plugin that keeps a daily journal of your conversations. When a session turn finishes, the exchange is summarized into a diary entry and appended to daily/YYYY-MM-DD.md. Record only, never inject: entries are plain Markdown on disk and are never added to session context unless you explicitly ask the agent to read them. Writes are concurrency-safe (file lock + append-only + idempotent markers, so simultaneous sessions never overwrite or duplicate each other), and summaries use a model fallback chain (session model, then other configured models, then a raw excerpt) so nothing is lost when a model is unavailable.

🔗 View Repository

deja V2 - Recall what you already did in your other coding agents

opencode remembers its own sessions. deja answers the other question: what you did in Claude Code, Codex, Cursor, Gemini CLI, Zed and fifteen more agents on this machine — including the months before you installed anything.

A local Go binary indexes the transcripts those agents already wrote to disk, and the plugin gives the model six tools over that index: recall past sessions by error or symbol, read the digest of one session, see which sessions discussed a file before you edit it, ask what this machine actually ran after a given error, and store a decision worth keeping. Recall also arrives unasked — once per session on the system prompt, and a relevance pass per prompt that stays silent when history has no answer.

BM25 over local files. No LLM, no embeddings, no network. Credentials are redacted at index time.

🔗 View Repository  •  📦 npm

Ditto unverified - Mines your local AI coding session logs into a personal profile your agent loads before every task

Mines local AI coding session logs (OpenCode, Claude Code, Codex, Copilot CLI) into a personal working profile. The OpenCode adapter reads OpenCode's SQLite session store and installs the profile into OpenCode's global rules so the agent loads it before every task. Single stdlib-only Python file, runs locally, MIT licensed.

🔗 View Repository

Ejentum unverified MCP - MCP server with reasoning, code, anti-deception, and memory tools for AI agents

MCP server with four tools (harness_reasoning, harness_code, harness_anti_deception, harness_memory) that AI agents can call on demand. Each tool returns a structured prompt the calling agent ingests before generating.

🔗 View Repository  •  📖 Docs

Episodic Memory V2 - Semantic search over your past conversations — local embeddings, no cloud

Indexes your OpenCode session history into a local SQLite store and lets the agent recall past discussions, decisions, and patterns via native episodic_search/episodic_read tools. Fully local and offline — embeddings run in-process via Transformers.js (Snowflake arctic-embed, model chosen by empirical eval), with vector, BM25, and hybrid RRF search modes. Auto-reindexes on session idle; opt-out marker for sensitive chats. Inspired by obra/episodic-memory for Claude Code, rebuilt natively for OpenCode primitives.

🔗 View Repository

Harness Memory V1 only - Persistent project memory - 73 percent fewer tokens than CLAUDE.md, with human review

Auto-captures evidence from tool interactions and materializes memories through a multi-gate pipeline. 4-layer activation engine selects the right memories per context. Replaces CLAUDE.md with structured, searchable, reviewable project memory. Local-first (sql.js WASM), zero cloud dependency.

🔗 View Repository

Honcho V2 - AI-native long-term memory for OpenCode

Give OpenCode persistent memory that survives context wipes, session restarts, and fresh chats. Honcho remembers what you're working on, durable preferences, and prior context across projects. Supports cloud and self-hosted deployments with configurable session strategies.

🔗 View Repository

Lemma unverified MCP - Persistent memory layer for LLMs via MCP - local, zero-dependency, works on all clients

Biological memory model for AI coding agents. Features confidence decay/boost, Fuse.js fuzzy dedup, guide system with usage tracking, cross-references, cumulative backup, virtual session tracking, and universal memory injection via tool descriptions (works on Claude Desktop, Cursor, VS Code, Gemini CLI, opencode, and any MCP client). No API keys, no cloud, fully local JSONL storage. 20 MCP tools, 110 tests, MIT licensed.

🔗 View Repository

Lore V1 only - Long-term memory for OpenCode with boot baseline and per-prompt recall

Native plugin that connects OpenCode to a Lore memory server. A fixed boot baseline loads identity, workflow and preference memories into the system context at session start, recall surfaces relevant memories before each prompt, and durable facts are written back through guarded lore_* tools to a URI-addressed memory graph shared with Claude Code, Codex and other runtimes. Self-hosted with Docker or use the managed Loremem service.

🔗 View Repository  •  📖 Docs

Mem0 OpenCode Plugin V1 only - Persistent memory plugin so your agent remembers across sessions

OpenCode plugin that gives your agent persistent memory powered by Mem0. Includes 9 MCP memory tools, 16 slash commands, and lifecycle hooks that auto-capture decisions, search prior context at session start, and pre-fetch error memories on failures. Install with `opencode plugin @mem0/opencode-plugin`.

🔗 View Repository

oc-mnemoria V1 only - Persistent shared memory (hive mind) for OpenCode agents across sessions

Gives all OpenCode agents a shared persistent memory store where every agent can read and write. Each entry is tagged with the creating agent (plan, build, ask, review) so no context is lost between roles. Powered by the mnemoria Rust engine with hybrid BM25 + semantic search, CRC32 checksum chains, and an append-only binary format. Includes 7 tools and /mn-* slash commands.

🔗 View Repository

oh-my-agentmemory V2 - Forces agents to actually write to agentmemory, not just read from it

Companion plugin to agentmemory-capture. Re-injects a memory policy directive into the system prompt every turn so the agent actually calls memory_save, bootstraps empty pinned slots on session start, catches bilingual intent keywords (EN/KR), auto-saves lessons from error-signal file edits, and suggests crystallizing finished work. Installs from npm with one line in opencode.json.

🔗 View Repository

OpenCode Claude Memory V2 - Claude Code-compatible memory

Share persistent Markdown memory between OpenCode and Claude Code using Claude Code-compatible paths and file formats.

🔗 View Repository

Opencode Mem V2 - Persistent local memory with Turso/libSQL vector search

A persistent memory system for AI coding agents that enables long-term context retention across sessions using embedded Turso/libSQL vector search. Features dual memory scopes, auto-capture, user profile learning, a full web UI, and multi-provider AI support.

🔗 View Repository

Opencode MemPalace Persistence V2 - Local-first persistent memory for OpenCode, backed by MemPalace

Saves every conversation to a local MemPalace palace and recalls it verbatim in later sessions. Mines run on idle, exit and startup — never mid-reply. Zero-config: the plugin registers its own read-only MCP entry, so multiple OpenCode tabs never starve each other on the palace writer lock. Ships a live sidebar status line and a companion local browser/curator (penfield) for timeline, knowledge graph, thread view and duplicate healing.

🔗 View Repository  •  📦 npm

opencode-short-term-memory V1 only - Maintain user instructions and preferences throughout long sessions. No more repeating yourself.

Automatically summarizes conversation context into structured session memory and injects it back into the system prompt every few turns — preserving user instructions, project context, decisions, and active references across long chats and compactions.

🔗 View Repository

Simple Memory V2 - Git-based memory

Simple plugin to manage memory inside a git repo that can be committed and reviewed by team members.

🔗 View Repository

Solo Memory V1 only - Transparent long-term memory — auto weave/ingest, no agent discipline required

A transparent long-term memory plugin for OpenCode. The chat.message hook auto-injects memory context before every turn, and the event hook auto-stores user+assistant pairs after replies — the agent never needs to remember to call memory tools. Local-first (SQLite + ChromaDB, zero cloud), bilingual (Chinese via BM25/jieba, English via semantic bge-large-en-v1.5), with a gap-detection → crawl → synthesize → verify learning loop and 7 MCP tools.

🔗 View Repository

SourcePier V2 - Persistent project documents across OpenCode sessions

Manage project-scoped Markdown documents in the OpenCode v2 native TUI. Import, search, edit, and export Sources across sessions, while agents can discover and read them through read-only tools.

🔗 View Repository

Unforgit V2 - Git-backed repository memory for AI agents

Unforgit gives OpenCode and other coding agents durable, repository-scoped memory through MCP. Store and recall project decisions, conventions, gotchas, and playbooks locally in a Git-backed memory base, with optional sync and curation workflows.

🔗 View Repository  •  📖 Docs

With Context MCP unverified MCP - Project-specific markdown notes

MCP server for managing project-specific markdown notes with templates, batch edits, and ignore patterns.

🔗 View Repository

Worktree Memory Sync V1 only - Auto-sync memory to git worktrees

Automatically copies .opencode/memory/ from the main repository into new git worktrees on session init. Detects worktrees via the .git file, resolves the main repo, and syncs memory files only when the destination is empty — so it never overwrites your changes.

🔗 View Repository

WPM V1 only - Confidence-weighted persistent memory with proofs

Local project memory maintained automatically during work. Every memory carries a 0-1 confidence score that only rises with external proofs; entries are validated or contradicted with evidence, never deleted, and deduplicated. Hybrid semantic + graph retrieval recalls the right memory when it matters.

🔗 View Repository

Context & Tokens ₍₁₃₎

cache-stats V1 only - Showing the active session's cache hit rate in a popup dialog

This plugin shows the overall cache hit rate, a per-model breakdown, a per subagent breakdown and the associated cost, making it easy to understand cache performance at a glance.

🔗 View Repository

Chisle V1 only - Cuts token use with terse answers, smallest-code-that-works rules and tool-output compression

Installs a ruleset for terse answers that keep every decisive fact and a "smallest code that works" ladder, Chisle skills, and a compression plugin that trims oversized tool output before it re-enters the context. The same ruleset ships for Claude Code, Cursor, Codex and Gemini CLI. Install with `npx chisle --only opencode`. MIT, zero dependencies, benchmarks with raw transcripts in the repository.

🔗 View Repository  •  📖 Docs

Context Analysis V1 only - Token usage analysis

An opencode plugin that provides detailed token usage analysis for your AI sessions.

🔗 View Repository

Context Optimizer V2 - Coding Agent Context Optimizer Plugin

Keep your coding agent's context small. When a session gets compacted, Context Optimizer reranks the relevant parts, drops duplicates, and compresses the rest with a local ML pipeline (LLMLingua-2 + Sentence Transformers) — so more of the window stays useful and fewer tokens get billed. Everything runs on your machine.

🔗 View Repository

ContextStream V2 MCP - Shared project context for OpenCode, Cursor, Claude Code, Codex, and Grok via MCP

Hosted MCP server that keeps coding agents on shared project context across sessions and teammates. Works with OpenCode and other MCP-compatible clients. Endpoint https://mcp.contextstream.io/mcp — Intelligence isn’t the bottleneck. Context is.

🔗 View Repository  •  📖 Docs

Dynamic Context Pruning V2 - Optimize token usage

Plugin that optimises token usage by pruning obsolete tool outputs from conversation context.

🔗 View Repository

Magic Context V2 - Lossless context management with background compression

Cache-aware context management that keeps long sessions productive. Background historian compresses old conversation into structured compartments while you keep working. Includes cross-session project memory, unified search across history/memories/facts, overnight dreamer for memory maintenance, and prompt-cache-safe deferred operations.

🔗 View Repository

OpenCode Autocompact V2 - Per-session autocompact — /autocompact 500k compacts a session and its children early

Claude Code's autocompact, with a knob. Compacts a session before the model's real context limit, at a token count you choose per session. OpenCode's built-in auto-compaction is configured globally and only relative to the model limit, so on a 256K or 1M context model it kicks in very late; this adds an earlier, deliberate ceiling for one session and its children, useful for cheaper requests, more headroom for large tool results, or a shorter tail for the summarizer. Thresholds are inherited by subagents by walking the parentID chain, `/autocompact off` reverts to normal behaviour, and with no threshold set the plugin has zero footprint. When a threshold applies, a chip joins the prompt footer and a side panel shows the numbers. Tested against 2.0.24.

🔗 View Repository  •  📦 npm

Opencode Ignore V1 only - Ignore files based on pattern

Plugin to ignore directory/file based on pattern.

🔗 View Repository

OpenCode Stats Pills V2 - Inline TUI stats bar for generation speed, tokens, cache hits, cost, and context usage

A compact stats bar under the OpenCode V2 input box showing streamed output speed (tok/s), average TTFT, token usage with cache hits and reasoning, session cost, and context-window usage, without opening the sidebar. Click any pill for a detail panel. Context usage follows OpenCode's own calculation, custom per-model prices are supported, and parent and subagent sessions are tracked separately. Tested on OpenCode 2.0.29; UI text is currently in Chinese.

🔗 View Repository

opencode-cache-compact V2 - Cache-friendly compaction for locally hosted models

Replaces OpenCode's built-in compaction, which rebuilds the prompt from scratch (no system prompt, no tools, history re-serialized) and therefore cold-prefills the whole conversation. This plugin appends the summary turn instead — a cache hit — and then rewrites only the outgoing message list down to that summary, so a locally hosted model reuses its KV cache instead of re-reading everything. Aimed at llama.cpp / LM Studio / vLLM setups where prefilling a large context is the expensive part.

🔗 View Repository

opencode-snip V1 only - OpenCode plugin that prefixes shell commands with snip to reduce LLM token consumption by 60-90%

Automatically prefixes supported shell commands (git, go, cargo, npm, docker, etc.) with snip to filter output before it reaches your LLM context window.

🔗 View Repository

Topic Compaction V2 - Appends late-position overrides to compaction prompts for topic-based session summarization.

This plugin improves session compaction continuity by restructuring how opencode summarizes file activity. Instead of a single linear stream that can lose track of pending work, the plugin clusters actual file activity into thematic topics during compaction. Each topic carries an explicit `blocked:` status field, and open blockers are guaranteed to survive across successive compaction cycles, ensuring unresolved work persists through session restarts.

🔗 View Repository

Multi-Agent & Orchestration ₍₃₂₎

@bluelovers/opencode-arise V1 only - 「⚔️ ARISE!」　A Solo Leveling themed orchestrator harness for OpenCode

A lightweight, token-efficient orchestrator layer. Enables parallel background task execution in OpenCode. Launch AI agents to work simultaneously on exploration and research while continuing with other tasks. Allows specifying custom models for each_agent via configuration.

🔗 View Repository

Agent Identity V1 only - Agent self-identity and per-message attribution for multi-agent sessions

Two plugins that improve agent identity awareness. AgentSelfIdentityPlugin injects a one-liner into the system prompt so the model knows which agent it's operating as. AgentAttributionToolPlugin exposes a tool for querying per-message agent attribution via the SDK, useful for agents that review multi-agent sessions.

🔗 View Repository

AIDevOps V2 - AI DevOps framework and OpenCode plugin

OpenCode-first AI DevOps framework with agents, skills, workflows, safety hooks, memory, GitHub worker dispatch, and integrations for code, infrastructure, content, SEO, and operations.

🔗 View Repository  •  📖 Docs

AION V2 - Time-series harness with multi-agent review, skills, tools, and hooks

A compiled TypeScript OpenCode plugin that turns OpenCode into a time-series research harness. Bundles six specialized agents (orchestrator, requirements analyst, information collector, coder, and two critic roles), seventeen skills, twenty-plus tools, and eight coordination protocols into one self-contained bundle installed via a one-command CLI. Enforces a four-layer model (task, workspace, execution, review) with hard gates for data-leakage detection, ablation-backed statistical rigor, and dual-critic governance before any output ships.

🔗 View Repository  •  📖 Docs

Background Agents V1 only - Async agent delegation

Claude Code-style background agents with async delegation and context persistence.

🔗 View Repository

CrewBee V1 only - Task-specific Agent Teams for OpenCode

CrewBee is an independent Agent Team framework for OpenCode. It lets users define reusable task/project-specific Agent Teams, project them into OpenCode agents, and switch between single-agent execution and multi-agent collaboration based on task complexity. It also includes built-in Team templates such as a Coding Team with review flow and completion criteria.

🔗 View Repository

FlowDeck V1 only - AI-powered multi-agent workflow orchestration with built-in safety intelligence

FlowDeck adds a structured, multi-agent development workflow to OpenCode. It coordinates 25 specialist agents through a four-phase cycle — discuss, plan, execute, review — with persistent state that survives session restarts.

Key features:

• 25 specialist agents (architect, planner, coder, reviewer, tester, debugger, risk-analyst, policy-enforcer, and more)
• 24 reusable workflow skills (TDD, security scan, deploy check, code review, and more)
• 17 workflow commands for all project operations
• 15 pre-built orchestration flows including Spec-Driven Development (SDD)
• Persistent state via .planning/STATE.md — resume exactly where you left off
• Wave-based parallel execution for independent tasks
• AI Safety layer: patch trust scoring, edit gates, phase gating, arch constraint enforcement, failure replay, and regression prediction
• Deep System Hooks: context monitoring, session idle summaries, shell environment injection
• Built-in MCPs: Context7 (docs), Exa (web search), Grep.app (code search)
• Ensemble Reasoning via /fd-council for synthesized consensus from multiple agents
• Persistent Memory with SQLite for tool executions and session summaries
  
  🔗 View Repository  •  📖 Docs

Froggy V2 - XP-guided multi-agent orchestration

Multi-agent delegation and orchestration guided by XP, KISS, and YAGNI, with specialized agents, independent code review, and a persistent loop to keep work moving toward the objective.

🔗 View Repository

hiai-opencode V1 only - Canonical 12-agent model with bundled skills, MCP, LSP, and ralph-loop

Unified plugin shipping 10 visible agents (Bob, Coder, Strategist, Critic, Guard, Researcher, Designer, Manager, Brainstormer, Vision) plus hidden Sub and Agent Skills. Bundles MCP wiring (playwright, stitch, sequential-thinking, firecrawl, rag, mempalace, context7, websearch, grep_app), LSP, skill materialization, and a multi-layer continuation system (todo enforcer, ralph-loop, ULTRAWORK auto-start) in one install.

🔗 View Repository

Oh My Kiro V2 - AWS Kiro-inspired multi-agent SDD coding agent harness

A spec-driven, multi-agent coding plugin for OpenCode inspired by AWS Kiro. Features requirements-first workflow, spec task execution, context gathering, background tasks, and multimodal support.

🔗 View Repository

Oh My OpenAgent V1 only - Agents & Pre-built tools

Background agents, pre-built tools (LSP/AST/MCP), curated agents, and a Claude Code compatible layer. (formerly Oh My Opencode)

🔗 View Repository

Oh My Opencode Slim V2 - Lightweight agent orchestration with reduced token usage

Slimmed-down fork of oh-my-opencode focused on core agent orchestration. Features specialized sub-agents (Explorer, Oracle, Librarian, Designer, etc.), background task management, LSP/AST tools, tmux integration for live agent visibility, and MCP servers. Optimized to consume significantly fewer tokens.

🔗 View Repository  •  📖 Docs

Open Conclave V1 only - Multi-agent debates, moderated by a captain agent until they reach consensus

In the style of Grok 4.20, multiple agents are dispatched to answer the same query. Each one has a different system prompt (logical, creative, research-focused) and have to reach a consensus for the final output. The user can specify which provider/model they want to use and override the system prompt of each agent via config.

🔗 View Repository

Open Dynamic Workflows V1 only - Dynamic multi-agent workflows for OpenCode — plan, orchestrate, and verify with the script as the orchestrator

An MIT-licensed engine bringing Claude-Code-style dynamic workflows and ultracode to OpenCode. A local daemon runs a generated orchestration script with concurrent agents, adversarial verification, and crash-resume; the OpenCode plugin adds workflow/ultracode triggers and a planning UI. Bring your own model — Anthropic, any OpenAI-compatible endpoint, or local Ollama.

🔗 View Repository

OpenCode Crosstalk V2 - Peer discovery, messaging, and expiring file leases between concurrent OpenCode sessions

Coordination for agents that share a workspace. Six tools let concurrent sessions discover each other, talk to each other, and take exclusive expiring leases on the files they are editing, so two agents never rewrite the same file at once.

• crosstalk_status / crosstalk_peers — declare a role, goal, and working set, then list who is running, what they declared, and what they hold
• crosstalk_send / crosstalk_inbox — message a session, a role, or everyone; mail lands in the recipient's mailbox and is injected into its current turn when the session is live
• crosstalk_claim — atomic lease on files or named resources; a refusal names the holder and when the lease expires
• crosstalk_wait — block until a peer goes idle or a resource is released, instead of sleeping and re-checking

Identity comes from the session rather than from tool input, so no session can release another's lease or read another's mail. All six tools share one permission action (crosstalk), so a project can allow or deny coordination with a single rule. The core coordination logic (presence registry, mailboxes, claim table) is host-free and unit tested — the plugin imports nothing from OpenCode outside its single entry point.

🔗 View Repository

OpenCode Ensemble V2 - Parallel agent teams for OpenCode

Coordinate parallel OpenCode agents with peer messaging, a shared task board, git worktree isolation, and a live dashboard.

🔗 View Repository  •  📦 npm

OpenCode Mesh V2 - The missing agent-to-agent messenger for OpenCode

The free OpenCode plugin that lets your agents communicate directly with any session on your machine. Separate projects, separate terminals, separate models, one connected team.

🔗 View Repository

OpenCode Mission Control V1 only - Command center for parallel agents — worktree isolation, DAG plans, merge train, PRs

Orchestrates parallel OpenCode agents in tmux-isolated git worktrees with live inspection (capture/attach/diff), a dashboard overview, agent status reporting, and in-chat notifications. Supports DAG-based plans with autopilot/copilot/supervisor modes and a merge train with test gating and automatic rollback.

🔗 View Repository

OpenCode Swarm V2 - Verification-gated swarm with architect, review, test, and security agents

Verification-gated OpenCode swarm with architect planning, independent review, test engineering, security checks, and resumable evidence.

🔗 View Repository

OpenCode TasksMD V1 only - Background task queue for OpenCode projects, powered by TASKS.md

Background task queue that reads TASKS.md — schedules recurring work, sets per-task permissions and agents, and persists state across restarts.

🔗 View Repository  •  📦 npm

OpenCode Teamwork V2 - Antigravity-style multi-agent orchestration with verification loops

Brings the Antigravity-style Teamwork orchestration model to OpenCode V2, where a Sentinel coordinator dispatches a DAG of specialist agents and a verifier gates every task until it passes. Ships 10 agents (Sentinel, Orchestrator, Crafter, Scout, Worker, Proof Worker, Proposer, Falsifier, Synthesizer, Verifier), 7 slash commands, per-agent git worktrees, cost tracking, and a 9-step prompt-crafting wizard.

🔗 View Repository  •  📦 npm

Opencode Workspace V1 only - Multi-agent orchestration

Bundled multi-agent orchestration harness with 16 components in one install.

🔗 View Repository

opencode-background-tasks V2 - A sidebar list of the background shells and subagents still running in an OpenCode v2 session

Adds a Background list to the OpenCode v2 session sidebar with every shell and subagent sent to the background, a live timer on each, and a done or failed mark when it ends.

🔗 View Repository

opencode-workflows V2 - Claude Code-style dynamic workflows for opencode v2 — a JS script fans work out to parallel subagents

A native opencode v2 plugin that reproduces Claude Code's dynamic workflows. The model writes a small JavaScript orchestration script (agent, parallel, pipeline, phase, budget, nested workflows) that runs in a sandboxed interpreter in the background, fans work out to subagents as child sessions, and can resume from its journal. Includes a live TUI tree to watch, steer, or stop running agents. Uses your configured opencode models; no extra daemon.

