Orkas — Your AI Team
Scale your business without hiring.
Download · User guide · Agents & skills · Source code
English · 简体中文
https://github.com/user-attachments/assets/c41678f0-9f50-48be-aae0-fae95977cff9
Orkas is an open-source, local-first desktop AI workspace for indie builders, one-person businesses and small teams. Describe your goal; Commander can handle the work, coordinate specialist agents and deliver reports, spreadsheets, presentations, code, images or videos for you to inspect and revise.
Orkas-Docs is the official guide catalog. Issues here are published articles, not a support queue. Every topic has separate English and Simplified Chinese versions. Open, locked Issues are maintained guides; closed Issues are archived content.
Complete your first task
- Download and open Orkas, then choose an available managed official model or connect your own supported provider.
- Open a new task. Describe the outcome, audience and format, and add only the files or resources needed.
- Supply any required decisions, inspect the deliverables and request specific revisions. Use a Project when future tasks should share context.
For example:
Prepare a monthly business review from this sales workbook and these meeting notes. Reconcile the numbers first, then deliver a one-page decision summary and an editable workbook. Flag missing data and unresolved questions. Do not send anything to anyone.
See the full user guide for models, projects, automations, connectors, Library and result management.
Find a guide by outcome
| What you want to do | Start here | Typical deliverables |
|---|---|---|
| Compare markets and assemble evidence | Market research · Document Q&A | Decision brief, citations, evidence table |
| Prepare a business or client review | Data analysis · Office reports | Workbook, report, presentation |
| Produce a product launch | Content · Images · Video | Copy, visuals, finished video |
| Develop software or a product workflow | Software development · Product workflow · UI design | Source, interface and acceptance evidence |
| Publish a customer self-service tool | Serve customers | Editable app, public link and publication details |
| Coordinate a team or client delivery | Shared projects · Client delivery | Shared rules, Library files, owner to-dos and a delivery pack |
| Review a store or recurring operations | Store review · Team updates | Checked data, proposed actions |
| Improve search and AI visibility | SEO and GEO | Audit evidence, prioritized fixes |
| Produce narration, transcripts or an audio episode | VoiceStudio | Narration, TXT/SRT and a revisable audio project |
Coordinate local agents
Orkas can coordinate supported local installations of Claude Code, Codex, OpenCode, OpenClaw and Hermes. Install and authenticate the tool first, then connect it through Orkas’s external-agent entry. External tools use their own model settings; native CLI permissions and Orkas connector permissions are separate controls.
Use the local-agent collaboration article for task-ownership ideas and the current user guide for app entry points and permissions. When agents could edit the same files, agree on ownership or use isolated working directories.
Team work and recurring tasks
Shared projects let members use common goals and rules, project memory, Library files and to-dos. Task conversations, Automation, CLI agents, model credentials and personal credits remain private; custom agents need explicit sharing. Personal cloud sync and project collaboration are separate, and a local save does not mean every member has received the update. See how shared projects work.
To run work later, separately create and enable a one-time or recurring Automation, select its device and inspect the first result. Closing the window can leave Orkas running in the background; quitting the app, sleep or shutdown affects execution. Scheduled work and to-do auto-advancement are configured separately; producing a deliverable does not enable them.
Compatible public browser apps can be published on an eligible plan, but cannot directly use desktop AI, local files or connectors. Check compatibility and plan limits before publication.
Choose a specialist
| Agent | Main use |
|---|---|
| Commander | Planning, coordination, follow-up and synthesis |
| DeepResearcher | Cited research and evidence collection |
| ContentWriter | Articles, reports and launch copy |
| OfficeWorker | Documents, spreadsheets, PDFs and reconciliation |
| PptMaker | Editable presentations |
| ProductDeveloper | Development, repairs and code review |
| UIDesigner | Product flows and editable interfaces |
| ImageStudio | Posters, covers and image production |
| VideoStudio | Production, editing, captions and voice |
| SeoGeoAgent | Search visibility audits and content improvements |
| ECommerceAnalyzer | Reconciled store metrics, profit drivers and stock risk |
| ECommerceOperator | Item-level store change plans and verified results |
| ECommerceResearcher | Product shortlists, unit economics and validation |
| ECommerceReviewer | Review themes, VOC evidence and improvement priorities |
| ECommerceWriter | Channel-specific listing copy and creative briefs |
| StockAnalyser | Sourced stock research, backtests and portfolio risk |
| VoiceStudio | Narration, local transcription and audio editing |
The agent guides below explain inputs, fit and deliverables. Find additional roles and skills in the marketplace.
Models, data and costs
- Chats, project files and deliverables stay on your device by default. Models, connectors, sync and sharing can send or store the data those features need.
- Own-provider credentials stay local and model calls go directly to that provider. Optional Orkas-managed models use the managed model service.
- Compatible local OpenAI-style model endpoints are supported. This does not automatically make search, external tools or the whole workflow offline.
- External CLIs use their own accounts and model settings. Managed models, your providers and other services have separate availability and costs; check the app and provider terms.
- Model API keys and task attachments do not move through cloud sync. Add supported reusable material to Library when appropriate and check the target device’s sync state.
The desktop source uses the MIT license. Read the security overview for data boundaries and pricing for current managed plans.
Published guides
These Open, locked Issues are the maintained English guides. Each guide explains the workflow, relevant inputs, deliverables and limits, with a practical task you can adapt.