🔗 View Repository  •  📦 npm

Orchestrator V2 - Cost-aware subagent routing skill plus injected orchestration instructions

Teaches the primary agent to route work to the right subagent tier: an `orchestrator` skill with routing rules, prompt templates and an escalation ladder; routing rules injected into primary sessions; subagent conduct rules injected into child sessions; and example agent-tier definitions.

🔗 View Repository

OrgX OpenCode Plugin V1 only - OrgX Work Graph peer for task dispatch, receipts, deviations, and passive reconciliation

OrgX OpenCode Plugin connects OpenCode sessions to the OrgX execution control plane. It receives OrgX task dispatches, drives the user's local OpenCode session, posts execution receipts and deviations back to OrgX, and writes compact Work Graph events for later audit-first reconciliation.

🔗 View Repository  •  📖 Docs

Pocket Universe V1 only - A subagent driven pocket universe for your primary agent

Async agents can be powerful, but orchestration is at best finicky; they fire and forget, orphan work, lose context, waste time... and tokens. This plugin extends the native opencode subagent paradigm to provide closed loop, resilient, async agents, blocking main thread execution. A "pocket universe". This ships with three tools creating a robust system for parallel subagents to communicate and coordinate work

🔗 View Repository

Sortie-dogs V2 - Agents write only the files they declared — manifest-gated orchestration loop

An opt-in OpenCode orchestration plugin. Every task starts with an explicit write manifest; edits and handoffs outside that scope are rejected. Three bounded scouts investigate read-only, a dedicated worker implements, and canonical validation plus risk-based review gate coordinator-owned completion. Dormant until you run /sortie — ordinary OpenCode sessions are untouched.

🔗 View Repository

Subagent Reporter V1 only - See exactly what your subagents are up to (in the terminal) when invoked during an `opencode run _____` session.

When opencode run <prompt> is invoked normally, the actions of the primary agent are visible in the terminal, but subagent actions are not. This makes it difficult to follow progress on the prompt, and makes it very difficult to use `opencode run` as part of an unattended process or script. This plugin pipes subagent actions/events directly to stdout, prefixes them with the subagents name, and enumerates subagents when run in parallel for easy identification. This allows the user to follow along in realtime when subagents are running, and effectively trace execution at a later time.

🔗 View Repository

Subtask2 V1 only - Orchestration system

Extend opencode /commands into a powerful orchestration system with granular flow control.

🔗 View Repository

Swarm Plugin V2 - Swarm intelligence

Swarm plugin for opencode enabling swarm-based agent coordination.

🔗 View Repository

telepathy V2 - OpenCode sessions message Claude Code, Codex, Gemini CLI and other agent sessions on the same machine

Adds telepathy_list_peers, telepathy_send_message and telepathy_read_messages as native OpenCode tools, so an OpenCode session can find and message the Claude Code, Codex, Gemini CLI, Copilot CLI, Cursor and other agent sessions running on the same machine. The plugin runs inside OpenCode and starts a turn in an idle session when a message arrives, so agents can coordinate without polling. Everything stays local (messages are JSON files in ~/.telepathy, with no network calls and no telemetry). MIT licensed.

🔗 View Repository

Workflows & Planning ₍₂₅₎

Agent Loop V2 - Goal-oriented agent loop that nudges agents toward objectives

Iteratively drives agents toward configurable objectives with turn/time budgets, marker/regex-based termination conditions, idle detection, and a /loop slash command for manual control.

🔗 View Repository

BRHP V1 only - Persistent planning state

Structured, persistent planning for OpenCode with local session state, /brhp commands, bounded planner history, and a TUI sidebar.

🔗 View Repository

ForLoop Planner V1 only - Autopilot development. Plan with ForLoop agent. Build, Test, Deploy on cloud in one go.

ForLoop offers an all-in-one platform that covers the entire development lifecycle, from initial planning and coding to automated testing and cloud deployment.

🔗 View Repository

Goal V2 - Persistent /goal mode that keeps a session working across turns

A port of OpenAI Codex's /goal for OpenCode v2. Registers a /goal command plus create_goal/get_goal/update_goal tools. The session keeps re-prompting itself until the objective is complete, with token budgets, completion audits, and a bundled goal skill.

🔗 View Repository

Godmode V2 - Discipline layer: measure → modify → verify → keep/revert

Skill plugin with 135 skills and 7 subagents that adds a measure → modify → verify → keep/revert discipline loop to AI coding agents. Works with OpenCode alongside Claude Code, Codex, Cursor, and Gemini CLI.

🔗 View Repository

GoopSpec V2 - Spec-driven development workflow

Transforms AI-assisted coding with spec-driven workflows. Features 5-phase workflow (Plan, Research, Specify, Execute, Accept), contract gates for user confirmation, 12 specialized subagents, persistent memory system, wave-based execution with atomic commits, and deviation rules for handling unexpected situations.

🔗 View Repository

Idea Inbox V2 - Capture stray thoughts into a sidebar backlog, then dispatch them to agents without derailing your session

A persistent idea backlog for the opencode v2 TUI. Mid-conversation, press leader+z (or run /idea) to park a thought in the Idea Inbox sidebar without leaving your current task — no notes app, no derailing the agent. When you have a free moment, press leader+i and pick a pending idea: a delegation mission lands in your current session and execution starts immediately, while the sidebar status flips from parked to running to done as the agent works. Agents can manage the backlog themselves through idea_add/idea_list/ idea_update/idea_start tools, and a "documented" status archives an idea after its result is written up. Storage is per-worktree SQLite, so the backlog survives restarts. Ships for opencode v2 (this entry) and as a legacy v1 line: opencode-idea-inbox@legacy-v1.

🔗 View Repository  •  📦 npm

keep-going V1 only - Sends an idle session back to work when a reviewer model judges the request unfinished

When a session goes idle, a reviewer model reads the owner's request and the last assistant message and answers CONTINUE or STOP; on CONTINUE the plugin sends a short fixed nudge such as "Keep going." as the next message. It stops for work that is done, blocked, or waiting on the owner's consent or answer, caps continuations per turn, and fails open on any error. Install with `npx github:ferdousbhai/keep-going --opencode`.

🔗 View Repository

Littlepowers V1 only - Proportional planning, recovery, and phase review gates for consequential work

A proportional planning, recovery, and engineering-discipline protocol. Scales planning depth to risk (direct, compact shape, or full brainstorm → spec → design → plan), pauses for review at each phase gate, recovers the last checkpoint after interruptions, and mirrors task checklists to OpenCode's native todo surface. Registers its skills directory through the config hook and injects read-only recovery context through the message-transform hook; the same implementation also ships for Codex, Claude Code, and Qoder.

🔗 View Repository

Micode V1 only - Brainstorm-Plan-Implement workflow

Structured workflow with session continuity, subagent orchestration, git worktree isolation, and AST-aware tools.

🔗 View Repository

Neuralyzer V1 only - Easier Ralph loop inside harness

Allow agent to wipe its own session context and re-run the first message. This allows easier and more ergonomic Ralph loop engineering.

🔗 View Repository

open-plan-annotator V2 - Annotate LLM plans like a Google Doc!

A fully local agentic coding plugin that intercepts plan mode and opens an annotation UI in your browser. Select text to strikethrough, replace, insert, or comment — then approve the plan or request changes

🔗 View Repository

OpenCode Chat Workflow Setup V2 - Configure chat workflows and reusable templates before skills run

Adds /setup-chat-workflows, /start-session, and /workflow-template commands. Workflows collect context, scope, gates, review modes, and reusable template choices before handing off to skills or actions.

🔗 View Repository

Opencode Hooks Plugin V1 only - Claude code compatible hooks

Use Claude code hooks definition and hook them to opencode hooks.

🔗 View Repository

OpenCode Plan Manager V1 only - AI-Native Implementation Planning for Modern Agentic Workflows

High-performance, minimalist plugin designed to bridge the gap between complex implementation strategies and autonomous execution. Selective context loading, zero-hallucination schemas, visible filesystem kanban.

🔗 View Repository

Opencode Roadmap V1 only - Strategic planning

Strategic roadmap planning and multi-agent coordination plugin. Provides project-wide planning capabilities.

🔗 View Repository

OpenCode Scratchpad V2 - A private per-project scratchpad you can capture and edit without leaving the TUI

One plain-Markdown scratchpad per project, keyed by absolute path. Append a line with `/scratch <text>` or open a modal editor with `/scratch`; the session sidebar shows a collapsible, read-only preview that refreshes on save. Notes live under `$OPENCODE_DATA_DIR/scratchpad` and are never written into your repo.

🔗 View Repository

OpenCode Todo Supervisor V2 - Session todo lists, plus a supervisor that keeps nudging idle sessions until the list is done

Gives every OpenCode V2 session a real todo list and then keeps the agent working on it. Ships todo_write, todo_read, todo_add and todo_remove tools (V2 ships none), stored per session with stable per-item ids, with a live progress strip in the composer. A supervisor re-prompts any session that goes idle while its list still has open items, and stays quiet while the agent is actually working, including during tool calls, subagent runs and background shells. Any markdown file can be subscribed so its GFM checklists count as todos, and the TUI panel can tick them off by hand. Nudges are delivered as synthetic session input, so the model reads them as ordinary instructions. Tested against 2.0.24.

🔗 View Repository  •  📦 npm

OpenCode Workaholic V1 only - Enforce mandatory working time and prevents premature "done" responses

- AI has an early-exit problem: it says "done" at 30 seconds while edge cases are untested, bugs remain, and docs are missing. - Workaholic enforces mandatory working time for AI until the timer hits zero. - Blocks premature "done" responses until timer expire, and require AI to call checkout() to end the task.

🔗 View Repository

opencode-bmad-workflow V1 only - BMAD workflow plugin — automates epic, feature, sprint and code review workflows using specialized AI agents

Brings the BMAD (Breakthrough Method of Agile AI-Driven Development) methodology to OpenCode. Provides specialized agents for product management, architecture, development, and QA roles. Automates epic planning, feature breakdown, sprint workflows, and code reviews through a structured multi-agent pipeline.

🔗 View Repository

OpenSpec V1 only - Add Architecture planning and specification agent for OpenSpec

An OpenCode plugin that integrates OpenSpec, providing a dedicated agent for planning and specifying software architecture.

🔗 View Repository

Plannotator V1 only - Interactive plan review UI

Plan review UI with visual annotation, private/offline sharing, and Obsidian/Bear integration.

🔗 View Repository

Ralph Wiggum unverified - Self-correcting agent loops

Iterative AI development loops with self-correcting agents based on the Ralph Wiggum technique.

🔗 View Repository

Superpowers V2 - Complete software development methodology with composable skills for coding agents

Gives your coding agent a structured methodology - brainstorm specs, write plans, run TDD, dispatch subagents, and review work. Skills trigger automatically across Claude Code, OpenCode, Codex, Gemini CLI, Cursor, and GitHub Copilot CLI.

🔗 View Repository

Todolist V2 - Restores todowrite/todoread with a sidebar todo strip

Brings back the V1 session todo tools as a server plugin. Lists are stored per session and a compact summary of open tasks is injected into the model context on every request, so the agent keeps track across long conversations and context compaction.

🔗 View Repository

Providers, Auth & Routing ₍₂₆₎

Antigravity Auth V1 only - Google Antigravity models

Use Gemini and Anthropic models for free via Google Antigravity IDE authentication.

🔗 View Repository

Antigravity Multi-Auth V1 only - Multiple Google accounts

Fork of opencode-antigravity-auth that allows using multiple Google accounts with automatic rotation when rate limited.

🔗 View Repository

Claude Code Switch (CCS) OpenCode Sync V1 only - Claude Code Switch (CCS) to OpenCode sync

An OpenCode plugin that reads your Claude Code Switch (CCS) configuration and automatically syncs the providers into your OpenCode config.

🔗 View Repository

Esuyo Custom Provider V2 - Auto-discover models from any OpenAI-compatible gateway on startup

Opencode plugin that syncs /v1/models from any OpenAI-compatible gateway into providers. Global ~/.config/opencode or project .opencode config, {env:} / {file:} secrets, include/exclude regex, and zero manual model maintenance.

🔗 View Repository

Free Models Router V2 - Free model discovery, prompt routing, and seamless quota fallback

Discovers free models from OpenRouter and models.dev, routes each prompt to the best free candidate by capability and score, and falls back to the next free model on rate limits, quota exhaustion, or 5xx errors. Ships cycle/stop/retry-last/wait fallback modes, cooldowns, configurable backoff, an optional circuit breaker, status/refresh/select tools, an RPC API, and an optional TUI plugin for toast notifications.

🔗 View Repository

Gemini Auth V2 - Google account auth

Authenticate the Opencode CLI with your Google account so you can use your existing Gemini plan.

🔗 View Repository

Kilo Gateway Auth V1 only - Kilo Gateway provider

Adds Kilo Gateway provider support to OpenCode.

🔗 View Repository

liter-llm V2 - Universal LLM API client for 143 providers

OpenCode plugin exposing liter-llm tools for chat, streaming, embeddings, and an OpenAI-compatible proxy across 143 providers.

🔗 View Repository  •  📦 npm

LM Studio Warm V1 only - Ensures your LM Studio model is loaded before every request

Guarantees the target LM Studio model is loaded and addressable before any request leaves opencode, healing cold starts, "no model loaded" errors, and mid-session idle-TTL evictions.

🔗 View Repository

Muse Auth V2 - Muse Spark billed to subscription

Use Muse Spark in opencode billed to the Muse Code monthly subscription via Meta device login instead of API billing.

🔗 View Repository

Omniroute Auth V1 only - Omniroute authentication provider

Connect and automatically fetch models from your Omniroute instance

🔗 View Repository

OpenAI Codex Auth V1 only - ChatGPT Plus/Pro OAuth

This plugin enables opencode to use OpenAI's Codex backend via ChatGPT Plus/Pro OAuth authentication.

🔗 View Repository

OpenCode Adaptive Thinking V1 only - Adaptive reasoning-effort control

OpenCode plugin that lets agents actively adjust model reasoning effort during a session, with configurable system guidance and a tool for switching between valid reasoning-effort variants.

🔗 View Repository

Opencode CMD Provider V2 - Use CommandCode plans (Go, GOAT, Pro, Max) as an OpenCode provider.

OpenCode provider plugin for the CommandCode Provider API. Authenticate via /connect browser flow or COMMANDCODE_API_KEY, auto-registers all CommandCode models ([CMD] prefix) with offline snapshot, pricing display, and per-plan deals/allowance intelligence in the sidebar and cmd_plan_summary tool.

🔗 View Repository

Opencode LiteLLM V2 - Auto-discover models from a LiteLLM proxy

Drop-in LiteLLM provider for OpenCode with zero configuration. Auto-detects a running LiteLLM proxy on common ports (4000, 8000, 8080), pulls every model from /v1/models, and registers them in OpenCode automatically — no model lists to hand-maintain. Smart name formatting, modality categorization (chat/embedding/image/audio), provider extraction, optional API-key auth, and a 5-second timeout so a slow proxy never blocks startup.

🔗 View Repository

OpenCode Model Alias V2 - Portable model aliases with variants and chains — one config entry, per-machine remapping

Define short model aliases (cheap, smart, reviewer...) in a central alias file and reference them in any agent or command model field. Each machine maps aliases to whatever provider/model it prefers, so shared agent and command configs stay portable across machines and providers. Supports model variants, alias chains (stable role names layered on machine-specific mappings), and a /alias command for managing aliases in-session.

🔗 View Repository

OpenCode Model Aliases V2 - Provider-scoped aliases for the newest eligible OpenCode models

Create stable model IDs within a provider using glob, capability, and context filters. On catalog refresh, each alias resolves to the newest eligible model already in OpenCode's catalog.

🔗 View Repository

Opencode Models Discovery V2 - Configurable model discovery and filtering without long manual config

OpenCode plugin for discovering models from OpenAI-compatible providers and API gateways, with provider and model filtering so users can avoid maintaining large configs or loading every model at once.

🔗 View Repository

OpenCode Provider Alias V1 only - Alias and curate OpenCode providers with model metadata from models.dev.

Lets users define local OpenCode provider and model aliases hydrated from existing models.dev metadata.

🔗 View Repository

OpenCode Provider Guard V2 - Restrict available providers by project directory

An OpenCode V2 plugin that lets you choose which providers are available in each project, so you can keep personal or pay-as-you-go providers out of selected projects.

🔗 View Repository

opencode-go-whoami V2 - Display local email labels for OpenCode Go/Zen accounts

Identify your OpenCode Go/Zen accounts through user-assigned email labels, a permanent sidebar panel, and /go-whoami. Runs locally and displays only masked keys and fingerprints.

🔗 View Repository

opencode-llm-proxy V2 - OpenCode-backed API gateway with tool calling

Local OpenCode plugin that exposes OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, and Gemini-compatible endpoints for any model configured in OpenCode. Supports streaming, model listing, and real tool/function calling including parallel tool calls, so tools like n8n, Open WebUI, LangChain, Continue, Zed, and coding agents can reuse the same OpenCode model setup.

🔗 View Repository  •  📦 npm

OpenHax Codex V1 only - OAuth authentication

OAuth authentication plugin for personal coding assistance with ChatGPT Plus/Pro subscriptions.

🔗 View Repository

Optimal Model Temps V1 only - Optimal sampling temperatures

Minimal plugin that nudges specific models to their preferred sampling temperature.

🔗 View Repository

Qoder Bridge for OpenCode V2 - Use Qoder AI models in OpenCode via the official Qoder Agent SDK, with tools, MCP, sessions, and quota tracking.

OpenCode plugin that integrates Qoder AI models through the official Qoder Agent SDK. Provides dynamic model discovery, streaming, tool calling, MCP integration, multimodal input, persistent sessions, live quota visibility, and usage tracking.

🔗 View Repository

Rehydra V2 - Prevent your coding agent from leaking secrets to LLM providers

Intercepts the conversation between OpenCode and the LLM, replacing secrets from .env files and detected API keys, tokens, and credentials with placeholders before they leave your machine. Secrets are transparently restored before any tool executes locally. Supports pattern-based detection for Stripe, OpenAI, AWS, GitHub, Slack, and more.

🔗 View Repository

Cost & Usage ₍₁₃₎

Cost Details V2 - Shows current and previous turn cost in the sidebar Context section

OpenCode TUI plugin that shows how much the current and previous prompt turns cost, right in the sidebar Context section next to the token count.

🔗 View Repository  •  📦 npm

DeepSeek Peak V2 - DeepSeek peak/off-peak pricing, Chinese holidays included

Shows whether DeepSeek is in its peak or off-peak pricing window. Applies the official rule including Chinese public holidays, treats a response pricing header as the source of truth and flags mismatches, adds a compact next-change time (local or UTC, with weekday) in a single non-wrapping sidebar line, works before the first request via a preview marker, and speaks English, Portuguese, Chinese, Japanese, and Korean. Supports OpenCode V2 and V1.

🔗 View Repository

OpenCode Multi-Usage V2 - One OpenCode plugin for multiple AI subscription quotas

An easily extensible OpenCode plugin for tracking AI subscription quotas. Add provider integrations with minimal effort; each is opt-in, with support for ChatGPT Codex, OpenCode Go, CommandCode and more.

🔗 View Repository

Opencode Quota V2 - Quota toasts and token tracking

Track quota and token usage across providers via automatic toasts and slash commands.

🔗 View Repository

OpenCode Session Cost V2 - Sidebar block showing true session cost including all subagents

Subagent costs hide from you by default in OpenCode's session cost display. This TUI plugin adds a sidebar block showing your true total — the active session plus every subagent (task) it spawned — and lets you expand it to see the cost of each one individually. Works for every session regardless of model or provider, no configuration required.

🔗 View Repository

OpenCode Token Tracker V1 only - Real-time token usage, cost, and latency tracking for every AI request in OpenCode

Displays a toast after each AI response with input/output tokens, cache hits, reasoning tokens, response latency, per-message cost, cumulative session cost, model identifier, and request count. Automatically aggregates tokens from subagent sessions spawned via the Task tool. Zero config — drop the file in and it works.

🔗 View Repository

Opencode Usage Bar V2 - Lightweight, configurable usage gauge with provider status alerts for the OpenCode TUI.

A minimal usage bar that shows how much of your coding-plan quota you've used and how long until it resets, in a full-width row below the session footer. Supports Claude Pro/Max and ChatGPT Plus/Pro (Codex), each toggleable per window (5h session, 7d weekly, per-model) via a simple TOML config, so you show only the bars you care about. Color-coded mini-bars escalate green/amber/red with usage, countdowns tick live, and last known values are cached so the bar appears instantly on restart. It also watches each vendor's public status page and prepends a color-coded "!" marker when there's an active incident — so you know it's them, not you. No slash commands, no sidebar, no token bookkeeping, just a lightweight always-visible gauge. Credentials are read-only from the vendors' own CLI logins or OpenCode's auth store. Installable via npm (@satas/opencode-usage-bar).

🔗 View Repository

Opencode Usage Monitor V1 only - Monitor OpenAI and Z.AI usage quotas in the TUI

OpenCode sidebar plugin that displays API usage quotas for OpenAI and Z.AI providers with per-model breakdowns, spend tracking, and configurable refresh controls.

🔗 View Repository

opencode-mystatus V1 only - Check AI subscription quotas

Check all your AI subscription quotas in one command. Supports OpenAI (Plus/Pro/Codex, etc.), Zhipu AI, Google Antigravity, and more.

🔗 View Repository

Quota Board V2 - Provider quotas, reset countdowns, and credit balances in the OpenCode V2 terminal

V2-native quota sidebar and /quota popup using existing active provider connections. Displays remaining allowances, reset countdowns, and credit balances, with collapsible providers, persistent visibility and pinning preferences, and stale-value retention after failed refreshes. Includes adapters for OpenAI, Anthropic, OpenCode Console, OpenRouter, Z.AI, Synthetic, and MiniMax.

🔗 View Repository

Sidebar Usage V2 - TUI sidebar showing per-turn and session-cumulative token usage, cache hit rate, and cost

Adds a collapsible "Usage" block to the opencode TUI sidebar showing token breakdown (input, output, reasoning, cache read, cache write), cache hit percentage, context window usage, and session cost. Updates reactively as new assistant messages arrive.

🔗 View Repository

Token Monitor V1 only - Token analysis & cost tracking with budgets, trends, and per-project analytics

Track token usage (input/output/reasoning/cache), estimate costs, view daily trends with ASCII charts, set budget alerts, and export data. Supports per-agent breakdown, agent×model cross-analysis, and project-scoped analytics.

🔗 View Repository

Tokenscope V1 only - Token analysis & cost tracking

Tokenscope, Comprehensive token usage analysis and cost tracking for opencode sessions.

🔗 View Repository

Notifications ₍₁₃₎

Opencode Caelestia Notify V1 only - Desktop notifications when opencode finishes or needs your attention

Sends a system notification when opencode finishes responding, hits an error, or waits for permission or input. Shows the session name so you know which task it is, and skips the done notification when your terminal is already focused. Built and tested on Caelestia shell, works on any Hyprland/Wayland desktop. Zero dependencies, single file.

🔗 View Repository

Opencode Host Notify Bridge V1 only - Devcontainer notifications bridged back to the host

OpenCode plugin and host helper that forward permission, question, and idle notifications from devcontainers to the host machine for desktop alerts and sound.

🔗 View Repository  •  📦 npm  •  📖 Docs

Opencode Notify V1 only - Native OS notifications

Native OS notifications for OpenCode - know when tasks complete.

🔗 View Repository

OpenCode ntfy.sh V1 only - Push notifications to keep you in the know, even when you're on the go.

An OpenCode plugin that adds push notifications through ntfy.sh.

🔗 View Repository

Opencode Status Popup V2 - Always-on-top status pill or tray icon that shows what the agent is doing (Windows)

A second screen for your agent. While a session is thinking, a small always-on-top pill types the word out letter by letter, or a tray icon draws the letter o pixel by pixel. It turns amber on provider retries, red when an execution fails, and violet when OpenCode is waiting for a permission, so you can be in another window and still notice. One PowerShell host starts on demand and is shared by every instance, and the renderer can be switched at runtime with the /popup-* commands or a tool.

🔗 View Repository  •  📦 npm

Opencode update notifier V2 - Notify about plugin updates.

Checks if your pinned plugins have newer versions available and shows a notification.

🔗 View Repository

Ring a Bell Example V1 only - Simple terminal bell plugin

A simple plugin to ring the terminal bell once a request is complete.

🔗 View Gist

Simple Notify V1 only - Native desktop notifications with near-zero dependencies

Lightweight desktop notification plugin for OpenCode. Sends native OS notifications via dbus (Linux) and osascript (macOS) when sessions complete, error, need approval, or wait for input. Shows project name, elapsed time, and message preview.

🔗 View Repository

Smart Voice Notify V1 only - Intelligent voice notifications

Smart voice notification plugin with multiple TTS engines (ElevenLabs, Edge TTS, SAPI) and intelligent reminder system.

🔗 View Repository

ul-opencode-event V1 only - Multi-channel notifications (SMTP, DingTalk, Feishu) and token usage analytics

Get notified when OpenCode sessions complete, error, or need your attention - via SMTP email, DingTalk robot, or Feishu robot. - Multi-channel: SMTP, DingTalk, Feishu, local JSONL file - Multi-level token analytics: message / session / total / project scopes - 74+ template variables with i18n support (zh/en) - Global + project config merge, automatic DB preload for crash recovery - Built-in CLI test tool

🔗 View Repository  •  📦 npm

Vibe Coding Slack Notifier V1 only - Slack DM alerts for OpenCode task completion

OpenCode-compatible Slack notifier plugin and toolkit for Codex, OpenCode, Claude Code, and Gemini workflows.

🔗 View Repository

Warcraft Notifications V1 only - Fun sound notifications

Notification plugin with Warcraft sounds for opencode completion alerts.

🔗 View Repository

WSL Notifications V1 only - Native Windows toast notifications from WSL

Enables OpenCode running inside WSL to send native Windows toast notifications for task completion, prompts, and errors using PowerShell and BurntToast.

🔗 View Repository

Sessions & History ₍₁₃₎

AutoRecord V2 - Automatic session recording

Automatically saves all OpenCode session records to Markdown files with centralized storage in ~/opencode-autorecord, plus generated HTML views (overview index + per-project pages) for browsing conversations.

🔗 View Repository

Autotitle V1 only - AI-powered automatic session naming

Two-phase session titling - instant keyword titles on user message, refined AI titles after response. Auto-selects cheapest model (flash/haiku/fast), respects custom titles, zero configuration needed.

🔗 View Repository

Handoff V1 only - Session handoff prompts

Creates focused handoff prompts for continuing work in a new session.

🔗 View Repository

Laya Drift V1 only - Semantic drift monitoring for long-running OpenCode agent sessions.

Laya Drift monitors whether an OpenCode agent is staying aligned with its original plan over long-running sessions. It uses the Laya decision model to embed session state, compares the current semantic state against a calibrated baseline with Jensen-Shannon divergence, and surfaces a live drift score in the OpenCode UI. When drift gets high, it can inject the score into the system prompt so the agent can self-correct. Sessions can also be recalibrated to establish a new baseline after an intentional change in direction.

🔗 View Repository

OpenCode Session Tracker V2 - TUI sidebar extension and Telegram integration to track active sessions/subagents and interact remotely.

A lightweight, robust TUI sidebar extension and interactive Telegram Integration for OpenCode that brings real-time session tracking, multi-process routing, and remote monitoring directly to your terminal and phone. Monitor and drive your terminal agents remotely from your mobile device, answer interactive questions, approve or deny commands, and track status streams.

🔗 View Repository

Opencode Sessions V1 only - Session management

Session management plugin for OpenCode with multi-agent collaboration support.

🔗 View Repository

OpenCode Topics V2 - Smart, zero-token session re-titling that tracks topic drift

Tracks topic drift across conversation turns and dynamically re-titles OpenCode sessions when the topic shifts, using zero-token local checks between assistant turns.

🔗 View Repository

opencode-ask V2 - Multi-turn /btw-style side chat that never adds to your session

Ask follow-up side questions about the current session in a full-screen view. Switch to the main conversation and back with shift+arrows, and close the side chat to forget it. Built for OpenCode 2.

🔗 View Repository

opencode-lens V1 only - TUI bridge and MCP toolkit for inspecting and controlling live opencode sessions

opencode-lens is an opencode TUI plugin plus MCP server that lets external agents inspect running opencode instances, send prompts, read messages/diffs/todos, handle permission and question prompts, and monitor session completion through a local Unix socket bridge.

🔗 View Repository  •  📦 npm  •  📖 Docs

opencode-session-tools V2 - OpenCode v2 session tools – session metadata, history recall, and Claude Code-style messaging between sessions

Two plugins for OpenCode v2, built on the public v2 plugin API. opencode-meta adds opencode_meta (bounded session metadata, relations, statistics and tool usage) and send_message, which lets one session send plain text to another by ID or title; the target receives it as a queued message marked as coming from a peer, never as user approval. opencode-recall-lite adds recall and recall_get to search and read back earlier sessions across projects, with a short reminder to check history before answering questions about past work.

🔗 View Repository

Session Time V2 - Session start / last answer timestamps and a clock

Adds a wall clock to the prompt footer, a "Session" block to the sidebar with the session start time and the time of the last model answer, and a /time timeline panel with every turn's timing.

🔗 View Repository

Smart Title V1 only - Auto-generate session titles

Auto-generates meaningful session titles using AI.

🔗 View Repository

Zellij Namer V1 only - Auto-rename Zellij sessions

Keeps your Zellij session name in sync with your work.

🔗 View Repository

Code Review & Quality ₍₉₎

Boffin V2 - Routed architectural guardrails for AI coding agents in OpenCode

Staff-engineer layer for AI coding agents: routes per-edit architectural constraints and requires verification. Not another AGENTS.md.

🔗 View Repository  •  📖 Docs

dejavu V2 - Cross-session error gates for recurring tool-call failures

Mechanically detects recurring tool-call failures (3 failures across 2 distinct sessions promote a gate) and enforces them: reminder on the next attempt, hard block on same-session repeat offense. Diagnostics never block, they annotate. Two-scope store with secret scrubbing and control-char stripping before persistence, 60-day TTL, self-healing reconcile, doctor report. One harness-neutral store shared with Claude Code, Codex CLI, Gemini CLI, Cursor, Copilot CLI, Crush, Devin CLI, Kiro and Cline. TypeScript + Bun, ships as source, no build step.

🔗 View Repository  •  📦 npm

kibi-opencode V1 only - OpenCode guidance for keeping requirements linked to code and tests

Kibi connects project requirements to code and tests. This optional OpenCode plugin gives agents relevant guidance and runs background sync and checks.

🔗 View Repository  •  📦 npm

lazy-clean V1 only - Minimal-code guidance, review commands, and a dependency-free TS/JS slop checker

Injects configurable minimal-code guidance into OpenCode each turn and adds slash commands for code review, audits, and test review. Includes a local, dependency-free TypeScript/JavaScript heuristic checker, run manually in OpenCode. Install from the full checkout to keep the shared hooks and skills directories alongside the plugin.

🔗 View Repository

opencode-review V1 only - Automatic structured code review with configurable dimensions and auto-fix

An automatic code review plugin for OpenCode CLI. Reviews staged changes when session goes idle, with configurable cooldown, multi-dimension analysis (code quality, security, performance, testing, documentation), severity levels, and auto-fix chain support.

🔗 View Repository

Semantic Anchors V1 only - Runtime contract enforcement for safe, traceable AI agents

Runtime steering plugin that enforces behavioral contracts via plugin hooks. Monitors chat messages and tool calls against configurable rules, supports BLOCK/WARN modes, built-in profiles (socratic, architecture), step confirmation intervals, override mechanism, dynamic role switching, and Zod-validated YAML configuration.

🔗 View Repository

Veris for Daytona V1 only - OpenCode sessions run in a Daytona sandbox whose vendor API calls are answered by Veris twins

Every OpenCode session runs in a Daytona sandbox whose calls to Stripe and the rest of your vendor stack are answered by stateful Veris twins, and the agent gets a receipt of what the vendor actually received. Unmodified code, real hostnames, no image to build.

🔗 View Repository  •  📦 npm  •  📖 Docs

Veris for E2B V1 only - OpenCode sessions run in an E2B sandbox whose vendor API calls are answered by Veris twins

Every OpenCode session runs in an E2B sandbox whose outbound vendor API calls are answered by stateful Veris twins, with a receipt of what the vendor actually received. One line in opencode.json - no template to register, no server to start.

🔗 View Repository  •  📦 npm  •  📖 Docs

Veris Sim V2 - Three engineer-invoked commands that test an application against Veris twins of its services

Three engineer-invoked commands that test an application against Veris twins of the services it calls. setup wires a repository once. build measures every vendor claim the task rests on before designing, then proves the change through veris-proxy. fix reproduces the failure before touching source and proves it closed. Ships the veris MCP server.

🔗 View Repository  •  📦 npm  •  📖 Docs

Security & Sandboxing ₍₁₃₎

Aegis-DevOps V1 only - Blocks kubectl delete, terraform destroy or DROP TABLE unless your signed policy allows it

A tool.execute.before plugin that checks every bash command OpenCode wants to run (kubectl, terraform/tofu, aws/az/gcloud, helm, argocd, git, SQL) against a signed, authority-checked policy before it runs. A rule only counts if it hasn't been edited since it was signed and its author was allowed to write that kind of rule, so a line planted in a ticket can't become policy. Only active in projects with a .aegis/ policy. Install with `pip install aegis-devops && aegis install opencode`.

🔗 View Repository

Aftermath V1 only - Post-failure recovery for agents — stop, revert cheap damage, return to last-known-good

Aftermath is a post-failure recovery skill for OpenCode. When something breaks — a test suite goes green to red, the same error recurs, a destructive command looms, or the user says "you broke it" — it stops forward motion, reverts the actual bad hunk instead of patching around it, and returns the session to last-known-good. Includes four intensity levels (lite/full/ultra/off), four slash commands (status, revert, audit, log), and lifecycle tripwires for test failures, repeat errors, and destructive commands.

🔗 View Repository

CC Safety Net V2 - Safety net catching destructive commands

A Claude Code plugin that acts as a safety net, catching destructive git and filesystem commands before they execute.

🔗 View Repository

DataCloak V1 only - Swap secrets and PII for realistic fakes before AI sees them; restore originals on return.

Local-first sensitive-data protection for OpenCode. Cloaks API keys, JWTs, PEMs, env vars, DSNs, emails, phones and IPs with Faker-generated synthetics pre-dispatch, blocks env-dump tool calls, redacts tool output, and auto-restores originals into file writes. Zero-config and fully automatic.

🔗 View Repository

Envsitter Guard V1 only - Prevent .env leaks

OpenCode plugin that prevents agents/tools from reading or editing sensitive .env* files, while still allowing safe inspection via EnvSitter (keys + deterministic fingerprints; never values).

🔗 View Repository

jes V1 only - Prompt-injection and jailbreak guardrails for AI agents, powered by decision models like Jev.

OpenCode plugin from the open-source jes library. It checks the user prompt, each tool call before it runs and each tool result, sending every check to a decision model such as Jev, which outputs a classification, not text. Secrets and PII are redacted locally before any model sees them. The Jev checks call a hosted API and need a TYPESAFE_API_KEY.

🔗 View Repository

Opencode Log Sanitizer V1 only - Sanitizes pasted logs by redacting long strings, JWTs, bcrypt hashes, and base64 blobs

Sanitizes pasted logs before sending them to AI by redacting long quoted strings, JWT tokens, bcrypt hashes, and base64 blobs to reduce token usage and remove irrelevant noise.

🔗 View Repository

OpenCode Seatbelt V2 - macOS kernel sandbox that blocks agents from reading secrets via .ocignore

Applies a macOS Seatbelt profile to the OpenCode process the moment the plugin loads, so every tool and child process (read, glob, grep, the shell, MCP servers, custom tools) is confined by the kernel. Patterns from .ocignore plus a built-in secret baseline are compiled to syscall-level rules instead of being matched against command text, so "tail" instead of "cat", symlinks, hardlinks, base64 and interpreters are all covered, and rename/unlink of protected files is blocked too. Requires macOS and OpenCode v2.

🔗 View Repository  •  📦 npm

opencode-pre-tool-review V2 - Model-assisted pre-execution review for OpenCode V2 tool permissions

Experimental OpenCode V2 plugin that reviews tool permission checks using a configurable reviewer model before execution. Integrates with OpenCode's native permission system, supports conservative read-only whitelisting, and offers user-confirmed one-time releases for blocked actions.

🔗 View Repository

opencode-safety-gates V2 - Run Claude Code bash-gate scripts inside OpenCode, deny self-revoking commands, catch fabricated CLI syntax.

Safety gates for OpenCode built on the V2 plugin API, with zero runtime dependencies. Wraps the built-in bash tool executor and runs your Claude Code PreToolUse/Bash gate scripts unchanged, from the same settings.json wiring — one script enforces in both tools. Ships a bundled example gate (options.exampleGates) that denies self-revoking commands — service stops that destroy the session's own access to the host (sssd, sshd, NetworkManager, firewalld, iptables flush, SSSD cache wipe). A config-driven verify-cli-syntax guard sends one corrective prompt per session when a reply recommends CLI syntax the session shows no verification for (--help, man page, or upstream source). Setup logs a health line with the active gate count, so a dead gate is never silent.

🔗 View Repository

opencode-shield-bash V2 - Gate bash calls through an LLM judge subagent session to allow or deny their execution

Gate every bash tool call through a second judge session. The judge applies a fixed policy covering destructive file operations, disk/device writes, piping remote output to interpreters, privilege escalation, reverse shells, secret exfiltration, resource bombs, and system-wide installs, and returns an allow/deny verdict before the command runs. Denials surface the reason to the calling agent; verdicts are cached per root session.

🔗 View Repository  •  📦 npm

Secret Scrubber V2 - Scrub secrets from the outbound LLM prompt before it reaches the provider

Redacts secrets from the whole outbound prompt (system, messages, tool calls, and tool results) at the V2 session hooks, so the model and provider never see them. Two detection layers - known-value matching from environment variables and an optional secrets file, plus a vendored high-signal Gitleaks pattern set. Includes local logging and sub-millisecond overhead per call.

🔗 View Repository

Secrets Env V2 - Inject secrets from a dotenv file into process.env, hot-reload, reconnect MCP servers

Loads secrets from ~/.config/opencode/secrets.env (dotenv format) into process.env at startup, so MCP servers ({env:NAME} references), other plugins, and OpenCode's own provider-key discovery can use them — keeping keys out of opencode.json(c) and your config safe to commit. Watches the file and hot-reloads changes in about a second, then reconnects only the MCP servers that reference a changed variable. Works with both OpenCode V1 (opencode-ai >= 1.14.34) and V2 (@opencode/cli).

🔗 View Repository  •  📦 npm

Git & GitHub ₍₅₎

GitHub Release V2 - Automated GitHub releases

Create and publish GitHub releases with semantic versioning, tag management, and auto-generated release notes.

🔗 View Repository

Opencode Worktree V1 only - Zero-friction git worktrees

Zero-friction git worktrees for OpenCode. Auto-spawns terminals, syncs files, cleans up on exit.

🔗 View Repository

PR Auto-Signature V2 - Automatically adds AI model signature to PRs, Issues, and Commits

Plugin that automatically appends AI model signature to Pull Requests, Issues, and git commit messages. Supports GitHub MCP tools, git CLI, and gh CLI. Recognizes 90+ AI models including Claude, GPT, Gemini, DeepSeek, Llama, Mistral, and more.

🔗 View Repository

PRs V2 - Lists GitHub pull requests referenced by the current session

OpenCode TUI plugin that lists GitHub pull requests opened by the current session in the sidebar, with titles scrolling on hover. Requires an installed and authenticated GitHub CLI.

🔗 View Repository  •  📦 npm

Worktrees V2 - Run subagents in isolated git worktrees so parallel editors never collide

Selected agents (default `worker`) each get their own git worktree branched from the default branch, with file copies via .worktreeinclude, automatic cleanup of unchanged worktrees, and enforcement against edits to the main checkout. Includes a bundled worktree-isolation skill.

🔗 View Repository

Remote & Mobile ₍₂₎

opencode-tg V2 - Control OpenCode from Telegram — every session gets its own thread

Runs inside the OpenCode server as a plugin (zero runtime dependencies) and mirrors every session into a Telegram chat: each session and subagent gets its own thread, and writing in the thread is prompting the agent. Answer the agent's questions and approve permissions with a tap, run shell commands in the session (a three-tier guard passes read-only commands, asks for confirmation on service-breaking ones, and blocks the irreversible), browse and attach project files, schedule tasks, switch models and agents, watch CI runs, and steer the message queue mid-turn. Optional extras: multi-bot topologies (one bot per project or per session), local whisper.cpp voice transcription, and a dead-prompt watchdog that reports the real cause when a turn never starts.

🔗 View Repository  •  📖 Docs

OpenCodeMirror V1 only - Bridge OpenCode sessions to chat platforms from your phone

Native OpenCode plugin that mirrors session output to Telegram and lets you approve permissions, answer questions, and send prompts from chat. Runs inside the OpenCode process with no separate server.

🔗 View Repository

Web & Browser ₍₇₎

Crawlberg V2 - Web crawling and scraping tools for opencode

OpenCode plugin exposing crawlberg CLI tools for scraping pages, crawling sites, and mapping URLs to JSON or Markdown.

🔗 View Repository  •  📦 npm

Google AI Search V1 only - Query Google AI Mode (SGE)

An opencode plugin that exposes a native tool for querying Google AI Mode (SGE).

🔗 View Repository

HTML to Markdown V2 - Fast, lossless HTML to Markdown conversion

OpenCode plugin exposing html-to-markdown CLI tools for converting HTML to clean Markdown with metadata, tables, and document structure.

🔗 View Repository  •  📦 npm

MrScraper V2 MCP - Web pages as HTML, Markdown, or screenshots, plus Google search, through the hosted MrScraper MCP server

OpenCode plugin that registers the hosted MrScraper MCP server (https://mcp.mrscraper.com/mcp, OAuth 2.1) and three on-demand skills. Its tools retrieve known pages as HTML, Markdown, or screenshots (scraperAPI), search Google (webSearch), and report account usage (status). The skills follow a retrieve-first workflow that preserves raw page content and derives structured output locally. Works in OpenCode 2 and OpenCode 1. A MrScraper account is required.

🔗 View Repository  •  📦 npm  •  📖 Docs

OpenCode Chromium Browser Plugin unverified - Browser automation for Chromium browsers with a readable extension and native host

OpenCode browser automation for Chromium-based browsers using a readable Manifest V3 extension, Node.js native messaging host, and OpenCode-native tools. Supports Chrome, Edge, Brave, Chromium, screenshots, CDP commands, DOM actions, downloads, console/network inspection, and controlled tab sessions.

🔗 View Repository  •  📖 Docs

opencode-serpapi V2 - Real-time web search for OpenCode via the SerpApi API

Adds a native serpapi_search tool to OpenCode, giving it real-time access to Google, Amazon, YouTube, and 100+ other search engines through SerpApi, without manual MCP configuration.

🔗 View Repository

Xquik unverified MCP - X/Twitter data skill & MCP server

X/Twitter data skill — MCP server, REST API, 20 extraction tools. Works with Claude Code, Cursor, Codex, and 40+ agents.

🔗 View Repository

UI & TUI ₍₁₃₎

aerovato/opencode-quotes-plugin V1 only - Display inspirational quotes instead of tips

Replaces the default home-page tips with inspirational quotes for a more motivating OpenCode experience.

🔗 View Repository

Cut V2 - Clipboard-aware Ctrl+X cut for prompt and dialog inputs

Makes Ctrl+X cut selected text in prompt and dialog inputs. Copies to the system clipboard first and deletes only after the write succeeds; leaves the input untouched if the selection or focused editor changes, and ignores repeated cut requests while a copy is pending.

🔗 View Repository

Lens V2 - Plugin injection framework for OpenCode Web

A local reverse proxy that loads web plugins and injects them into the OpenCode Web HTML response. Adds script injection, an integrated settings UI for toggling plugins, and a simple manifest format for building browser-side OpenCode Web plugins. Enables extending the OpenCode Web UI with custom JavaScript modules, HTML injection, and interactive controls.

🔗 View Repository

Opencode Agents Sidebar V1 only - Browse configured OhMyOpenAgent agents in the TUI

OpenCode sidebar plugin that displays configured OhMyOpenAgent agents with lifecycle-based categories, collapsible sections, descriptions, and model information.

🔗 View Repository

Opencode Canvas V1 only - Interactive terminal canvases in tmux splits

Interactive terminal canvases (calendars, documents, flight booking) in tmux splits. Port of claude-canvas for OpenCode.

🔗 View Repository

Opencode Visualizer V1 only - 2D pixel-art office for AI agents

Turning raw OpenCode terminal logs into cozy 2D pixel office chaos. Watch your agents work, idle, and celebrate success in a bustling virtual office.

🔗 View Repository

opencode-ascii V1 only - Strip unicode characters from output and replace them by their ASCII equivalents

Substitute unicode characters for ASCII. Em-dash (—) and en-dash (–) for hyphens (-); right arrow (→) for (->) and so on.

🔗 View Repository

opencode-dir-tree-tui V2 - File tree sidebar with Git status and quick file actions

Browse project files in the OpenCode V2 sidebar with Git status indicators, custom ignore rules, path copying, and quick file and folder opening.

🔗 View Repository

opencode-fold-tools V2 - Expand tool outputs on click, with one row per call and native file diffs.

Keeps each OpenCode TUI tool call on its own line in conversation order. Click to reveal output underneath while preserving the input summary, or use Ctrl+O to toggle all matched rows. Retains native colored file diffs and provides a full tool-record viewer. Includes npm installation and a GIF demo.

🔗 View Repository

opencode-shortcuts V2 - A keyboard shortcut cheatsheet and rebinder for the OpenCode v1 and v2 TUI

Press ctrl+/ to see every key you can use in the screen you're on, including your overrides. Rebind any built-in command in place, with clash handling and comment-preserving config edits. OpenCode v1 and v2.

🔗 View Repository

opencode-skills-tui V2 - Browse skills, track loaded state, and preview SKILL.md

OpenCode V2 skills sidebar with loaded-state tracking, SKILL.md previews, token estimates, and skill mentions in prompts.

🔗 View Repository

Shell Magazine V2 - Real-time shell command monitor in the TUI sidebar

Sidebar panel that tracks every shell command in real time. Reads the host shell registry and session.shell.* events for true status (running/exited/timeout/killed, exit code, PID, elapsed time), tails background output, backfills history from session messages, and offers kill, copy, open-output, and finish notifications. Multi-TUI safe through the host's atomic storage.

🔗 View Repository

UNMOJI unverified - Strip emojis from output

A simple plugin that strips ALL emojis from agent outputs in Opencode.

🔗 View Repository

Voice ₍₄₎

Listen STT V2 - Offline dictation, 99 languages

Voice dictation for OpenCode V2 on Windows. Runs whisper.cpp locally across 99 languages, pins the spoken language so it is not mis-detected, and uses CUDA or Vulkan automatically. Nothing leaves the machine.

🔗 View Repository

Opencode TTS V1 only - Voice summaries for idle responses

Speaks assistant responses aloud when a session goes idle, with summary or full-text modes and local TTS backends.

🔗 View Repository

OpenCode Web Voice Plugin V2 - Voice input for OpenCode Web through Lens

Adds a microphone button to the OpenCode Web prompt composer using the Web Speech API. Record speech, insert transcripts directly into the prompt editor, and send naturally. Requires Lens to load.

🔗 View Repository

Speak TTS V2 - Free neural voices, no API key

Reads OpenCode answers aloud on Windows using Microsoft's free Edge neural voices. Condenses long answers through the model you already have connected, and falls back to the built-in Windows voice when Edge is unreachable.

🔗 View Repository

Media & Documents ₍₉₎

GPT Imagegen V1 only - gpt-image-2 in OpenCode — no API cost when using your ChatGPT subscription

Brings gpt-image-2 (ChatGPT Images 2) image generation to OpenCode. When you sign into OpenCode with your ChatGPT account, generations are billed against your existing Plus / Pro / Business plan — no per-image API cost. An OpenAI API key path is also planned. Supports reference images for style guidance and edits.

🔗 View Repository

Image Preview V2 - Inline image previews from URLs in OpenCode Web

A Lens plugin that renders images from URLs inline in OpenCode Web chat. Supports markdown image syntax (\) and raw image URLs. Features click-to-expand, lazy loading, code block protection, and error fallbacks.

🔗 View Repository

Image Reader Relay V1 only - Let text-only models read pasted images via a vision model (OpenCode plugin)

OpenCode plugin that lets text-only models read images pasted into a session. Stashes pasted images in memory, injects a note into the model's context only, and registers a `read_image` tool the main model calls with its own question — relaying the image to a configurable vision model in a throwaway session, then cleaning up. Models that support images natively are left untouched.

🔗 View Repository

Modellix V2 - Unified API for AI image and video generation

OpenCode-compatible plugin/skill for async image and video generation via Modellix CLI (model run --wait, task download) with default T2I/T2V/I2I/I2V/V2V/TTS/STT/STS models.

🔗 View Repository

Oh My Cassette V2 - Chat your raw clips into a finished cut

Adds a local stdio MCP server that turns a chat message into an edited video: beat-synced cuts, auto-matched background music, subtitles, transitions, and picture-in-picture. Ingest local clips, then drive a persistent multi-turn editing thread — every turn returns a timeline delta and a contact-sheet preview, so an edit can be reviewed before anything renders. Editing and rendering run on the hosted Cassette service, which requires an account; the plugin, skills, and MCP server are MIT licensed.

🔗 View Repository  •  📖 Docs

OpenCode Vision V2 - Give text-only OpenCode models the ability to see images

Intercepts images pasted into chat, routes them through a configurable vision backend (an MCP server or a local CLI like MiniMax mmx), and injects the resulting text description back into the conversation so non-vision models can reason about images. Ships no vision of its own — you provide the backend, it provides the wiring.

🔗 View Repository

SpriteCook MCP V2 MCP - Generate 2D game assets from OpenCode with SpriteCook

Connect OpenCode to SpriteCook to generate game assets with AI, including sprites, pixel art, icons, tilesets, characters, and animations.

🔗 View Repository  •  📖 Docs

Vision V2 - Screen recording and video analysis tools (record_screen, watch_video, video_frames)

Gives OpenCode agents eyes: record the full screen or a single window with ffmpeg, send clips to a video-capable model for timestamped Q&A, or turn videos into timestamped contact sheets for image-only vision models. Windows-focused, with a bundled video-analysis skill.

🔗 View Repository

Xberg V2 - Local document extraction for 97+ formats

OpenCode plugin exposing xberg CLI tools for document extraction, MIME detection, and supported-format discovery.

🔗 View Repository  •  📦 npm

Docs & Research ₍₃₎

opencode-craft V1 only - Senior pair programmer, craftsmanship coach, and knowledge ledger for OpenCode

Transforms OpenCode into a seasoned Principal Engineer / Technical Craft Lead who builds software at full velocity while coaching system architecture, trade-offs, and interview readiness. Includes auto-synced LEARNING_PATH.md and @expand / @interview-angle skills.

🔗 View Repository  •  📦 npm

Plugin Template V1 only - CICD setup for plugins

Focuses on providing the CICD setup with generator script, release please, bun publish, npm trusted publishing, and mise tasks.

🔗 View Repository

Research Papers V1 only - Search arXiv and OpenAlex for research papers with recency and citation filtering

OpenCode plugin that adds a research_papers tool for searching scholarly literature. Uses arXiv for fresh preprints and OpenAlex for broader metadata, citation counts, and open-access links. Supports filtering by latest, trending, or top-cited papers.

🔗 View Repository

Environment & Setup ₍₁₁₎

Awake V2 - Keeps your computer awake while a session works, even with the lid closed

OpenCode plugin that keeps the computer from sleeping while a session is working, including lid-close sleep on MacBooks. Other sleep plugins only block idle sleep with caffeinate; this one also blocks lid sleep via pmset on macOS and systemd-inhibit on Linux, and releases it as soon as every session is idle.

🔗 View Repository  •  📦 npm

Background V1 only - Background process management

Background process management plugin for opencode.

🔗 View Repository

Devcontainers V1 only - Multi-branch devcontainers

Plugin for running multiple devcontainer instances with auto-assigned ports and branch-based isolation.

🔗 View Repository

Direnv V1 only - Load direnv variables

Automatically loads direnv environment variables at session start. Perfect for Nix flakes.

🔗 View Repository

OpenCode 2 Shell Tasks V2 - Claude Code-style background shell jobs for OpenCode 2 (V2 plugin API)

Run long commands without blocking the current turn: detached processes that return a task ID immediately, combined stdout/stderr logs with bounded tails, exit-code reconciliation after a restart, process-group kill, and a native /tasks TUI panel. Built on the OpenCode 2 V2 plugin API.

🔗 View Repository

OpenCode Agent Tmux unverified - Real-time tmux panes for OpenCode agents with auto-launch, streaming, and cleanup.

Smart tmux integration for OpenCode that auto-spawns panes to stream agent output, supports flexible layouts and multi-port setups, and cleans up when sessions finish.

🔗 View Repository

OpenCode Docker Panel V2 - Docker containers in the TUI sidebar, with control and logs

Shows the running containers in the session sidebar with their state, published ports and compose project, and a pin keeps a container visible whatever its state. Clicking a row opens the actions docker accepts for that container's current state - start, stop, restart, tearing down its compose project behind a confirmation, or starting the stack declared by the compose file in the agent directory - and the same window doubles as a scrollable log view.

🔗 View Repository

Opencode Synced V2 - Sync configs across machines

Enables syncing global opencode configurations across machines with public/private visibility options.

🔗 View Repository

OpenCode Zellij Indicator V2 - Show each OpenCode session's status and title on its Zellij tab

Puts a status icon (working, needs-you, done-unseen, done-seen) and the session title on each Zellij tab, so you can see at a glance which of your parallel OpenCode agents is busy, blocked on a prompt, or finished — without switching tabs. Includes an optional stopwatch and sound notifications.

🔗 View Repository

Shell Strategy unverified - Avoid interactive shell hangs

Instructions file that teaches LLMs how to avoid interactive shell commands that hang in non-TTY environments.

🔗 View Repository

Stay Awake V2 - Keep the computer awake only while a session is generating

Long agent runs (model streaming, tool execution, compaction, rate-limit retries) routinely take minutes with no user input, and default OS power settings treat that as idle and suspend mid-prompt. This plugin holds a system sleep inhibitor for exactly as long as any session is generating, then releases it — caffeinate -dims -w <server pid> on macOS, systemd-inhibit on Linux, no-op elsewhere. Busy detection is event-driven rather than polling, so a tool call that runs for minutes keeps its work item open instead of being released mid-run. Every inhibitor is bound to the server pid and a stale cap bounds any lost end event, so the machine can never be wedged awake. Zero dependencies, MIT.

🔗 View Repository

Observability ₍₁₃₎

@telemetry-dev/opencode V1 only - OpenTelemetry traces for OpenCode agent turns, model calls, and tools

Records OpenCode agent turns, model calls, tool executions, and session events with OpenTelemetry for telemetry.dev. Available as the @telemetry-dev/opencode npm package; source and setup instructions are in packages/opencode of the linked SDK repository.

🔗 View Repository

Godmode V1 only - A local, tamper-evident record of what a coding agent did, what it claimed, and what was verified.

Godmode keeps a hash-chained local record of what a coding agent did, what it claimed, and what was verified, and refuses "done" without a citation. On OpenCode it runs as an AGENTS.md instruction adapter plus an optional Bun plugin shim (adapters/opencode/godmode.opencode.js) that routes every bash, write, edit and patch call through the same pre-tool gate the other hosts use and throws before a refused tool runs. Zero runtime dependencies, zero network use, nothing leaves the machine.

🔗 View Repository

MARGINAL V1 only - Runtime governance for OpenCode that detects no-progress repetition and records local evidence.

MARGINAL is a local-first runtime governor for AI coding agents with an OpenCode observe-only integration. It observes tool activity, detects proven no-progress repetition, and records decision evidence without blocking OpenCode actions.

🔗 View Repository

OpenCode Obvious Grid V1 only - The obvious grid — know what every agent is doing from across the room

Ever squinted at five terminals trying to figure out who's actually working? This plugin puts every opencode instance on one big screen as a giant color-coded card — orange = working, blue = waiting on you, red = errored, green = idle. Readable from across the room, so "obvious" is not just a vibe, it's a font size. Stick it on the wall display and you'll never walk over to squint again. It's a grid, but make it fashion: each card carries live tokens, cost, cache rates, speed graphs, git branch, and deep links into web sessions — plus built-in ntfy push and alarm-sound notifications, because even an obvious grid needs to yell at you sometimes. Zero infrastructure: the plugin serves its own page. One install. Infinite peace of mind. Your agents will never know they're being watched.

🔗 View Repository  •  📦 npm

Opencode Telemetry V1 only - Passive cross-session telemetry to local SQLite; cost rollups across agent orchestration chains

Logs every opencode session to a local SQLite database automatically — tokens, tool calls, skills, per-turn cost. Aggregates costs across orchestration chains (conductor + child sessions), offers turn-by-turn forensic inspection, and ships an `octm` CLI for reports without spending model tokens. Zero cloud, zero network, no message bodies stored.

🔗 View Repository

OpenCode Throughput V1 only - Real-time LLM performance monitoring

Real-time LLM performance monitoring plugin for OpenCode. Tracks TTFT, TPS, latency, token usage, and cost per model with toast notifications and a TUI sidebar display.

🔗 View Repository

OpenCode Vitals V2 - Turns, steps and streaming tok/s in a small always-on-top bar

A desktop bar for the session you are working in - turns, steps, and average streaming tokens per second, measured as generated tokens over active stream time so tool waiting stays out of the denominator. One-command install with no configuration file edit and no network calls; standard-library Python only, built for OpenCode V2.

🔗 View Repository

opencode-mlflow-plugin V1 only - Track OpenCode session metrics and tool execution timings in MLflow.

Records OpenCode sessions as MLflow runs with message and tool counts, tool execution timings, and agent, model, and provider tags. Connects to a local MLflow tracking server for comparing session metrics. Available on npm as opencode-mlflow-plugin.

🔗 View Repository

opencode-plugin-otel V2 - OpenTelemetry telemetry exporter for opencode sessions, mirroring Claude Code monitoring signals

Exports metrics, logs, and traces from opencode sessions via OTLP/gRPC to any OpenTelemetry-compatible backend (Datadog, Honeycomb, Grafana Cloud, etc.). Instruments session lifecycle, token usage, cost, tool durations, and git commits — mirroring the same signals as Claude Code's monitoring.

🔗 View Repository

opencode-trace V2 - Capture raw json LLM requests+responses as jsonl traces, with interactive HTML viewer

OpenCode plugin that logs raw provider request and response payloads to ~/opencode-trace as self-contained HTML files with an interactive viewer for debugging and inspection.

🔗 View Repository

opencode-trace V1 only - Automatic tracing, visualization, and replay for OpenCode AI conversations with token and latency metrics

A comprehensive tool suite that automatically records every interaction between OpenCode and AI Providers (OpenAI, Anthropic), including complete request/response payloads, token usage, and latency metrics. Features a real-time Web Viewer for conversation history and statistics, per-session tracing control via slash commands and agent tools, CLI for export/query, and support for both global and project-level storage.

🔗 View Repository

System Prompt Logger V1 only - System prompt logger

OpenCode plugin that intercepts the system prompt before it's sent to the LLM and logs it.

🔗 View Repository

WakaTime V2 - WakaTime integration

WakaTime integration plugin for tracking coding activity in opencode sessions.

🔗 View Repository

Integrations ₍₄₎

Beads Plugin V1 only - Beads issue tracker integration

Integration for Steve Yegge's beads issue tracker with /bd-* commands.

🔗 View Repository

Dodo Payments V1 only - Payments, subscriptions, webhooks, and billing for OpenCode agents

Official Dodo Payments plugin for OpenCode (also Claude Code, Codex, Cursor). Bundles eight integration skills (checkout, subscriptions, webhooks, usage-based billing, credits, license keys, BillingSDK, best practices) plus two MCP servers — a live API server with browser OAuth and a documentation search server. Distributed via npm as @dodopayments/opencode-plugin.

🔗 View Repository

Internet Court V2 - Trust layer for agent-to-agent commerce: mandates, payments, escrow, dispute resolution.

The trust layer for agent-to-agent commerce — natural-language mandates, delegated permissions, x402 payments, escrow, and dispute resolution as one open, catch-all skill. By the Internet Court Consortium (internetcourt.org). MIT licensed.

🔗 View Repository  •  📖 Docs

Pilot unverified - Automation daemon

Automation daemon that polls for work from GitHub issues and Linear tickets.

🔗 View Repository

Code Intelligence ₍₄₎

Morph Fast Apply V2 - 10,500+ tokens/sec code editing

Integrates Morph's Fast Apply API for faster code editing with lazy edit markers and unified diff output.

🔗 View Repository

OpenCode Indexing V2 - Semantic code search for OpenCode that reuses Kilo Code's index and keeps its own incremental one.

Adds semantic (vector) code search to OpenCode. It reads Kilo Code's existing index (Qdrant ws-* collections and Kilo's local LanceDB) without ever modifying it, and can import those vectors and chunks at zero embedding cost, then keep its own incremental index (Qdrant oc-* collections or embedded LanceDB) keyed on a per-file SHA-256. Search merges results from both indexes, so semantic search keeps working even when Kilo is not installed, is disabled, or its index is stale.

• Reuse Kilo Code's index read-only, importing vectors without re-embedding
• Own incremental index with manifest-based freshness (new/changed/deleted only)
• Vector-store agnostic: Qdrant or embedded LanceDB
• Settings and index actions from the TUI, Desktop and web clients
• Supported embedding providers: Mistral, OpenAI, Ollama, Gemini, Voyage, OpenRouter and any OpenAI-compatible endpoint
  
  🔗 View Repository  •  📖 Docs

OpenCodeRAG V2 - Local-first RAG plugin for semantic code search with tree-sitter chunking and LanceDB

A local-first Retrieval-Augmented Generation plugin for OpenCode that enables semantic code search across your workspace. Features AST-aware chunking via tree-sitter (19 languages), incremental indexing with file-change detection, watch mode for live re-indexing, and configurable embeddings via Ollama (default) or OpenAI. Includes a CLI (index, query, status, clear) and an OpenCode plugin with a chunk retrieval tool + automatic chat.message file suggestions. All processing stays local by default — no data leaves the machine.

🔗 View Repository

Tree-sitter Language Pack V2 - Code intelligence for 300+ languages

OpenCode plugin exposing tree-sitter-language-pack tools for parsing and extracting structure, imports, and symbols from 300+ languages.

🔗 View Repository  •  📦 npm

Prompting & Skills ₍₁₂₎

Agent Skills (JDT) V1 only - Dynamic skills loader

Dynamic skills loader that discovers skills from project, user, and plugin directories.

🔗 View Repository

Command Inject V1 only - Auto-inject project commands into OpenCode

Automatically discovers and injects Makefile targets, npm/pnpm/yarn/bun scripts, and local skills at startup. Type / to see and run all project commands.

🔗 View Repository

homie V2 - Four-level chat-voice personality layer — same code, different voice (off / yo / dawg / mafa)

A prompt-injection personality layer that changes how the agent talks in prose while leaving technical answers, code, and commands identical. Ships yo, dawg, and mafa registers plus an off switch, set with /homie. Same package also runs on Claude Code, Codex, and Pi.

🔗 View Repository

Manage Skills V1 only - Wizard-driven skills management for OpenCode.

Modal-based skill install, remove, list, and update workflow that avoids ANSI prompts while wrapping the skills CLI inside OpenCode.

🔗 View Repository

Model Announcer V1 only - Model self-awareness

Automatically injects the current model name into the chat context so the LLM is self-aware.

🔗 View Repository

OpenCode Clarify V2 - Rewrite rough prompts into precise technical prompts

Rewrites rough, plain-language prompts into precise technical prompts with /clarify, /clarify:model, or a -clarify marker before you send them to the model. OpenCode 2 TUI plugin, ported from pi-clarify.

🔗 View Repository

Opencode Snippets V2 - Instant inline text expansion

Instant inline text expansion for OpenCode. Type #snippet anywhere in your message and watch it transform. Brings DRY principles to prompt engineering with composable, shell-enabled snippets.

🔗 View Repository

opencode-personality V1 only - A configurable personality and mood system plugin for OpenCode.

Give your AI assistant a distinct personality with customizable moods that drift over time.

🔗 View Repository

Openskills unverified - Alternative skills manager

Alternative skills management plugin for opencode with enhanced features.

🔗 View Repository

Plain V2 - Plain-English replies: everyday words, no jargon, no code in explanations

Puts the assistant into plain-English mode. While active, a short ruleset is injected into the system prompt so it explains its work in everyday words and full sentences, with no jargon or acronyms and no code in explanations. On by default in every session; /plain toggles it, with /plain-commit and /plain-review commands for commits and diffs. Small, dependency-free, inspired by caveman - inverted.

🔗 View Repository

toon-config V1 only - Loads TOON based, AGENTS.toon rather than Markdown

Use TOON based agent instructions. Command to analyize and improve agent instructions.

🔗 View Repository

TypeUI V1 only - Design systems, UI prompts, and layout variation guidance

Open-source design layer for OpenCode that provides curated design skills, UI prompts, and layout variation guidance so agents can generate more consistent interfaces. Includes an OpenCode setup guide (linked as Docs).

🔗 View Repository  •  📖 Docs

➕ Add a Plugin

🎨 THEMES (9)

Ayu Dark - Port of the popular Ayu Dark color scheme with golden yellow accent.

Port of the popular Ayu Dark color scheme with golden yellow accent.

🔗 View Repository

Charcoal - Deep-black grayscale theme for OpenCode — no hues, all shades of gray

Charcoal is a pure grayscale theme for OpenCode with a near-black (#0d0d0d) background. All 56 theme color keys use shades of gray — no hues anywhere. Designed to match a minimal dark terminal aesthetic.

Also available for bat and Ghostty in the same repository.

🔗 View Repository

Lavi - A soft and sweet colorscheme for Opencode and 15+ other apps

A soft, sweet dark theme for Opencode with rich purple tones and carefully tuned syntax and diff colors. Part of the Lavi colorscheme family, which also provides matching themes for Neovim, Alacritty, Ghostty, Kitty, Wezterm, Zellij, and other tools, with Nix flake and home-manager support.

🔗 View Repository

MiMoCode Theme - Warm dark/light OpenCode theme with signature orange accents.

MiMoCode is a warm dark/light theme for OpenCode with signature orange accents, ported from the Zed MimoCode-theme. Dark mode uses warm-black surfaces, light mode uses warm cream surfaces.

🔗 View Repository

Moonlight - A moonlight color theme for OpenCode based on the popular moonlight-vscode-theme.

A moonlight color theme for OpenCode based on the popular moonlight-vscode-theme by atomiks. Features a dark, cool-toned palette inspired by moonlight aesthetics.

🔗 View Repository

OpenCode Light Themes - A curated collection of 21 light color themes for OpenCode

The largest collection of light themes for OpenCode. Includes 21 carefully crafted light themes converted from popular VS Code themes — scaefy-light (Atom One Light), Solarized Light, Vivid Light, Coffee Cream, GitHub Light, Quiet Light, Bluloco Light, Brackets Light Pro, NetBeans Light, Ysgrifennwr, Hop Light, GitHub Plus, Classic Light, HC Flurry, all Milkshake variants, and more. Bilingual (EN/TR) documentation and one-command install script included.

🔗 View Repository

opencode-material-theme - Material Design 3 themes for the OpenCode TUI.

Material Design 3 themes for the OpenCode TUI: 194 static JSON themes across 9 variants x 12 hues, plain + OLED.

🔗 View Repository

Poimandres Theme - Poimandres theme

Poimandres theme for opencode.

🔗 View Repository

VS Code Themes - Ports of all built-in VS Code themes (Modern, Plus, VS, High Contrast).

OpenCode ports of all built-in VS Code themes from the vscode.theme-defaults extension. Includes vscode-modern, vscode-plus, vscode-vs, and vscode-hc with automatic dark/light variant selection. One-line install script included.

🔗 View Repository

➕ Add a Theme

🤖 AGENTS (29)

Agents ₍₁₅₎

Agentic - Modular AI agents

Modular AI agents and commands for structured software development with opencode.

🔗 View Repository

Claude Subagents - Claude Code subagents

Comprehensive reference repository for production-ready Claude Code subagents.

🔗 View Repository

deliberation - Ask GPT, Gemini, Grok, or OpenRouter for a second opinion or a fix, as expert subagents over MCP

Get a delegated second opinion or an actual fix from GPT (Codex), Gemini, Grok (xAI), or OpenRouter. Seven expert subagents - Architect, Plan Reviewer, Scope Analyst, Code Reviewer, Security Analyst, Researcher, and Debugger - plus the ask-gpt, ask-gemini, ask-grok, ask-openrouter, ask-all, and consensus commands. Each one can advise (read-only) or implement. ask-all asks every model at once and compares the answers; consensus runs an arbiter loop until the models agree.

🔗 View Repository

Gem Team - Self-Learning Multi-agent orchestration harness for spec-driven development and automated verification.

Self-Learning Multi-agent orchestration harness for spec-driven development and automated verification.

🔗 View Repository

Goblin-Logic-Manager - Model-aware OpenCode agent with skill routing, bounded execution, durable closeout, and cold-session recovery.

A model-expandable OpenCode development-agent system that separates stable development behavior from thin model-specific overlays, with skill routing, bounded waits, evidence-backed verification, durable chunk closeout, and cold-session recovery.

🔗 View Repository

handoff - Delegate tasks to OpenCode / DeepSeek V4 right inside your Claude Code / Codex sessions.

Cross-agent task dispatcher — hand off work to DeepSeek V4 (via OpenCode), Codex, or Claude Opus without leaving your current session. Runs in the background; result comes back automatically.

🔗 View Repository

NERV - Invisible engineering infrastructure for AI agents

Spec-Driven Development pipeline, A2A task delegation hub, persistent semantic memory (MAGI), and 9 specialized subagents for OpenCode. Scaffolds 45+ files via `nerv init` including SDD skills, slash commands, MCP servers, and lifecycle plugins.

🔗 View Repository

Opencode Agents V1 only - Enhanced workflows

A set of opencode configurations, prompts, agents, and plugins for enhanced development workflows.

🔗 View Repository

Python Expert Agent for OpenCode - Python Expert Agent toolkit for OpenCode with subagents and skills

A custom configurable agent toolkit includes 1 Primary Custom Agent python-expert with intelligent skill loading, Predefined Specialized Subagents for Code generation, review, testing, and exploration, On-Demand Skills for Python development projects, Context Files for Standards, patterns, and security guidelines.

🔗 View Repository

Redstone - AI-built Minecraft plugins

an Opencode agent that simplifies Minecraft plugin development and deployment.

🔗 View Repository

review-pro - Tiered code review - triage picks the relevant specialists, they run in parallel, synthesis returns one verdict

Splits code review into three stages. A triage pass classifies the diff and selects which of thirteen specialist reviewers to dispatch - security, correctness, craft, ai-antipatterns, dry, performance, backend, frontend, a11y, db, api-contract, tests, spec - each with its own rubric and scoped context. The selected reviewers run in parallel, then a synthesis pass dedups overlapping findings, resolves conflicts by domain ownership, calibrates severity, and emits one verdict. Sixteen stack packs add language and framework specific signals from the repository's .review-pro/ directory.

🔗 View Repository

SEC-AF - Multi-agent security auditor — orchestrates a 250+ agent swarm that proves exploitability, SARIF for CI

Multi-agent security auditor. Orchestrates a 250+ agent swarm to find vulnerabilities and prove they're exploitable — not just flag patterns. SARIF output for GitHub Actions integration. Apache 2.0.

🔗 View Repository

server-manager - Non-blocking background server management — start, track, and stop local dev servers without blocking your AI session.

Start, track, and stop local dev servers in the background — without blocking your AI conversation. Supports Node.js, Python, Go, .NET, Java, Rust, PHP and any framework that runs a dev server. JSON state persistence, per-project log files, HTTP health checks, and log tail/grep for debugging.

🔗 View Repository  •  📖 Docs

StringRay - OpenCode orchestration framework with AI agents

A production-ready orchestration framework for OpenCode with multi-agent coordination, enforcement pipeline, codex compliance, and intelligent task routing. Includes 21 built-in processors, 60+ codex validators, and autonomous reporting.

🔗 View Repository

SWE-AF - Autonomous software engineering agent team — orchestrates a 400+ agent swarm for planning, coding, testing, and review

Autonomous software engineering agent team. Orchestrates a 400+ agent swarm across the full development lifecycle from a single API call. Apache 2.0.

🔗 View Repository

Skills ₍₁₄₎

AwareCode - Documentation-aware coding protocol for OpenCode

A 5-phase meta-skill that enforces documentation-first coding. Scans project docs before writing code, remembers root causes of errors, reviews code quality and security, synchronizes docs after changes, and outputs structured summaries. Designed for OpenCode's skill system.

🔗 View Repository

crowdsource-debugger - Debug stubborn errors with community help

Stops agents from going in circles. When stuck on the same error 3+ times, guides searching Stack Overflow, GitHub Issues, Reddit, and Discourse forums for vetted solutions.

🔗 View Repository

Emdash Skills - 14-category autonomous product-building OS

Turns one-line prompts into deployed products on Cloudflare Workers. 18 specialized agents (architect, test-writer, deploy-verifier, security-reviewer, visual-qa, etc.) self-coordinate through parallel build phases. Full-stack skills covering architecture, planning, quality verification, brand, media, observability, and more. Compatible with Claude Code, OpenCode, Gemini CLI, and Windsurf.

🔗 View Repository

iamhumans - Teaches Claude to stop performing empathy and start being present. 99/100 on a 100-case eval.

An opencode skill (~200 lines) that removes sycophancy, lecturing, and performed empathy from Claude's emotional responses. Teaches the shape of human conversation — when to be short, when to sit with something, when the right reply is "oh." Ships with a 100-case eval corpus and a baseline comparison showing +89.4 point skill delta over default Claude (1/20 PASS baseline vs 99/100 with skill). Load for human-shaped conversations (grief, decisions, vent, small talk). Don't load for code generation.

🔗 View Repository

llm-roundtable - Six models from five vendors vote on choices, plans and PRs when an unattended agent would stop to ask you

Agent Skill plus a stdlib-only Python script (MIT). When an unattended coding agent would stop to ask its user to choose, approve, or review, it convenes a council through each vendor's CLI: Claude Fable and Opus (claude), GPT (codex), Grok (grok), Gemini (agy), and the newest free OpenCode model (opencode, no login needed). Round 1 is blind, round 2 is an anonymous cross-examination (M1..Mn), peers rate each other's evidence 0-2, and a weighted vote decides with no chairman model; an undecided council falls back to a safe default. Seats run read-only and sandboxed. Review mode returns APPROVE/CHANGES/REJECT for plans and PRs. Runs from OpenCode as the host too; --private drops free-model seats that may train on prompts.

🔗 View Repository  •  📖 Docs

MorningAI - AI news tracking skill

AI news tracking skill that monitors 80+ entities across 6 free sources (Reddit, HN, GitHub, HuggingFace, arXiv, X/Twitter). Generates scored daily reports with infographics and message digests. Works via SKILL.md — no Docker or servers required.

🔗 View Repository

OpenCode Agent Orchestration Kit - Structured product-development workflows

A reproducible OpenCode configuration kit with specialized agents, structured handoffs for research, design, specification, implementation, and review, plus local skills, validation checks, safe installation, and optional design integrations.

🔗 View Repository

opencode-mcp-registry - Register MCP servers in opencode

Guides agents through installing and registering MCP (Model Context Protocol) servers. Supports pre-built binaries, pip, npm, Docker, and remote HTTP endpoints with proper auth handling.

🔗 View Repository

opencode-roundtable - Multi-agent roundtable discussion skill with per-member model routing and meeting minutes

A roundtable discussion (圆桌讨论) skill for opencode — TUI, CLI and Desktop. Pick members one by one via inline cards (role, persona, model — any provider); each member is a real subagent run via opencode run -m provider/model, a neutral host summarizes each round, and on termination a structured meeting-minutes Markdown file is written. Uses only built-in primitives — no plugin required.

🔗 View Repository

PDCA Skills - PDCA orchestration skills: PLAN → DO → CHECK → ACT with evidence-based verification

A family of four OpenCode skills that drive a task through a PDCA cycle (PLAN → DO → CHECK → ACT) with transition gates, evidence-based verification, escalation accounting, durable status, and recovery after compaction.

• pdca — domain-neutral cycle; the task's PLAN selects the means and the verification method.
• pdca-coder — stack-agnostic executor for software tasks (code, tests, docs, refactoring, config).
• pdca-dotnet — the same cycle tuned for .NET/C# repositories.
• pdca-collection — a batch of tasks run as parallel PDCA lanes (worktree per group, --no-ff merge of completed groups).

Roles (planner, coder, check, scout, escalate) are bound by the host, so the skills stay model- and provider-neutral.

🔗 View Repository

PPTX from Layouts - PowerPoint from slide master layouts

Generate consultant-quality PowerPoint presentations from markdown outlines using your template's actual slide master layouts, not text overlays. Includes template profiling, visual types, and surgical edit mode.

🔗 View Repository

sensitive-operations - Gatekeeper for privileged operations

Detects sensitive operations (passwords, sudo, tokens, destructive commands) and delegates them to the user for manual execution with verification.

🔗 View Repository

ShipFrame - AI coding workflows for OpenCode, Claude Code, and Codex

AI coding workflow toolkit with installable OpenCode skills, converted agents, project profiles, release evidence, and disciplined PR/MR workflows.

🔗 View Repository  •  📖 Docs

uv-skill - Use uv for all Python package management

Mandates using uv for ALL Python dependency, virtual environment, and project management tasks. Replaces pip, poetry, pipx, pyenv, and virtualenv.

🔗 View Repository

➕ Add an Agent

🛠 PROJECTS (191)

Agents ₍₁₎

Goblin-D.M.T. V2 - Model-agnostic OpenCode runtime for bounded exploration, evidence-driven decisions, and user-gated execution.

An actor-driven OpenCode control runtime that turns model exploration, branching, uncertainty, and tool use into bounded work with lazy skill routing, persistent state, evidence-driven decisions, verification, checkpointing, and user-gated execution.

🔗 View Repository

Skills ₍₁₀₎

Aide - Spec-driven workflow as opencode skills, with a local dashboard that runs specs across projects

Skills (/aide-create, /aide-analyze, /aide-implement, /aide-archive) that write each change as four Markdown files — description, analysis, solution, status — before any code, then implement it test-first. The installer sets the skills up for opencode alongside Claude Code, Codex and GitHub Copilot. A local dashboard queues specs across projects and runs each step headless on the model chosen for it, including opencode's provider/model models, and merges only once the project's tests pass.

🔗 View Repository

assay - Deterministic QA that drives every control on a web page and reports where it contradicts itself

assay opens a web page in Chromium through Playwright, drives every control it finds, and reports where the page contradicts itself. Runs are deterministic and use no LLM, so the same page gives the same result every time. For opencode you install its skill by copying checking-a-page into .opencode/skills/, and it tells the agent to run assay after it changes a page and report the result. MIT licensed, Python 3.10+, installed from PyPI as assay-ui.

🔗 View Repository

Claude Learn - Self-improving behavioral playbook plugin for Claude Code

Self-improving behavioral playbook plugin for Claude Code. Captures corrections, failures, and discoveries as scored rules that strengthen or decay over sessions — a closed feedback loop for continuous improvement.

🔗 View Repository

Comfanion Workflow - AI-assisted development workflow with agents, skills, and semantic search

Complete workflow system for AI-assisted development. Includes specialized agents (analyst, PM, architect, developer), reusable skills, slash commands, semantic code search with local vectorizer, and sprint management with Jira integration.

🔗 View Repository

growth-architect-foundry - Ad-agency growth ops skill for OpenCode: audits, client reports, ROAS maths, human-approved live writes

Deterministic growth-ops commands for ad agencies (audit, diagnose, EN/AR client reports, COD-aware economics) as an Agent Skill plus AGENTS.md. Live Google/Meta budget or pause changes are enforced in code to need a person typing the campaign key at a terminal, so an agent cannot self-approve. Installs into .agents/skills with one script and is checked against the real OpenCode CLI in CI.

🔗 View Repository

Hipocampo - Dual memory system with BIRE search for AI agents

A dual-memory system (PostgreSQL + pgvector + Gemini Embeddings) for AI coding agents. Implements the BIRE algorithm (Búsqueda Integrada por Relevancia Expansiva) — unified semantic + lexical search with auto-tagging, hybrid scoring, GIN trigram indexes, and cross-system 768d embeddings. Ships as an opencode skill with SKILL.md, Python scripts, and full DDL schema.

🔗 View Repository

PPT Design Skill - One-liner PPT generation —40,000+ styles, AI images, fully editable .pptx

AI-powered PPT generation skill for OpenCode, Claude Code, Codex, Cursor, and 13+ AI coding platforms. Three modes: FreeStyle (one-liner quick generation), Build Script (pixel-perfect delivery with python-pptx), and VI Build (enterprise template compliance). Features 40,000+ style combinations via natural language (--style "dark cyberpunk"), 5 AI image engines (Seedream/GPT Image/DALL-E/Wanx/Kimi), 10 diagram types, 28 design quality upgrades, advanced effects (3D shapes, gradient text, frosted glass, morph transitions), and CJK font support. Outputs fully editable .pptx files directly —356x faster than HTML-to-screenshot approaches. Includes a component library with 5,500+ SmartArt/GroupShape templates and one-command installation for all platforms.

🔗 View Repository

Professor - Fleet-boost discipline layer for Claude Code, Codex & OpenCode agents

An LLM-harness fleet boost framework that turns your AI coding agents into a disciplined engineering team you can see, message, and hold to the rules. Ships a compiled .opencode/ layer (agents, commands, skill symlinks, opencode.jsonc with pinned denies) generated from its Claude sources, plus pfm, a Go CLI/TUI that lists OpenCode chats alongside Claude Code and Codex sessions.

🔗 View Repository  •  📖 Docs

SDLC-Assist - SDLC phase router skill - infers the phase from the repo, asks one question, hands off

Agent skills that tell opencode which SDLC phase a request is in (analysis, planning, development, testing, deployment) from what the repo already contains, show the evidence, ask one confirmation and route to spec-driven-development, a QA gate and a release skill. Node only, no dependencies.

🔗 View Repository

UIZZE - UI quality workflow for OpenCode with an optional authenticated MCP

Help OpenCode avoid generic UI with a free anti-UI-slop Agent Skill and an optional authenticated UIZZE MCP for focused UI references from 800,000+ real web and iOS screens.

🔗 View Repository

MCP Servers ₍₃₀₎

/call-me - Let OpenCode ring your iPhone, read you a question and return your spoken answer

A hosted remote MCP server for OpenCode. The agent can ring your iPhone, read a question or status update aloud, and get your spoken reply back as text, or send you a text and wait for your answer. Useful when a long session needs a decision while you are away from the keyboard. Needs the /call-me iPhone app, which gives you the number the agent calls.

🔗 View Repository  •  📖 Docs

Agent Nexus - Remote MCP registry of 700+ APIs, MCP servers and CLIs, each live-probed so agents pick tools that actually work

Remote MCP server and discovery registry for AI agents. Every entry (APIs, MCP servers, CLIs) is continuously probed for liveness, capability and response schema, so an agent can search, compare and pick a tool based on live evidence instead of a README. Agents can also follow an entry and get alerted when it goes down or changes shape. Works with opencode via remote URL (https://agentnexus.app/api/public/mcp, free tier without key) and publishes a ready-to-use config at https://agentnexus.app/.well-known/opencode.

🔗 View Repository

Agent QA - The self-improving QA agent for software teams

Open-source QA harness for authoring and running natural-language web and mobile tests, with persistent memory, self-healing, a dashboard, Agent Skills, and an MCP server. Configure its `agent-qa mcp` command in OpenCode to expose tools for creating, executing, inspecting, triaging, and repairing tests.

🔗 View Repository

AgentDeals - MCP server aggregating free tiers, credits, and referral codes across 1,500+ developer tools

Remote MCP server that surfaces free tiers, credits, trials, and referral codes for developer infrastructure — Railway, Vercel, Fly, OpenAI, Anthropic, and 1,500+ other vendors. Exposes four tools (search_deals, plan_stack, compare_vendors, track_changes) so an agent can find the cheapest way to ship a given stack. Works with opencode via remote URL (https://agentdeals-production.up.railway.app/mcp). Also tracks pricing changes over time.

🔗 View Repository

ax - Local telemetry graph for AI coding agents

Local-first telemetry and taste graph for AI coding agents. Ingests OpenCode, Claude Code, Codex, Cursor, and Pi session data plus OTLP, then exposes recall, cost, skill/hook usage, churn, and workflow-extraction views through a CLI, dashboard, and MCP server.

🔗 View Repository

Cortex - Codebase intelligence engine for coding agents via MCP

Indexes your project into a structured knowledge graph of files, symbols and dependencies using tree-sitter AST parsing, and exposes it to OpenCode over MCP. Includes semantic search, dependency-graph context engine, architecture analysis, persistent project memory, task planning, and specialized agents (architect, reviewer, security, tester) producing unified reports. TypeScript, Node 22+, MIT.

🔗 View Repository

db-semantic-mcp - Read-only PostgreSQL and SQL Server schema context for OpenCode agents

A local MCP server that gives OpenCode agents table and column metadata, database comments, bounded sample data with redaction controls, and optional semantic schema search. Supports PostgreSQL and SQL Server without arbitrary SQL execution or a vector database.

🔗 View Repository

Dialbot MCP - Local MCP browser control for OpenCode with an unpacked Chrome extension

Control Chrome tabs from OpenCode through a local stdio MCP server and a Chrome Manifest V3 extension. Includes fast DOM and selector actions, plus visual screenshot, mouse and keyboard control. Also works with Pi and other MCP stdio clients.

🔗 View Repository

dsh-skills-anywhere - Serve the skills installed for 60+ agents to OpenCode over MCP, zero copies

MCP server that reads the Agent Skills (SKILL.md) already installed for OpenCode, Claude Code, Codex, Cursor, Gemini CLI and 60+ other agents, plus Claude Code plugin marketplaces and any GitHub skills repo, and exposes them as find_skills/open_skill tools and skill:// resources without copying or symlinking.

🔗 View Repository

Employed by Guaq AI - Live job search MCP server and skills for OpenCode, from about 5,000 verified employer ATS boards

Remote MCP server plus Agent Skills for searching live job postings read from employers' own applicant-tracking systems, researching an employer's hiring and tracking applications. Signs in with OAuth through opencode mcp auth, so no API key is needed. Every result links to the employer's own apply page.

🔗 View Repository  •  📖 Docs

GoodMemory - Local-first, auditable memory for OpenCode through MCP

Gives OpenCode durable user and project memory across sessions through a standalone MCP server. Uses SQLite by default, keeps recall inspectable, and makes memory writes opt-in.

🔗 View Repository  •  📖 Docs

Kranz - Inspect and control live development services from OpenCode over MCP

Kranz runs local development stacks with a terminal UI and CLI. Its built-in stdio MCP server lets OpenCode inspect services, readiness, logs, ports, actions, and run history, then perform authorized lifecycle operations against the same live runtime. The companion kranz-services Agent Skill teaches the agent to reuse existing runtimes and preview scoped changes.

🔗 View Repository  •  📖 Docs

Lians - Local-first memory for OpenCode and any MCP-compatible AI agent

Gives OpenCode durable memory across chats through its native MCP support. Ships a ready-to-copy OpenCode configuration and five-tool starter profile for remembering, recalling, inspecting, correcting, and explicitly erasing local memory. No Lians account, API key, model download, or cloud service is required; memory stays in a local SQLite database and can be shared with other compatible agents.

🔗 View Repository

magents - Read, hand off to, message, and spawn OpenCode, Claude Code, Codex, Cursor, and Gemini sessions from one MCP server

Rust MCP server and CLI over the session files coding agents already write. From OpenCode you can list and read other live sessions, search transcripts, hand off compact state into another chat, send into a specific session, or spawn an independent headless session that replies through a shared inbox. Reads OpenCode's session database directly; `magents install --all` registers the MCP server in opencode.json and adds the magents skill. Homebrew, APT, binaries, and Docker. MIT licensed.

🔗 View Repository

Mnemoverse - MCP memory server that re-ranks recall from reported outcomes, backed by a hosted engine

MIT-licensed MCP server for agent memory, backed by a hosted engine that re-ranks recall from reported outcomes rather than similarity alone. OpenCode can use it as a local MCP server (npm `@mnemoverse/mcp-memory-server`) or as a remote MCP server at https://mcp.mnemoverse.com/mcp with OAuth sign-in, and the same memory is shared with Claude Code, Cursor, VS Code, and ChatGPT. The engine is hosted and needs a Mnemoverse account, with a free tier.

🔗 View Repository  •  📖 Docs

MonadicSharp - Railway-Oriented Programming for C#

MCP server + Green Code Agent + MonadicForge CLI analyzer for C#/.NET. Includes 5 MonadicSharp resources, 21 code generation scenarios, inline ROP compliance review, and skills for green-code-advisor and saas-module-design. .NET 8+, MIT license.

🔗 View Repository

myc - Local task queue and decision memory for coding agents, wired into opencode with one command

A local task-and-memory layer for coding agents (opencode, Claude Code, Codex, Kimi). One SQLite file next to the repo holds a task queue with dependencies, blockers and atomic claims, an oplog of decisions with hybrid search, and a code index; `myc wire` installs the opencode plugin and MCP server. No network calls of its own, no API key, MIT.

🔗 View Repository

Opencode Chat Export - MCP server to export chat history to Markdown

MCP server that exports OpenCode chat history to Markdown files with tool calls, reasoning, and cost summaries. Export current session, specific sessions by ID, or bulk exports. pip installable.

🔗 View Repository

Openhook - Give OpenCode a callback inbox to test webhook integrations through MCP.

Open-source callback inbox with a hosted Streamable HTTP MCP endpoint and an OpenCode setup recipe. Create a disposable inbox, wait for a webhook, inspect the received body and headers, then clean up the test inbox. Private per-inbox tokens control access to captured events. Also supports email and DNS capture, with self-hosting available.

🔗 View Repository

PatchWarden - Local-first MCP safety layer for OpenCode and other AI coding agents

PatchWarden is a local-first MCP safety and verification layer for AI coding agents. It provides workspace confinement, command allowlists, scope-violation detection, and auditable task evidence for OpenCode and other MCP clients.

🔗 View Repository  •  📖 Docs

PickySteve - Local skill router and context picker for coding agents

A cheap local model routes each prompt to the right skill via hybrid retrieval (BM25 + embeddings, RRF-fused) with cross-encoder reranking and an LLM judge, then assembles a minimal context bundle for the execution model. Runs fully offline on local Ollama by default and ships a fail-closed prompt-injection gate (ONNX classifier) that scans both requests and retrieved documents. Exposes an MCP stdio server and an OpenAI-compatible proxy, with a one-command installer wiring 18 coding agents including Claude Code, Cursor, and OpenCode.

🔗 View Repository

PriceWin - Remote MCP server and skill for live hotel and flight prices across Booking.com, Agoda, Trip.com and Traveloka

Compares live hotel and flight prices in USD across Booking.com, Agoda, Trip.com, Traveloka and Google Flights through one remote MCP server (https://mcp.price.win/mcp, no account or API key). Add it to opencode.json as a remote MCP server and copy the bundled pricewin-travel-search skill to ~/.agents/skills, which OpenCode loads automatically. Results name their source and link to it for booking.

🔗 View Repository

Prompt to Asset - Generate visual assets across 30+ image models via MCP

MCP server that generates production-ready visual assets (app icons, favicons, OG images, logos, wordmarks) by routing prompts across 30+ image generation models. Three execution modes: inline SVG, external prompt, and full API generation. Zero API key required for first run via Pollinations and Stable Horde free tiers.

🔗 View Repository

SCVD General Store - Skills and MCP tools for checking x402 endpoints before payment and signed receipts afterward

An evidence observatory for agentic commerce with portable skills and a remote MCP service. OpenCode can load the verification skill and run a free endpoint preflight without a wallet or API key. Guidance covers signed-receipt checks and the limits of unpaid observations. Optional paid observations require an explicit spending decision; installation grants no payment authority.

🔗 View Repository  •  📖 Docs

semantic-search-mcp - Semantic code search MCP server - find code by meaning across thousands of files

Adds semantic code search to OpenCode. Find code by meaning across 6,900+ files in 2 seconds. 100% local 80MB model (bge-small-en-v1.5), zero API calls. Function-level extraction for PHP, JS, TS, Python, Go, Rust, Java. Hybrid scoring (70% embedding + 30% keyword), TUI config wizard, live progress bar, per-project caching.

🔗 View Repository

showagent - Browse OpenCode history and copy conversation messages between coding agents.

A terminal picker, CLI, and MCP server for local coding-agent sessions. Browse OpenCode sessions alongside Codex, Claude Code, Gemini CLI, jcode, and Pi; preview message transfers and create new native sessions. OpenCode operations use the installed OpenCode CLI. Transfers preserve original sessions and copy extracted user and assistant messages, excluding tool and runtime state.

🔗 View Repository

SpringBrand - Remote MCP server with go-to-market tools for agents

Hosted remote MCP server that gives OpenCode go-to-market tools: social listening, traffic and SEO research, lead discovery, and creator discovery.

Exposes five tools — search_tools, get_tool_schemas, execute_tools, manage_connections, and get_execution — so the agent discovers and calls capabilities on demand instead of loading every tool schema into context.

Transport is Streamable HTTP. Auth is OAuth 2.1 with Dynamic Client Registration (RFC 7591) and PKCE, which OpenCode handles automatically on the first 401, so no API key goes in the config file.

🔗 View Repository  •  📖 Docs

Stacktree - Publish HTML from your agent to a private, unguessable link

Turns HTML an agent produced into a link a person can open — a report, a dashboard, a brief. Links are unguessable and uncrawlable by default, with an optional passcode and expiry, and republishing replaces the page in place so a shared URL stays current. Connects as a local MCP server, and the skill is fetchable over HTTP so it stays current without reinstalling. Free anonymous pages need no account at all; agents with a wallet can pay per page over x402 instead of signing up.

🔗 View Repository  •  📖 Docs

SuperLocalMemory - The governed memory brain for every AI agent. It lives on your machine.

Local-first memory for AI agents, added to OpenCode as an MCP server in opencode.json. It recalls what was saved, keeps it true over time, and answers that it does not have the answer instead of guessing.

🔗 View Repository

Wuniq - Local-first structured project memory for OpenCode through MCP

Connects OpenCode to the free Wuniq desktop app through @wuniq/mcp. Wuniq keeps decisions, constraints, intent, and human reasoning in .wuniq files beside project files, then exposes progressive read and sync tools to the model. The public repository includes a validated opencode.json setup and AGENTS.md workflow. No account, cloud storage, or telemetry.

🔗 View Repository

Clients & GUIs ₍₆₀₎

Agent of Empires - Multi-session TUI for OpenCode

A terminal UI for managing multiple OpenCode sessions in tmux with git worktree integration and Docker sandboxing.

🔗 View Repository

Agent Sessions - Local OpenCode session-history browser

Local-first macOS app for browsing and full-text searching OpenCode session history. It reads local OpenCode SQLite history in read-only mode, opens old runs as readable timelines, and can resume sessions where the OpenCode CLI supports it.

🔗 View Repository  •  📖 Docs

Agent Workbench - Local web UI for OpenCode and three other agent CLIs, with one shared history

Runs OpenCode through its own opencode serve process, next to Claude Code, Codex and Antigravity CLI, each in a browser tab. Shows OpenCode sessions as conversation cards read from its database in read-only mode, lets you answer its questions and see its permission waits from the UI, hands a conversation over to another CLI and searches across all of them. Never opens auth.json or opencode.json, no telemetry.

🔗 View Repository

agent-manager - Terminal UI that runs OpenCode and other coding-agent CLIs side by side in persistent tmux sessions

Terminal UI that runs OpenCode alongside Claude Code, Codex and other coding-agent CLIs, each in its own persistent tmux session. It launches the installed OpenCode CLI unmodified, so login, config and MCP servers carry over. One list shows every session's live status, prompts go into a pane without attaching, a session can spawn into its own git worktree, and a full-file diff review sends line comments back to the agent. Runs on macOS, Linux and Windows via WSL2. Apache-2.0.

🔗 View Repository  •  📖 Docs

agenttrace - Local TUI for OpenCode session cost, health, and trace reports

Local-first Bubble Tea TUI and report CLI for inspecting OpenCode and other AI coding-agent logs with cost, tokens, latency, tool failures, anomalies, health gates, diffs, and reports.

🔗 View Repository

AionUi - Free, open-source multi-agent desktop client that gives OpenCode (and other CLIs) a GUI.

Cross-platform (macOS/Windows/Linux) desktop app that wraps OpenCode and other agent CLIs in a chat-based GUI. Supports any API key and multiple LLM providers with API-key polling, multi-session and multi-agent workflows, MCP and ACP protocols, file management and preview, code diff view, scheduled tasks, local data storage, and remote WebUI access.

🔗 View Repository  •  📖 Docs

Calyx - Native macOS terminal for running and supervising OpenCode and other coding agents in parallel

Calyx is an MIT-licensed native macOS terminal built on libghostty. It runs OpenCode alongside Claude Code, Codex, Hermes, Grok, and pi in parallel panes, shows each agent's status and subagents in a sidebar, lets you review diffs and send line comments back to the OpenCode pane, and provides a built-in MCP server for cross-pane messaging.

🔗 View Repository

Caprock - Local dashboard showing OpenCode sessions next to Claude Code and Codex, with cost per repository

Reads OpenCode's own database read-only and shows its sessions on the same screens as Claude Code, Codex and DeepSeek Harness, with an all/opencode filter. Costs are shown as OpenCode reports them. Searchable history, single Go binary, no telemetry.

🔗 View Repository

Chronotile - Local usage analytics dashboard for OpenCode — cost, tokens, models, agents, tools, and sessions

Native desktop app (Tauri 2, Rust backend, React UI) that turns OpenCode's local SQLite history into a dashboard — spend, token breakdown including reasoning and cache, per-model, per-agent, per-tool and per-skill analytics, project and file activity, quota burn rate, a full-year spend heatmap, and a session browser with full transcripts. OpenCode records $0 for subscription and OAuth-plan traffic, so Chronotile also estimates cost from models.dev rates and shows reported and estimated figures side by side. Databases are opened read-only, an incremental rollup cache keeps queries instant on multi-gigabyte histories, and no usage data ever leaves your machine.

🔗 View Repository  •  📖 Docs

Claudescope - Local, read-only web UI to browse, search, and analyze OpenCode (and other agents') sessions

Local-first CLI that serves a web UI to browse, read, search, and analyze AI coding-agent transcripts — OpenCode sessions alongside Claude Code, Codex, Junie, pi, and Copilot CLI. Sessions are merged by working directory; includes DuckDB full-text search and token-cost analytics. Read-only and cross-platform, distributed as a single npm package (MIT).

🔗 View Repository

ClawMetry - Zero-config observability dashboard for OpenCode and 20+ agent runtimes

Local web dashboard that auto-detects OpenCode session history and shows live sessions, token usage, costs, tool calls, and transcripts alongside 20+ other agent runtimes. One pip install, no configuration.

🔗 View Repository  •  📖 Docs

CodeWalk - Native cross-platform Flutter client for OpenCode server mode

A fast, native Flutter client for OpenCode server mode with realtime streaming chat, speech-to-text on all platforms, multi-server profiles, model selection, and a Material 3 responsive UI for Linux, macOS, Windows, Web, and Android.

🔗 View Repository

Crit - Review OpenCode plans and diffs in a local browser UI and send comments back to the agent.

Crit opens agent plans, code diffs, HTML artifacts and a running dev server in a browser review UI. You leave inline comments and they go back to the agent as structured feedback. Ships an OpenCode plugin (v1 and v2) and skills. Runs locally as a single binary.

🔗 View Repository

Deck - Local cockpit for OpenCode-powered autonomous agents with secure sandboxes and live desktop visibility.

Open-source desktop cockpit that runs OpenCode-integrated agents in isolated Docker sandboxes with chat plus live noVNC desktop monitoring, tool permission controls, and a Pilot bridge suite for orchestrating agents from Telegram, Slack, WhatsApp, and more.

🔗 View Repository  •  📖 Docs

dotpals - A desktop pal and notch that shows what your OpenCode sessions did, in plain words

An open-source desktop companion that follows OpenCode through an OpenCode plugin and integrates with Claude Code, Codex, Cursor, Gemini CLI, and Copilot CLI through their respective hooks or local logs. Each request becomes a few readable with a live view of the file being edited in a notch at the top of the screen, and flags for .env changes, force-pushes and two agents editing the same file. Runs locally, no telemetry.

🔗 View Repository

Duplex - A browser shared by a human and an AI — same tabs, same live session

Electron desktop browser with a bundled stdio MCP bridge (24 tools — snapshot, click, type, screenshot, tabs, annotations) plus an opencode mirror plugin that streams session events into a side panel and injects panel messages back into the same session. Built-in multi-protocol model support (OpenAI / Anthropic / Responses / Gemini) and an Esc takeover for the human.

🔗 View Repository

Handy - Speech to Text

Easy Open Source Speech to Text.

🔗 View Repository

Heym - Visual AI workflows with OpenCode coding agents and Go usage tracking

Heym is a source available AI workflow platform you can self host. Its OpenCode Go node runs the OpenCode CLI against GitHub repositories and can return patches or open pull requests. It also provides native OpenCode Go credentials, live model selection, and subscription usage tracking.

🔗 View Repository  •  📖 Docs

Infiknit - Local stdio launcher for Infiknit desktop image and video canvas workflows

Connect OpenCode to Infiknit on this computer. Generate images and videos into a project or build canvas workflows. Keep Infiknit open. Cloud agents cannot reach it. Official registry io.github.Synthenova/infiknit.

🔗 View Repository  •  📖 Docs

intentic - Self-hosted browser workspace that runs OpenCode agents in parallel sandboxes

A browser workspace for running coding agents on hardware you own, with OpenCode connected over ACP as a chat provider. Each agent gets its own Docker sandbox and git worktree, terminals survive disconnects, any browser or phone reopens the same fleet, and every change is reviewed as a per-file, per-hunk diff before it lands.

🔗 View Repository

Ivy Tendril - AI coding orchestrator with plan-based lifecycle and verification gates

Open-source desktop app that orchestrates OpenCode, Claude Code, Codex, Antigravity, and Copilot through a structured plan-based lifecycle. Features verification gates, self-improving memory, human-in-the-loop checkpoints, and git worktree isolation. Local-first, agent-agnostic, stack-agnostic.

🔗 View Repository  •  📖 Docs

JustSessions - Native Mac workspace for finding, reading, and resuming OpenCode sessions.

Groups OpenCode and other coding-agent sessions by project, with searchable conversation history and embedded terminals that preserve the original CLI. Requires macOS 14 or later, Apple Silicon, and an installed, authenticated CLI.

🔗 View Repository

Kimaki - Discord bot controller

A Discord bot to control opencode sessions on any computer via Discord.

🔗 View Repository

LoopTroop V1 only - Local GUI orchestrator for autonomous AI coding using multi-model planning councils and isolated worktrees

Local-first, open-source GUI orchestrator that runs long-running AI coding tickets end-to-end. An LLM Council plans the work, tasks are decomposed into atomic units (Beads) executed in isolated git worktrees, and a Ralph Loop recovers from errors with a fresh context. Built on OpenCode.

🔗 View Repository  •  📖 Docs

MCP Voice Interface V1 only MCP - Talk to AI assistants

Talk to AI assistants using your voice through a web browser. Compatible with Claude Desktop and opencode.

🔗 View Repository

MiMo-Code - Voice-driven multimodal desktop coding agent for MiMo

A native desktop coding agent for the MiMo model family — Windows & macOS, no TUI — built on the OpenCode harness and adapted to be MiMo-first. Adds native multimodal understanding (image, PDF, video) plus voice dictation (ASR) and speech synthesis (TTS), with cost-aware model selection.

🔗 View Repository  •  📖 Docs

NextProb - Desktop notebook for vibe-coded websites, edited by your local AI coding-agent CLI (OpenCode and others)

Electron desktop app that turns a workspace folder of plain .html files into an AI-editable notebook. Agent-agnostic — edit notes with OpenCode or any other agent CLI in an embedded terminal. Notes are real HTML pages. Each note is a self-contained folder that can include a SQLite database, key/value store, attached files, and scripts — so it can grow from a static page into a sortable table, interactive dashboard, or research tool.

🔗 View Repository

OC Manager - Metadata TUI

Terminal UI for inspecting, filtering, and pruning OpenCode metadata stored on disk.

🔗 View Repository

oc-monitor - Real-time terminal dashboard for tracking OpenCode session costs and usage.

A terminal UI (built with Go and Bubble Tea) that reads your local OpenCode SQLite database to show live session activity, burn-rate charts, and cost breakdowns by agent, model, and project. Includes daily/hourly usage trends with a calendar heatmap, per-session token and tool-call inspection, full-text search across sessions, and JSON export. All data stays local — the only network call is an optional lookup to models.dev for model context-window sizes.

🔗 View Repository

Octto V1 only - Interactive browser UI for AI brainstorming

Interactive browser UI for AI brainstorming with multi-question forms, parallel exploration branches, and visual feedback.

🔗 View Repository

Open Agent - Self-hosted control plane

Self-hosted control plane for OpenCode agents with isolated Linux workspaces (systemd-nspawn), git-backed Library configuration, and multi-platform dashboards (Next.js web, SwiftUI iOS).

🔗 View Repository

Open Dispatch - Control agents from Slack, Teams, or Discord — locally or via Fly.io Sprites

Bridge app connecting chat platforms (Slack/Teams/Discord) to AI coding assistants. Run agents locally or spin up isolated Fly.io Sprites for parallel execution. Supports 75+ AI providers via OpenCode, Claude CLI, real-time output streaming, session persistence, and smart message routing.

🔗 View Repository

OpenChamber V2 - GUI for OpenCode

A fan-made web and desktop interface for OpenCode with VS Code extension, multiple sessions, and git worktrees management.

🔗 View Repository

OpenChamber Design Studio - Open, local alternative to Claude Design inside OpenChamber

An OpenChamber extension that turns design work into a chat loop. It opens your running app in a real Chrome window beside the conversation; Cmd/Ctrl-click any element to attach it to the chat as a reference (selector, markup, computed styles), and the agent edits the real project while the preview reloads. Ships Design Mode (/design), per-project harness plugin, DESIGN.md / tokens.css authority with a vendored design-system preset catalog, project scaffolding, and a .design-studio.json project config. MIT, install from a git URL via Settings → Extensions.

🔗 View Repository

OpenClient - Native iPhone and iPad companion for self-hosted OpenCode servers

Native SwiftUI client for browsing projects, resuming sessions, streaming chat, handling permission and question prompts, and following active work through Live Activities and widgets.

🔗 View Repository  •  📖 Docs

OpenCode Dashboard - Single-file live token usage dashboard with zero dependencies

A live usage dashboard served by one Python file — no framework, no build step, no dependencies. Token-first views of sessions, models, projects, tokens by changed file/subsystem, tokens per git commit, and a 52-week activity heatmap, with click-to-filter and JSON/CSV export.

🔗 View Repository

OpenCode Mobile V2 - Your OpenCode workflow in your pocket. Manage AI coding agents from Android and iOS.

OpenCode Mobile is a free, open-source, community-built companion app for OpenCode on Android and iOS. Connect to your self-hosted OpenCode server to start coding sessions, monitor agent progress, approve permissions, review code changes, browse files, access the terminal, and manage projects and AI models directly from your phone. Optional secure remote connectivity lets you stay connected to your development environment from anywhere, while your code and workloads remain on your own computer. No cloud-hosted development environment required.

🔗 View Repository

OpenCode Mobile - Android client for OpenCode — drive your self-hosted agent from your phone

Free, open-source (MIT) Android app that connects to an opencode server you run yourself. Stream coding sessions in real time, review file diffs, and approve tool calls from your phone. Install via the F-Droid repo or a signed APK.

🔗 View Repository  •  📖 Docs

Opencode Neovim - Neovim plugin

Neovim plugin for making convenient editor-aware prompts.

🔗 View Repository

OpenCode Sentry - Android companion: scan-to-pair live OpenCode console + screen-off alerts over Tailscale

The Android companion for OpenCode. Scan one QR printed by the self-hosted bridge and the official web console opens in-app: watch every session live, send prompts, follow reasoning, tool runs and streaming output, switch models — all directly over your own Tailscale network. A foreground service also rings, vibrates and speaks when a task finishes, needs permission, asks a question or fails, with a three-layer keep-alive (accessibility keeper, persistent Job, exact-alarm watchdog) that survives aggressive vendor-ROM clean-ups. Everything stays on your own tailnet — no third-party server. Not affiliated with the OpenCode team.

🔗 View Repository

OpenCode Sidebar V1 only - Tmux sidebar for managing OpenCode sessions

A tmux-backed sidebar launcher for OpenCode that keeps the session list, preview pane, and OpenCode TUI together in one workspace. It supports session recall, background sessions, search, project folders, deletion, and sound notifications.

🔗 View Repository

Opencode Telegram Bot V2 - Telegram bot client for OpenCode CLI

Run and monitor AI coding tasks from your phone while everything runs locally on your machine.

🔗 View Repository

Opencode Web V1 only - Browser-based access

Web interface for opencode - browser-based access to AI coding agent.

🔗 View Repository

opencode-kanban - Kanban-style task management for opencode

Kanban tool that allows you to stay in terminal.

🔗 View Repository

opencode-telegram V1 only - Control OpenCode sessions from Telegram

A Telegram bot that bridges your chats to an OpenCode server with real-time SSE progress, session management, and access control via user allowlists.

🔗 View Repository

OpenMate - Offline-first Android client for OpenCode, connect from anywhere

Native Android (Kotlin/Compose) client for a self-hosted opencode server. Connects over your local network or automatically through a cloud relay, so you can reach your agent from anywhere. A lightweight Bridge streams incremental, seq-based sync from opencode's own database into a local copy on your phone, so you can read and reply to sessions offline. Supports opencode v2 only.

🔗 View Repository

OpenTab - Terminal UI for browsing your OpenCode spend by month, day, project, session, and model

A terminal UI that reads OpenCode's local SQLite database (read-only) to show where your tokens and money went — by month, day, project, session, and model, with spend-over-time trends and drill-down into your most expensive sessions (model mix and subagent costs included). Standard-library Python only, offline by default. Also reads Claude Code, Codex, the Copilot CLI, and more.

🔗 View Repository

OpenWork V1 only - Desktop GUI for OpenCode workflows

Open-source alternative to Claude Cowork built on top of OpenCode. Provides a polished desktop UI for sessions, skills, plugins, and templates.

🔗 View Repository  •  📖 Docs

Orkas - Local-first multi-agent desktop workspace with OpenCode orchestration

Orkas is an open-source desktop AI workforce whose Commander coordinates specialist agents and can run OpenCode alongside Claude Code, Codex, and Cline from one chat. Conversations, files, agent configurations, and model keys stay local.

🔗 View Repository

P4OC (Pocket for OpenCode) - Android client for OpenCode

Native Android app for controlling OpenCode from your phone. Chat with AI, manage sessions, browse files with syntax highlighting, view diffs, and use an embedded terminal — all styled with a terminal UI aesthetic. Available on Google Play.

🔗 View Repository

Parallel Code - Orchestrate multiple AI coding agents simultaneously

Desktop app for orchestrating multiple AI coding agents (Claude Code, Codex CLI, Gemini CLI) simultaneously in isolated git worktrees. Run agents in parallel, each in their own branch, and merge results when ready.

🔗 View Repository

Promption - Manage AI agent configs

A desktop app + CLI for managing AI agent configurations across projects. Stop copy-pasting .cursorrules.

🔗 View Repository

qwen-audio-agent - Full-duplex voice runtime for OpenCode and other ACP agents

Realtime full-duplex voice interface that connects to OpenCode natively over ACP, with one-click install and DashScope configuration. Conversations keep going while OpenCode runs tasks in the background, and task results flow back into the same voice conversation. Also supports Claude Code, Codex, Kimi Code, and more. Ships a macOS desktop app, terminal TUI, and web UI.

🔗 View Repository

Ritemark - Markdown writing app for macOS and Windows where OpenCode works directly in your local documents.

Open-source (MIT) desktop writing app with a visual editor for plain Markdown files in a folder on your computer. OpenCode, with your own API keys, is one of its built-in agents (beside Claude Code and Codex) and edits the documents in place, with a built-in terminal beside the document. Each window is one folder with its own conversations, AGENTS.md, agents and skills. Mermaid and draw.io diagrams, and export to Word and PDF.

🔗 View Repository  •  📖 Docs

Solenta - Desktop app for OpenCode and other coding agents with one shared local memory

A local-first desktop app (macOS, Windows, Linux) that runs OpenCode, Claude Code, Codex, Cursor, Kimi and Grok threads in git worktrees. A local memory server is injected into every session over MCP, so what one thread learns the next one knows; the plan is your GitHub issues, with Git review, terminal and browser panes, and verify-gated completion.

🔗 View Repository

swe-mux - Run OpenCode sessions side by side with a status each, and on your phone over Tailscale

Terminal multiplexer for coding agents. Runs OpenCode (plus Claude Code, Codex and pi) in real terminals on your own machine, shows which session is waiting on you, and serves the same live sessions to your phone over your own Tailscale network. Sessions survive app restarts and updates.

🔗 View Repository  •  📖 Docs

TelegramCode - Telegram bot for OpenCode and Claude Code, run coding agents from your phone

Run your coding agents on your own machine or VPS and drive them from Telegram, by voice or text. Each forum topic is its own project session, like terminal tabs on your phone. Self-hosted, no open ports, MIT.

🔗 View Repository

Untether - Telegram bridge for remote task control

Telegram bridge for OpenCode (and 5 other agents). Send tasks by voice or text, stream progress live, and manage sessions remotely. Supports 75+ providers via OpenCode's model system.

🔗 View Repository

Vibe Kanban - Manage AI in parallel

A Kanban board to manage and orchestrate AI coding agents in parallel.

🔗 View Repository

VSC OpenCode GUI - The opencode agent, native in VS Code - chat in an editor tab instead of a terminal

A VS Code extension that runs the opencode agent in an editor tab with command-palette parity and full theme sync. It spawns a headless `opencode serve` and drives it with its own webview UI over HTTP/SSE - feature parity with the opencode app, without leaving the editor.

🔗 View Repository

Tools & Utilities ₍₈₇₎

Agent Run Inspector - Local, read-only reports for OpenCode session exports - repeated tool calls, failures, completion

Generates a self-contained HTML (and JSON) report from `opencode export <sessionID>` data: repeated identical tool calls, recovered vs unresolved failures, completion state, and tool/usage statistics. Python standard library only, no network calls, and the input file is never modified. Tested with OpenCode 1.18.x; other versions are best-effort with an explicit note in the report. The tool itself is a paid download (USD $49 individual / $149 team); the linked repository contains the documentation, screenshots, and a free sample pack of synthetic inputs and example reports.

🔗 View Repository

Agent Teams AI - Multi-team control plane for OpenCode and other coding agents

Open-source orchestration platform for autonomous coding-agent teams with first-class OpenCode support. Agents manage Kanban tasks, message each other, review work, and coordinate across teams while users steer with high-level commands. Also supports Codex, Claude Code, Cursor, Grok, GitHub Copilot, Kiro, and other providers.

🔗 View Repository  •  📖 Docs

agent-dotfiles - Write AI coding rules once, sync to every agent

Write AI coding rules once, sync to every agent. Supports Command Code, Claude Code, Cursor, Copilot, Codex, and OpenCode.

🔗 View Repository

agent-harness - Reusable agent asset lifecycle for OpenCode workspaces

A Node.js CLI for discovering, curating, staging, activating, and wiring reusable AI-agent assets into developer workspaces, including OpenCode.

🔗 View Repository  •  📦 npm

agent-historian - Search and read past AI coding-agent sessions from the CLI

A read-only CLI (ochist) plus an Agent Skill so OpenCode and other agents can search and re-read their own past conversation history before redoing research. Reads transcripts the agents already persist locally — no memory layer, embeddings, index, daemon, or network. Project- or global-scoped, subagent-aware, zero runtime dependencies.

🔗 View Repository

agent-sessions - Share OpenCode, Claude Code and Codex sessions with your team through git

Push the current OpenCode session into the repo as a redacted transcript plus a handoff summary under .claude/sessions/, and pull teammates' sessions from any coding agent straight into your chat. Ships as an Agent Skills skill that OpenCode loads from ~/.agents/skills, plus a Claude Code plugin and a CLI on PyPI (agent-sessions-cli).

🔗 View Repository

AgentBox - Run multiple coding agents in parallel, each in its own sandboxed VM

Run OpenCode, Claude Code, and Codex in parallel — each teleported into its own sandboxed VM (local Docker, self-hosted, or cloud: Hetzner / Daytona / Vercel / E2B / DigitalOcean). Sub-1s checkpoint startup, per-box browser/VS Code/persistent shells, and git credentials stay on the host. MIT.

🔗 View Repository

AgentField - Control plane for orchestrating AI agent swarms as microservices — best-in-class DX

Open-source infrastructure for orchestrating AI agent swarms at scale. Best-in-class developer experience with SDKs for Python, Go, and TypeScript. Handles async execution, durable state, and observability.

🔗 View Repository

agentledger - Local-first cost ledger — what your coding agents really cost in tokens and money

Reads the session logs OpenCode, Claude Code, Codex and Gemini CLI already leave on disk and turns them into token and dollar totals. Prices cache reads and cache writes at their own rates instead of counting them as input, attributes spend by project/branch/model, detects context churn, and projects a budget. Zero dependencies, no network calls — your entire coding history never leaves the machine.

🔗 View Repository

AgenTopology - Declarative multi-agent topology language

Generate platform configs for OpenCode/OpenClaw and 6 other platforms (Claude Code, Codex, Gemini CLI, Copilot CLI, Kiro) from a single declarative .at file. Define agents, flows, gates, hooks, and group chats. Ships with a Claude Code skill and interactive visualizer. Apache 2.0.

🔗 View Repository

aGiTrack - Turns every agent turn into a traceable git commit, with a live dashboard

A terminal tool that drives OpenCode (or Claude Code or Codex) and automatically commits each agent turn, recording the prompt, the model, and the token cost alongside the diff, so the AI's work leaves a clean and reviewable git history. Ships with a live web dashboard for browsing commits, sessions, and token spend.

🔗 View Repository

Antigravity Local Bridge - Local OpenAI-compatible API bridge for Antigravity

An Antigravity extension that exposes a local OpenAI-compatible HTTP server on localhost:11435. Enables any OpenAI-compatible tool (OpenCode, Aider, Cline, Continue.dev) to use Antigravity's Gemini, Claude, and GPT models without additional API keys.

🔗 View Repository

Argus - Banking compliance SDLC framework for AI agents

Agent-agnostic framework that installs a complete banking compliance SDLC into any codebase. Deploys 5 specialized agents (Guardian, Compliance, Builder, Tester, Reviewer) to enforce GDPR, DORA, MiCA, PSD2, PCI-DSS and other regulations across the entire software lifecycle.

🔗 View Repository  •  📖 Docs

axi-memory (mem-axi) - Filesystem + git agent memory CLI with TOON output

A bash CLI (@masculinecache/mem-axi) giving terminal coding agents durable cross-session memory stored as markdown + YAML frontmatter in a git repository — no daemon, no database, no lock-in. Adds are structured records (decision, howto, failure, constraint, preference) with confidence and priority, retrieved via ripgrep-backed keyword search over frontmatter; output follows the AXI 10 principles for agent ergonomics (TOON, minimal schemas, definitive empty states, structured errors, exit 0/1/2). Composable with OpenCode sessions as a memory bridge — skill entry points can route capture and recall through the same CLI, and dedicated never-write-secrets handling redacts common secret patterns from stored records.

🔗 View Repository

Azdaja - Local context evaluator for OpenCode with bounded semantic calls and validated results.

A Rust binary that keeps source material outside the root prompt, runs exact analysis in a local Python sandbox, and requests bounded semantic judgments through OpenCode when needed.

🔗 View Repository  •  📖 Docs

Beads - Project task management

Steve Yegge's project/task management system for agents (with beads_viewer UI).

🔗 View Repository

Bifrost - OpenCode gateway for multi-provider routing

Route OpenCode requests through Bifrost's OpenAI-compatible gateway with provider selection, virtual-key controls, and request logging.

🔗 View Repository  •  📖 Docs

brood-box - Hardware-isolated microVMs for coding agents

Run AI coding agents (Claude Code, Codex, OpenCode) inside hardware-isolated microVMs with snapshot isolation, egress control, and MCP authorization.

🔗 View Repository

bx - macOS file-system sandbox for AI coding agents

Launch OpenCode, VSCode, Claude Code, or any command in a macOS sandbox (sandbox-exec). Dynamically blocks access to everything except the project directory — no containers, no VMs, just one command.

🔗 View Repository

claudex-switch - OpenCode V1 Go profiles share session history; experimental V2 profiles isolate credentials and history

Source-available CLI for OpenCode V1 Go account aliases and provider-reported quotas. V1 profiles share normal session history; experimental V2 profiles keep credentials and SQLite history per alias. Also supports Claude Code and Codex. Licensed under MIT + Commons Clause v1.0: ordinary work use is allowed, but charging third parties for products or services whose value derives entirely or substantially from this tool requires separate permission.

🔗 View Repository

Claudexor - Run OpenCode alongside Claude Code, Codex, and Cursor in one local-first control plane

A local-first control plane that runs OpenCode alongside Claude Code, Codex, and Cursor while keeping one Claudexor thread across harness lanes. Adds isolated workspaces, Best-of-N runs, cross-model review, and typed apply gates; quota-aware multi-subscription rotation is available for Claude Code and Codex.

🔗 View Repository  •  📖 Docs

CLI Proxy API - Multi-model proxy

A proxy server providing compatible API interfaces for multiple model CLIs.

🔗 View Repository

cliproxy-rs - Use your Claude and ChatGPT accounts in OpenCode through one local endpoint, with each account's limits on a dashboard

A Rust port of CLIProxyAPI. OpenCode points at it as an OpenAI-compatible custom provider, and it translates to the Anthropic, OpenAI and Gemini APIs while spreading requests across the accounts and API keys you already have. The built-in dashboard shows each Claude and Codex account's 5-hour and weekly limits. Reads CLIProxyAPI's config and auth files. MIT.

🔗 View Repository

Codex Proxy Server - Local API proxy

A proxy server that provides a local API proxy for Codex/ChatGPT-like models.

🔗 View Repository

Cognitive Lead HQ - OpenCode multi-agent HQ with Brain/Hands split, FastMCP and 31 skills

Documentation-only OpenCode HQ with Brain/Hands split, 3 FastMCP servers (context/memory/lint), Kanban ZAC workflow and 31 reusable skills.

🔗 View Repository  •  📖 Docs

cowork-to-code-bridge - Run agent tasks on your own machine from a sandboxed caller, with no inbound port

A local-first async bridge that lets an agent running somewhere else — a cloud sandbox, a container, a hosted chat session, a CI job — execute work on the developer's real machine. There is no network listener, no inbound port and no tunnel: a daemon watches a shared directory, the remote side drops a JSON task file in, and the daemon runs it locally and writes a result file back. Any shared folder works (bind mount, Syncthing, Dropbox, NFS). Execution is allow-listed — only scripts installed into the daemon's scripts directory can run, so a remote caller cannot request arbitrary shell. Tasks are non-blocking and cancellable (SIGTERM to the whole process group, then SIGKILL after a grace period), output is size-capped while streaming with truncation reported, and results carry real exit codes. Ships ~25 scripts (host health, disk, ports, docker ps/logs, git status, package updates), a wrapper that hands a whole task to a local CLI coding agent, and an MCP proxy that relays JSON-RPC to local stdio MCP servers. Python client, stdlib-only, MIT, macOS and Linux.

🔗 View Repository

Cupcake - Policy enforcement layer

A native policy-layer for AI coding agents built on OPA/Rego with native OpenCode plugin support.

🔗 View Repository

DevIntern - Turn tracker tickets into OpenCode-authored, self-reviewed pull requests

Wires the task tracker your team already uses (Jira, Linear, Trello, Asana, Azure DevOps, GitHub Issues, or plain markdown files) to OpenCode. Running `devintern PROJ-123 --create-pr` picks up the ticket, runs OpenCode non-interactively in your repo, self-reviews the diff, and opens a pull request. A feasibility gate bounces vague tickets back to the tracker with questions instead of producing a confidently wrong PR. An optional unattended mode fully automates ticket handling: scheduled ticket pickup plus webhook-driven handling of PR review comments, so reviewer feedback becomes commits on the same branch. Includes a companion `devpm` tool that turns rough prompts, logs, or Figma frames into well-specified, codebase-grounded tickets. Runs on your machines with your own model keys.

🔗 View Repository  •  📖 Docs

djimitflo - Djimit Agentic OS control plane with verified OpenCode executor integration

A TypeScript monorepo control plane for goal-driven autonomous loops, governance gates, and multi-runtime execution. Includes a verified OpenCode executor runtime, loop lifecycle management, maker/checker/security leases, and a dashboard for tracking goals and loop runs.

🔗 View Repository

dsh-movein - Preview and move OpenCode setup into DeepSeek Harness

Preview and migrate global and project OpenCode configuration into DeepSeek Harness. Handles JSONC precedence, singular and plural asset directories, inline agents and commands, local and remote MCP servers, disabled entries, and environment placeholders without copying secret values.

🔗 View Repository

eval-harness - Behavior-regression testing for OpenCode skills

Detects when an OpenCode skill's behavior drifts from a baseline, attributes the cause across 4 classes (skill change / fixture stale / model change / unknown drift), and emits a 6-field FAIL schema with transcript-span and env-delta evidence. Ships a git pre-push hook, a composite GitHub Action, and a $-cost hard ceiling for CI safety. Bash + jq + python3 stdlib only; no daemon, no SaaS. MIT.

🔗 View Repository

Foreman - Jev-powered agent supervisor for OpenCode workers

Runs OpenCode workers through opencode run while TypeSafe AI's Jev independently assesses requirements, implementation completion, tests, progress, and human escalation. Foreman can continue, stop, retry, or launch independent verification workers. The OpenCode backend supports stop/retry interventions.

🔗 View Repository  •  📖 Docs

Forgeo - Scheduled software factory that runs your coding agent on a backlog

A scheduled, agent-driven software factory for a repository. Keep a plain JSON backlog of tasks and Forgeo runs your configured coding agent (opencode, Claude Code, Codex, or any CLI) on each one, commits the result to the default branch, and runs a refactoring pass when the backlog is empty. Pauses and writes BLOCKER.md when the agent needs a human decision, with a web dashboard for every instance.

🔗 View Repository  •  📖 Docs

fractal - Hierarchical coding-agent loops with recursive delegation and per-node worktrees

Supports OpenCode as one of five coding-agent backends. Each node runs in its own Git worktree, delegates separable subtasks recursively, and records state and cost data in a local SQLite database with live terminal controls.

🔗 View Repository

Gemini CLI to API - Gemini proxy

A proxy that converts the Gemini CLI tool into OpenAI-compatible endpoints.

🔗 View Repository

Geneseed - Compiles a disciplined agent harness for OpenCode - laws enforced by plugins at the tool boundary

One source renders into a ready-made OpenCode harness - a constitution of laws and doctrines, 18 capability agents, 47 native skills, a memory convention and 7 plugins that hard-block force-pushes, reset --hard and credentials written into tracked files before the tool runs. Install once globally and every repo inherits it; 14 themes change only the voice. Zero runtime dependencies. Also emits for Claude Code, Bob, OpenClaude and plain AGENT.md tools.

🔗 View Repository  •  📦 npm

GolemBot - Unified AI assistant framework for multiple Coding Agent CLIs

Wraps Claude Code, Cursor, OpenCode, and Codex behind a single API with a Skill system, IM channel adapters (Feishu/Slack/Telegram/Discord), fleet mode, and an interactive onboard wizard.

🔗 View Repository

hcom V2 - Let AI agents message, watch, and spawn each other across terminals

Claude Code, Gemini CLI, Codex CLI, and OpenCode agents in separate terminals can message each other, detect file edit collisions, read transcripts, view terminal screens, subscribe to activity, and spawn/fork/resume agents. First-class OpenCode support with native plugin. Includes TUI dashboard, cross-device relay, Python API, and multi-agent workflow scripts. pip installable, MIT licensed.

🔗 View Repository

herdr-reviewr - Review your coding agent's diff in a terminal pane and send line comments back to OpenCode

A terminal code-review pane that runs beside OpenCode in herdr. Read the agent's diff, comment on lines or ranges, and send the comments straight back to the agent's input. Also works with Claude Code, Codex and Pi.

🔗 View Repository

Hexis - Git-backed skills, tools, and context for OpenCode

Manages shared agent skills, tool access, and context in Git, then makes approved resources available to OpenCode through a remote OAuth MCP connection.

🔗 View Repository

Hivemind - Delegate grunt work to headless opencode worker swarms while your main agent plans and reviews

Runs opencode as a disposable worker fleet driven by a planning agent. Every worker returns exactly one compact JSON line (result, tokens, cost, duration), so raw agent streams never flood the orchestrator's context. Writing workers each get their own git worktree so parallel edits cannot corrupt each other, and an explicit fallback ladder retries once, then falls back to the orchestrator doing the subtask itself, so a swarm never fails a task a single agent could have finished. Ships scout, coder and tester agent definitions plus slash commands for parallel review panels, research sweeps, batched migrations and partitioned test runs.

🔗 View Repository

jailoc - Sandboxed Docker environments for OpenCode agents

CLI tool that wraps OpenCode agents in isolated Docker Compose environments with file isolation (bind-mounted workspaces only), network isolation (private subnets blocked by default with host allowlisting), and a per-workspace DinD sidecar. Zero config to start.

🔗 View Repository

llm-panel - Put one question or diff to several LLM CLIs in parallel — opencode runs as a verified read-only agent

Put one question or diff to several LLM CLIs in parallel — opencode judges run as a verified read-only agent alongside codex, claude and ollama — with every answer shown in full, an anonymised rebuttal round, cost accounting, a GitHub Action, and a measured miss rate on AACR-Bench.

🔗 View Repository  •  📖 Docs

nsq - Run OpenCode, Claude Code and Codex side by side in one terminal and see which one needs you

Open-source CLI that runs OpenCode (1.x and 2.x), Claude Code, Codex or any command side by side in one terminal. Status comes from each CLI's own events (working / needs you, with the question / finished), and a permission prompt can be answered y/a/n without opening the agent. A daemon keeps agents alive when the terminal closes, each agent can get its own git worktree, and cost is tracked per agent. Never writes to opencode.json; each session gets its own config layer. OpenRouter and local model servers supported. No account, no telemetry. MIT, `npx neurosquad`.

🔗 View Repository

OC Context (occtx) - Switch contexts quickly

A command-line tool for switching between different opencode contexts quickly.

🔗 View Repository

OC Monitor Share - CLI monitoring tool

A CLI tool for monitoring and analysing opencode AI coding usage.

🔗 View Repository

ocm - Plugin marketplaces for OpenCode - install skills, agents, commands and MCP servers from any git repo, globally

Turn any git repository into a plugin marketplace. `ocm add <url>` installs its skills, agents, commands, JS plugins and MCP servers once into your global OpenCode config, so they work in every project, and `ocm update` plus a startup auto-sync keep them current. Includes a /ocm TUI for browsing, asks before any plugin code runs, and never overwrites your own files.

🔗 View Repository

OCP - Isolated OpenCode profiles

Juggling work and personal OpenCode accounts means hand-swapping env vars and clobbering your sessions. ocp gives each project its own auth, session history, and config — and switches automatically based on the directory you're in. Think direnv for OpenCode.

🔗 View Repository

ocp - Manage OpenCode config profiles in Git

A CLI for switching OpenCode configuration profiles stored in Git repositories.

🔗 View Repository

OCX V2 - OpenCode package manager

The missing package manager for OpenCode extensions - ShadCN model with Ghost Mode.

🔗 View Repository

oos - Find any OpenCode session across all projects and resume it instantly

A TUI fuzzy finder over every OpenCode session from every project — type keywords to filter in real time, pick with ↑/↓, Enter to jump back into the conversation. Go, zero config, single binary.

🔗 View Repository

Opencode A2A - Full A2A Protocol support for OpenCode.

A production-ready A2A Protocol implementation that exposes OpenCode as a standardized Agent-to-Agent service, enabling seamless cross-agent collaboration and automated workflows.

🔗 View Repository

Opencode Actions - Reusable GitHub Actions for installing and running OpenCode in CI/CD workflows

A collection of reusable GitHub Actions that make it easy to integrate OpenCode into your CI/CD pipelines. Includes actions for PR review, architecture review, spec coverage checking, and one-step opencode setup and execution. Supports caching, retry logic, and multiple LLM providers.

🔗 View Repository

OpenCode Database Cleaner - Safely analyze and reclaim unused space in OpenCode's SQLite database

A zero-dependency Python CLI that reports live data and freelist bloat in OpenCode's local SQLite database, then safely reclaims unused disk space with process checks, WAL checkpointing, verified backups, VACUUM, and integrity checks. It does not delete sessions.

🔗 View Repository

Opencode DDEV V1 only - DDEV container wrapper

Wraps bash commands to execute inside the DDEV container (Docker-based PHP development environments).

🔗 View Repository

Opencode Helper - Cleanup, analysis & visualization CLI for OpenCode

Zero-dependency Python CLI that cleans old sessions and temp files, analyzes model/tool/MCP/skill usage patterns with AI interpretation, and generates visual workflow diagrams. pip install opencode-helper.

🔗 View Repository

Opencode History Search V1 only - Search OpenCode conversations with keyword, regex, and fuzzy search

Search through your OpenCode conversation history with powerful keyword, regex, and fuzzy search capabilities. Features date filtering, multiple match types (titles, messages, tools, file paths), and fast performance (<50ms queries).

🔗 View Repository

Opencode Profile Router - Profile router for auth and selective config isolation

Shell wrapper for OpenCode that selects named profiles by path and can isolate provider login state, config directories, and data roots selectively and independently per profile.

🔗 View Repository

Opencode Session Manager - Session viewer & manager

View & Manage sessions for opencode, also detect orphan sessions for deletion

🔗 View Repository

Opencode Skills V1 only - Skills management

Skills management system for organising and tracking opencode capabilities.

🔗 View Repository

OpenCode Unlimited Proxy - Self-hosted OpenAI-compatible gateway for the OpenCode free tier, with a de-duplicated egress pool

Serves OpenCode's anonymous free tier (`Bearer public`) through an OpenAI-compatible API. Documents and implements the four server-side conditions those models gate on, so callers outside the CLI don't hit `403 FreeTierError`: a tool named exactly `read` (renaming it does not count), forced `stream: true`, a versioned `User-Agent`, and a canonically-encoded session id. Each one is independently probeable with `scripts/gate-probe.py`, which ships negative controls so a pass cannot be confused with a broken harness. In front of the gateway sits a de-duplicated pool of rotating egress addresses, so six nodes are six quota buckets rather than three.

🔗 View Repository

opencode-agent for Cowork - Cowork plugin that turns Claude into an intelligent orchestrator for OpenCode sessions

A Cowork (Claude desktop) plugin that bridges natural language conversations with OpenCode. Handles agent routing (build/plan/@general), stall detection, result validation (diff coherence, tests, build, semantic alignment), per-project operational memory, provider fallback chain (anthropic → openrouter → openai), and MCP composability with existing user tools. Ships 8 orchestration skills, 3 scheduled-task recipes, and 1 custom agent template (cowork-with-github). Install: drag-and-drop .plugin into Cowork.

🔗 View Repository

opencode-handoff-p2p - Transfer OpenCode session share URLs P2P via private GitHub inbox repos

An OpenCode skill that transfers session share URLs between machines or collaborators peer-to-peer using personal GitHub private inbox repositories. Features a 5-tier verification pipeline (filename regex, trusted-sender allowlist, byte-level content validation, GitHub commit author+committer attribution, GPG signature), transactional fetch-before-delete (URL fetch failure preserves the inbox file for retry), and a Chinese-language trust-boundary preamble that prevents fetched share content from autonomously triggering local actions. Each user owns one private inbox repo; senders push directly into it via collaborator access. Includes a reference Python verification script for deterministic cross-shell behavior.

🔗 View Repository

OpenKrak - Pre-computes your repo structure via Dorchester engine before the LLM sees a single token

OpenKrak is an MCP server that runs the Dorchester static analysis engine on your repository and delivers a structured Mahadata brief to your coding agent before it reads any files. Cuts token usage 20-80% by replacing raw file reads with a pre-computed dependency graph, hotspot ranking, and blast radius map. Works with OpenCode. Free tier includes 15 queries per 24h rolling window. Install via npm install -g openkrak-mcp.

🔗 View Repository

OpenSpec - Spec-driven development

Spec-driven development with opencode - structured specification management.

🔗 View Repository

openwts - Isolated git worktrees for AI coding agents — opencode, Claude Code, and more

Stop context-switching. openwts creates isolated git worktrees for each task and opens your AI coding agent inside them — no stashing, no partial commits, no "what branch was I on?"

One command — openwts opencode hotfix — creates a worktree, opens opencode inside it, and cleans up when you exit. Jump between tasks like it's nothing.

Features

• Agent-agnostic: opencode, Claude Code, and any AI CLI on PATH
• Auto-detects opencode, falls back to an interactive picker
• Works on all platforms: Linux, macOS, Windows
• Manifest tracking — never touches worktrees it didn't create
• Cleanup modes: auto-remove, prompt, or leave

Zero config. One env var if you want a default agent.

🔗 View Repository

OrcaReplay - Record an OpenCode run, replay it offline with the network off, fork it onto another model

Records an OpenCode session without modifying it, then replays that run with the network off and reproduces it with the network off, or forks it from any step so the model is the only thing that changed. Because OpenCode keeps its provider origin in opencode.json rather than an environment variable, there is nothing to redirect: the adapter terminates the TLS OpenCode opened itself using a per-run CA that only the launched process trusts and that is deleted when the run ends, and your config is left untouched. Alongside the model traffic it records shell exit codes and timing, per-turn file changes and MCP calls, all on one timeline. Apache-2.0, Node 20+, and the trace format is a published spec so other readers can be written against it.

🔗 View Repository

Qwen Code OAI Proxy - Qwen model proxy

An OpenAI-Compatible Proxy Server for Qwen models.

🔗 View Repository

Reflection-3 V1 only - Judge layer that catches premature agent stops

LLM-as-judge plugin for OpenCode and Claude Code: fires on every agent stop, classifies stops as premature or legitimate using a rubric mined from 227 real sessions (78% were premature), and re-prompts the agent with targeted feedback. Enforces workflow gates (tests, CI, PRs). Implements the Reflexion self-improvement loop. Works on both OpenCode (session.idle event) and Claude Code (Stop hook).

🔗 View Repository

Roborev - Background code review agent with OpenCode agent-hook support

A multi-agent background commit-review tool that runs automated code reviews and exposes a $roborev-fix skill via OpenCode's agent hook. Useful for catching issues before they land while keeping the review loop inside the OpenCode workflow.

🔗 View Repository

rolecraft - The package manager for AI agents — install, test, publish, and manage skills & MCP servers across 86+ agents

Zero-dependency CLI for installing AI agent skills as roles & behaviors from any source (local folders, GitHub, GitLab, SSH). Supports 86 agent targets (26 verified), lockfile-based CI re-install with `rolecraft ci`, `rolecraft doctor` for troubleshooting, and bundled team onboarding. Works with opencode, claude-code, cursor, and more.

🔗 View Repository

Sabi - Adaptive per-round inference routing for coding agents — model, effort and provider scheduled against the trajectory

Sits between a coding harness and its model providers and makes a fresh routing decision per inference round, not per session. Each round is classified from trajectory state — round position, tool calls and their results, failure evidence, context pressure — and a deterministic policy maps class to tier (failure → strong, exploration → cheap, first-turn → mid). An optional semantic judge (TypeSafe Jev) runs only on rounds the policy cannot classify, and is fail-open. The harness loop is untouched. Exposes an OpenAI-compatible local proxy at 127.0.0.1:8787/v1. From OpenCode, `npm run connect:opencode` writes only a provider.sabi entry and the adaptive alias is `sabi/sabi-code`; fixed baselines `sabi/sabi-cheap`, `sabi/sabi-mid` and `sabi/sabi-strong` are available for comparison. Borrowed-auth passthrough is supported: the harness keeps its own credential and Sabi rewrites only model. Decisions and cost are inspectable via `npm run report`.

🔗 View Repository  •  📖 Docs

Session Coherence - Cross-tool session memory for AI coding assistants

Cross-tool session memory for AI coding assistants. One markdown chronicle shared by 9 tools (Claude Code, Cursor, Codex, Gemini CLI, Aider, and more). Zero dependencies, ~300 token fixed cost, 100% local.

🔗 View Repository

SpecRalph - Spec-Driven Ralph-Wiggum AI coding automation CLI

AI Loops runner for OpenCode and other agent CLIs. Enables spec-driven development workflows with structured prompt engineering and loop automation for AI coding agents.

🔗 View Repository

squirrelscan MCP - Website audit CLI, MCP server and skill for coding agents (SEO, performance, security, accessibility)

CLI website audit tool built for coding agents. Runs 260+ rules across SEO, performance, security, accessibility and agent experience, and hands the agent exact, source-mapped fixes. Works with OpenCode via MCP (hosted https://mcp.squirrelscan.com/mcp with OAuth, or local `squirrel mcp` over stdio) to run audits, read reports and fix findings, or via the agent-skills format with `npx skills add squirrelscan/squirrelscan`. Free local audits, GitHub Action, MIT CLI and rules.

🔗 View Repository  •  📖 Docs

staffetta - Model-decided context handoff: close a milestone, write memory to disk, restart with a clean session

A relay-race alternative to compaction for long-running agents. Instead of a threshold rewriting the conversation in place, the agent itself decides when a milestone is closed, writes what the next session needs to disk, and hands over — the session ends and a fresh one picks up the baton. Memory lives in files (MEMORY.md overwritten, JOURNAL.md append-only, AGENTS.md for rules), so it is readable, diffable, correctable and survives the session. Four tools the agent calls itself: checkpoint (verifies by checksum that memory actually changed, rejects journal entries with no numbers in them, detects tampering with older entries), ask (pauses for a human decision, requires the agent to state what it would do without an answer), resume (refuses until the decision is written down), and loop (reopens the session and nothing else — it ends when the model declares --done or --blocked, not when a counter infers it). Measured on a local KAT-Coder V2.5 via llama.cpp: 7 milestones closed autonomously, 7 of 7 checkpoints invoked unprompted, 587 tests green, ~5,500 tokens of context per round instead of a compacted session restarting from 142,000 — and it does not grow across rounds. Pure bash, no dependencies; adapters/opencode/loop.sh is the opencode adapter and everything under core/ works with any agent that can run a shell command.

🔗 View Repository

supamem MCP - Project-agnostic dual-memory MCP CLI for OpenCode

Adds semantic memory (Qdrant tuned hybrid retrieval — RRF fusion of dense MiniLM + sparse BM25) and structural memory hooks to OpenCode via MCP. Ships an installer that wires `qdrant-find` / `qdrant-store` tools and a snapshot hook into your project's OpenCode config. Python 3.12+, MIT, on PyPI as `supamem`.

🔗 View Repository

SwarmClaw - Self-hosted multi-agent runtime with first-class OpenCode delegation

Self-hosted runtime for autonomous AI agents with heartbeats, schedules, delegation, memory, runtime skills, and reviewed conversation-to-skill learning. Orchestrates OpenCode alongside Claude Code, Codex, Gemini CLI, Copilot CLI, Cursor Agent, Goose, Qwen Code, Droid, and 20+ LLM providers. Ships as an Electron desktop app, CLI, and Docker image. MCP-native (server and client). MIT, TypeScript.

🔗 View Repository

SwarmVault MCP - Local-first RAG knowledge base compiler with bundled skill and MCP server

Turns raw docs, research, and code into a persistent markdown wiki, knowledge graph, and hybrid SQLite FTS + embeddings search. Ships a bundled skill (schema-first, graph-query-first conventions) and an MCP server, so it works with OpenCode, Claude Code, and Codex. Offline heuristic provider by default; optional Ollama for local embeddings.

🔗 View Repository

Tokscale - Token usage tracking CLI

A CLI tool for tracking token usage from OpenCode and other coding agents (Claude Code, Codex, Gemini CLI, and Cursor IDE).

🔗 View Repository

Tree Ring Memory - Local-first memory lifecycle CLI for AI agents

Rust-native memory lifecycle layer for AI agents with project-scoped SQLite/FTS recall, forgetting, audit, consolidation, DOX/Revolve sync adapters, terminal TUI, and agent-framework discovery for OpenCode-style workflows.

🔗 View Repository  •  📖 Docs

trustless - Credential broker CLI — injects secrets into subprocesses and proxies so AI agents never see plaintext keys

trustless decouples AI agents from credentials. Agents reference secrets by name; the broker resolves them at the process or transport layer with output sanitization, integrated DLP (outbound redaction), OAuth token refresh, structured audit logs, and pass/Bitwarden backends. Zero-dependency single binary with bundled agent skills for OpenCode, Claude Code, Codex, and Hermes.

🔗 View Repository

Universal Agent Plugins - Manage Agent Plugins across OpenCode and other AI coding agents

A cross-platform CLI for installing, updating, repairing, and removing Agent Plugins 1.0 across OpenCode, Codex, Claude Code, Cursor, Gemini CLI, and other compatible agents. It also includes tools and an SDK for building portable plugins.

🔗 View Repository

Universal LLM API Proxy - Universal multi-model proxy and library - made with Opencode community

Universal LLM Gateway: One API, every LLM. OpenAI/Anthropic-compatible endpoints with multi-provider translation and intelligent load-balancing. Works with any application that supports custom OpenAI/Anthropic base URLs—no code changes required in your existing tools. Best support for Antigravity/Gemini CLI out of the competition. Deploy anywhere.

🔗 View Repository

Unship - Compare AI-made UI variants locally

Local CLI and browser picker for comparing temporary UI variants created by coding agents like OpenCode, then keeping one and cleaning up the rest.

🔗 View Repository

Vestige - Local persistent memory for OpenCode agents

Adds local and project-scoped MCP memory to OpenCode, so agents can remember decisions, preferences, architecture context, and previous fixes across sessions.

🔗 View Repository  •  📖 Docs

yoink - Connect an API-key provider to opencode and 12 other coding harnesses, and switch AI coding logins

CLI and macOS menu bar app. Add a provider once (OpenAI, Kimi Code, Moonshot, OpenRouter, DeepSeek, z.ai, Ollama or any OpenAI- or Anthropic-compatible base URL), pick its models, and tick opencode; yoink writes the provider into ~/.config/opencode/opencode.json (or opencode.jsonc, edited in place), alongside pi, codex, Claude Code, Qwen Code, Kilo Code, Droid and others. It also saves and switches Claude Code, ChatGPT (Codex), Kimi Code, Gemini and GitHub Copilot logins.

🔗 View Repository

Forks ₍₃₎

OpenCode Vim V2 - OpenCode fork with built-in Vim mode

An unofficial OpenCode fork that adds Vim motions, navigation, and editing directly to the OpenCode TUI.

🔗 View Repository  •  📖 Docs

opencode-rust - The open source AI coding agent, in Rust - CLI, TUI and HTTP server

An independent Rust implementation of the TypeScript opencode, providing a CLI, a TUI and an HTTP server for local AI agents, with no Docker required. MIT licensed, 21 crates and 571 tests. The command is `opr`; install with `npm install -g opencode-rust` or `brew install vakandi/tap/opencode-rust`. Not affiliated with the OpenCode team. Docs: https://vakandi.github.io/opencode-rust/

🔗 View Repository

RedCode - TUI + GUI dual-interface AI coding assistant with MCP ecosystem

A feature-rich opencode fork with dual-interface (TUI terminal + GUI desktop), built-in MCP ecosystem (CodeGraph, TypeGraph, jCodeMunch, Browser, Vision, Web Search), automated memory system, AI personality system, skill extensions, and 3-tier permission guardrails. Privacy-first design with personal data fully separated from public templates.

🔗 View Repository

➕ Add a Project

📚 RESOURCES (28)

Skills ₍₇₎

Claude Skills Pro - 15 battle-tested engineering skills (review, testing, debugging, security) that install into OpenCode in one command.

A pack of 15 expert-workflow skills - seven-dimension code review, root-cause debugging, tests that catch real bugs, feature specs, security audit, safe refactoring, API design, zero-downtime migrations and more. Each skill encodes trigger conditions, a step-by-step procedure, and hard anti-patterns, so the agent follows senior-engineer discipline automatically. 7 skills are free (MIT). Install into OpenCode via npx skills add Hahaknight/claude-skills-pro --agent opencode.

🔗 View Repository

DataEcho Skills - Deploy files, sites, and Dockerfile apps to live URLs + persistent agent memory, from opencode.

Two zero-dependency Agent Skills for the DataEcho platform (dataecho.ai). `dataecho` publishes anything the agent builds — a single file, a static site, or a server-side app via Dockerfile — to a live URL in seconds, with anonymous publishing (24h claim flow), incremental deploys, and private cloud Drives with scoped share tokens. `dataecho-memory` gives the agent durable memory across sessions and machines — an indexed fact store on a private versioned drive with atomic concurrent-safe writes, history/restore, and one-command handoff. Install with `npx skills add mohocp/dataecho`; opencode discovers them via its standard skill paths. Also available as an MCP server (https://dataecho.ai/mcp).

🔗 View Repository

Jev Social - Browser-grounded social research with Jev and the local socai CLI

An OpenCode-compatible Agent Skill that routes an Instagram, TikTok, or LinkedIn research goal through Jev, runs read-only searches with the user's local socai CLI and existing browser session, and returns source-linked evidence plus a Markdown report. Requires Node 20+, a current socai CLI, a browser session, and either OpenRouter Jev or a local System One-compatible endpoint.

🔗 View Repository

Juhe Skills - Juhe MCP 深度调研与快速查询 skill 集

提供两个 AI agent skill：juhe-deep-researcher（四档可选深度调研分析，支持轻量/标准/深度/旗舰）和 juhe-quick-researcher（快速全网资讯搜索）。配合 Juhe MCP 远程服务使用，支持 RSS 订阅源检索、文章全文获取、AI 摘要、趋势分析等功能。适用于 OpenCode、Claude Code、Cursor 等支持技能和 MCP 协议的编码 agent。

🔗 View Repository

Meanwhile - Optional human side quests while OpenCode continues long work

A portable Agent Skill that offers one bounded, opt-in learning, recovery, play, or reality-check quest while the coding agent continues. It requires no account, callback, or completion tracking.

🔗 View Repository

repo-cover - Agent skill that designs your GitHub social preview as one self-contained HTML file

The agent writes a 1280x640 og:image card following a constrained design system, with four editorial moods, CJK-first typography, a deterministic checker for contrast and line breaking, and PNG export through headless Chrome. Installs into OpenCode via the Agent Skills CLI.

🔗 View Repository

SuperSpec - Requirement or production alert in, reviewed pull request out - 20 skills, one SKILL.md source

A toolkit of 20 Agent Skills that combines the engineering discipline of superpowers with the living spec lifecycle of OpenSpec, and adds the two things neither has: end-to-end workflows that carry a requirement, or a production alert, all the way to a reviewed pull request, and multi-repo orchestration that runs one headless agent per repository. Implementation is dispatched to parallel test-driven subagents; a review panel with distinct lenses (quality, spec and guardrail compliance, cross-module integration) then returns a severity-graded verdict and drives a bounded fix loop until the code is approved. Root-cause analysis is evidence-based rather than a guess: competing hypotheses are falsified against logs, traces, metrics, code and git history before any fix is written. Capability specs follow the OpenSpec directory convention, so a repository initialized by the OpenSpec CLI works as-is. Nothing is written into your project - the guardrails are read in place at runtime, and there is no CLI or state machine to install.

🔗 View Repository

MCP Servers ₍₇₎

Baizhi Cloud Agent Toolkit - Remote web search and page extraction MCP for OpenCode agents

For OpenCode V2, configure a remote MCP server at https://agent-toolkit.app.baizhi.cloud/mcp with oauth disabled and Authorization: Bearer {env:BAIZHI_API_KEY}. It offers web search, page reading, and structured extraction. The linked MIT repository publishes integration configuration and documentation, not the hosted backend. Users need their own Baizhi account and API key; calls may consume paid credits.

🔗 View Repository

Find MCP - Search 17,000+ MCP servers from the official MCP registry

Find MCP is an MCP server for discovering other MCP servers. It searches the agentage MCP Catalog, which syncs 17,000+ servers from the official MCP registry (registry.modelcontextprotocol.io), through the mcp_search, mcp_get, and mcp_categories tools. Works with OpenCode as a generic MCP server, either remote over Streamable HTTP (https://catalog.agentage.io/mcp, no auth needed for search) or stdio via npx -y @agentage/find-mcp.

🔗 View Repository

LWC - Source-grounded project memory for OpenCode with an Agent Skill, CLI, and read-only MCP server

Local-first project memory for coding agents with bounded recall, source citations, provenance, atomic changesets, and optional document and code graphs. Includes the installable using-lwc Agent Skill and a stdio MCP server exposing one bounded, read-only exploration tool.

🔗 View Repository

OutageDeck MCP - Vendor-published cloud and SaaS outage checks for OpenCode agents

OutageDeck gives OpenCode agents current provider status, incident timelines, and observed uptime history for cloud and SaaS vendors, based on their official status feeds. Its hosted remote MCP server connects without a package, daemon, account, or API key for public status checks, so an agent can rule out dependency outages before changing code.

🔗 View Repository  •  📖 Docs

Pixeltable Developer MCP - Local MCP server for Pixeltable applications

Local MCP server for Pixeltable, a multimodal data store with incremental computed columns and embedding search for agents.

🔗 View Repository

RunAPI MCP - MCP server for model API jobs

RunAPI MCP lets coding agents discover models, check pricing, and create image, video, music/audio, text-to-speech, and LLM tasks through RunAPI.

🔗 View Repository

Tapetide MCP - Indian stock market data (NSE and BSE) for OpenCode agents

MCP server that gives OpenCode structured Indian stock market data for about 8,200 NSE and BSE stocks, including quotes, quarterly financials, shareholding, a 326-ratio screener, FII/DII flows, option chains and company filings. Runs locally over stdio via npx with a free personal token, or remotely at https://mcp.tapetide.com/mcp with OAuth. MIT licensed, free plan available.

🔗 View Repository  •  📖 Docs

Tools & Utilities ₍₂₎

Akephalos - Local-first markdown passport for portable agent context and memory

A local-first, markdown-first `.akephalos` passport for portable agent preferences, tool notes, rules, project context, and durable memories. It uses plain files and Git so MCP-capable coding-agent workflows can inspect and carry context across tools and machines without hosted sync.

🔗 View Repository

GoTTY - Turn CLI into Web App

A simple command-line tool that turns your CLI tools, like opencode, into web applications.

🔗 View Repository

Configs & Starters ₍₈₎

AI Agent Automations - Automation prompts for OpenCode, Codex, Claude Code, and Cursor workflows

A curated collection of agent automation prompts for real-world engineering and operations workflows. Covers OpenCode, Codex, Claude Code, and Cursor patterns for code maintenance, security reviews, research, support, billing, and work triage.

🔗 View Repository  •  📖 Docs

Cosmosmith - Multi-agent workflow initializer with OpenCode adapters

Generates AGENTS.md, task.md, proposal/design templates, and adapters for OpenCode, Claude Code, Cursor, Copilot, and Trae so mixed-agent projects can share one governance layer and verification ledger.

🔗 View Repository

Infer Client Examples - OpenCode configurations for Infer's GPT-6 Astra and GPT-5.6 Sol models

Configures Infer as a custom Responses API provider in OpenCode using an environment-based key. Includes Astra and Sol selectors, a standard-library Python client with an output-token cap, and offline tests. Live usage requires paid Infer wallet credit.

🔗 View Repository  •  📖 Docs

kickstart.opencode - A heavily commented OpenCode starter config that teaches you what everything does.

Inspired by kickstart.nvim. A minimal, well-commented OpenCode configuration designed as a starting point, not a framework. Every decision is explained. Includes free-only models and MCPs, a plan/execute workflow with interview-style clarification, and a careful agent for supervised AI execution. Read it, understand it, make it yours.

🔗 View Repository

looot MCP config - opencode.json snippet for the looot remote MCP server, 2,500+ data API endpoints

A ready opencode.json block that adds the looot remote MCP server (https://api.looot.ai/mcp) with browser sign-in, plus a variant that reads a token from {env:LOOOT_TOKEN}. looot gives an agent one key and one prepaid balance for work-email lookup, enrichment, SERP, backlinks and web page data. The same repo has snippets for other agent platforms and a starter SKILL.md that opencode loads from .agents/skills.

🔗 View Repository

my-opencode-deepseek-config - Optimal DeepSeek V4 dual-model config — 11 agents, 15 skills, 25 commands

A pure-config, zero-dependency OpenCode configuration optimized for DeepSeek V4 dual-model (Pro + Flash). Includes an orchestrator with 17 intent-gate patterns, delegation decision table, and cost-aware routing; 11 specialized subagents (planner, deep-worker, oracle, reviewer, consultant, ui-builder, explore, librarian, light-orchestrator, generalist); 15 on-demand skills (code-review, spec-workflow, deepwork, reflect, and more); and 25 slash commands. Designed for maximum token efficiency with model-isolated routing, background dispatch, read-only isolation for analysis agents, and execution-only discipline for worker agents. Iterated across 14 versions, continuously incorporating best practices from 6 upstream reference projects.

🔗 View Repository

Opencode Config Starter - Flexible config starting point

A powerful custom opencode configuration with agents, commands, rules, skills, and pre-configured MCP server.

🔗 View Repository

OpenCode Ultra Setup - A full OpenCode setup with a team of 11 agents, persistent memory, context compression, and a real terminal.

An "ultra" OpenCode setup combining oh-my-openagent (11 agents with team mode), supermemory (persistent memory), headroom (context compression up to ~57%), opencode-pty (interactive terminal), cited web search, and wakatime metrics. Includes Cortex and GitHub MCPs, an OmniRoute free AI gateway, and Karpathy + anti-slop guidelines in a global AGENTS.md.

🔗 View Repository

Guides ₍₄₎

Coding Agent Orchestration - OpenCode Setup Guide / Orchestration

An opinionated but practical handbook that argues for treating AI coding tools as a composable, multi-agent system rather than a single assistant, and then walks through how to implement that mindset using OpenCode's CLI, configuration system, and ecosystem (agents, tools, skills, and plugins).

🔗 View Repository

Debug Log to Text File - Troubleshooting guide

How to output a debug log from opencode to a text file for troubleshooting.

🔗 View Discussion

OpenCode CN Deployment Guide - Chinese free-tier multi-model deployment guide (DeepSeek/Kimi/Zhipu/SiliconFlow/OpenRouter) + MCP + CN network tips

A practical Chinese-language guide for deploying OpenCode in mainland China on a budget. Covers a free-tier model matrix — DeepSeek V4 Flash for daily tasks, Kimi K2.x for coding, GLM-4-Flash / Qwen2.5-7B (permanently free) as fallback, plus OpenRouter free models — along with a ready-to-use opencode.example.json template, MCP (GitHub/ComfyUI) configuration, and mainland-China networking notes (npm mirror, proxy).

🔗 View Repository

OpenCode Primer - Everything you need to know about OpenCode — a verified, release-tracked community guide.

A community field guide that goes from first prompt to power use — setup, prompt engineering, custom commands, agent skills, plugins, custom agents, MCP servers, Zen and Go billing, TUI mastery, and headless CI. Ships a worked example .opencode/ directory (agents, commands, plugins, skills), seven drop-in specialist agent prompts, and MCP walkthroughs. Every version pin, price, and flag is verified against the current OpenCode release with a visible last-reviewed date.

🔗 View Repository

➕ Add a Resource

🤝 Contributing

Found an Awesome Opencode project?
Submit a Pull Request to add it to the list!

_{Released under CC0 1.0 Universal.}