Agent guides
| Agent | Published guide |
|---|---|
| Commander — Orchestrate an AI agent team in Orkas | Read the Commander guide → |
| ContentWriter — Publication-ready AI writing in Orkas | Read the ContentWriter guide → |
| DeepResearcher — Evidence-grounded AI research in Orkas | Read the DeepResearcher guide → |
| ImageStudio — Create and edit images in Orkas | Read the ImageStudio guide → |
| OfficeWorker — Multi-file office delivery in Orkas | Read the OfficeWorker guide → |
| PptMaker — Editable, reviewable AI presentations in Orkas | Read the PptMaker guide → |
| ProductDeveloper — Repository-aware AI engineering | Read the ProductDeveloper guide → |
| SeoGeoAgent - Prioritize SEO Issues and Fix Them in the Repo | Read the SeoGeoAgent guide → |
| UIDesigner — HTML-first AI UI design in Orkas | Read the UIDesigner guide → |
| VideoStudio — Create and intelligently edit video in Orkas | Read the VideoStudio guide → |
| ECommerceAnalyzer — Reconcile Store Metrics and Explain Profit Changes | Read the guide → |
| ECommerceOperator — Review Store Changes and Verify Each Result | Read the guide → |
| ECommerceResearcher — Product Shortlists, Unit Economics and Validation | Read the guide → |
| ECommerceReviewer — Review Themes, VOC Evidence and Improvement Priorities | Read the guide → |
| ECommerceWriter — Channel-Specific Listing Copy from Verified Product Facts | Read the guide → |
| StockAnalyser — Traceable Research, Backtests and Portfolio Risk | Read the guide → |
| VoiceStudio — Narration, Transcripts and Revisable Audio Edits | Read the guide → |
Use-case guides
| Workflow | Published guide |
|---|---|
| AI Weekly Team Operations — Updates, Decisions and Follow-up | Read the guide → |
| Build and Publish a Customer Self-Service Web App with Orkas | Read the guide → |
| Document Q&A with Citations, Versions and Escalation | Read the guide → |
| Create a Consistent Marketing Pack — Copy, Images and Video | Read the guide → |
| AI Data Analysis — Checked CSV, Excel and Sheets Results | Read the guide → |
| Ship a Product Workflow — Design, Code and Acceptance Evidence | Read the guide → |
| AI Image Production — Accurate Layouts, Generation and Edits | Read the guide → |
| AI Video Editing — Footage, Edit Decisions and Reviewable Output | Read the guide → |
| Monthly Business Review — Reconciled Workbook, Report and Slides | Read the guide → |
| AI Market Research — Evidence for a Market-Entry Decision | Read the guide → |
| SEO and AI Visibility — Citations, Evidence and Prioritized Fixes | Read the guide → |
| AI UI Design — Editable Interfaces and Complete Product States | Read the guide → |
| E-commerce Store Review — Weekly Priorities, Draft Replies and To-dos | Read the guide → |
| Client Delivery in Orkas — Repeatable Reports with Separate Client Context | Read the guide → |
| Software Development with Orkas — From UI Requirements to Verified Code | Read the guide → |
| Small-Team AI Collaboration — Shared Rules, Owners and Reviewable Files | Read the guide → |
Comparison guides
| Comparison topic | Published guide |
|---|---|
| Activepieces vs Orkas — Repeatable Flows and AI Deliverables | Read the comparison → |
| AnythingLLM vs Orkas — Document Answers and Reviewable Deliverables | Read the comparison → |
| AutoGPT vs Orkas — Autonomous Workflows and a Desktop AI Team | Read the comparison → |
| Bolt.new vs Orkas: an app builder in a browser tab, or an agent team on your machine. | Read the comparison → |
| Orkas vs ChatGPT: your machine, or the cloud. | Read the comparison → |
| Orkas vs Claude Code: an agent team, or a terminal. | Read the comparison → |
| Orkas vs Cline: a multi-agent desktop team, or a coding agent in your editor. | Read the comparison → |
| Codex vs Orkas: OpenAI's coding agent, and a desktop team that can run it for you. | Read the comparison → |
| CrewAI vs Orkas: a framework you write code with, or an app you talk to. | Read the comparison → |
| Orkas vs DeepSeek Harness: a finished desktop AI team, or a plugin-composed agent runtime. | Read the comparison → |
| Devin vs Orkas: a hosted AI software engineer, or an open-source agent team on your machine. | Read the comparison → |
| Dify vs Orkas — AI Services, Desktop Work and Published Browser Apps | Read the comparison → |
| Flowise vs Orkas — Visual AI Flows and Conversation-Led Delivery | Read the comparison → |
| Goose vs Orkas — A Local Agent and a Specialist Desktop Team | Read the comparison → |
| Hermes Agent vs Orkas — Agent Runtime, Desktop Team and Scheduled Work | Read the comparison → |
| Orkas vs LangChain: use the app, or build with the framework. | Read the comparison → |
| LibreChat vs Orkas — Shared Chat Deployment and Desktop Project Delivery | Read the comparison → |
| Lovable vs Orkas: one tool builds the product, the other does everything the product needs around it. | Read the comparison → |
| n8n vs Orkas: data plumbing, or an agent team that produces the deliverable. | Read the comparison → |
| Orkas vs OpenClaw: desktop AI workforce or self-hosted assistant. | Read the comparison → |
| OpenCode vs Orkas: a coding agent in your terminal, and a desktop team that can run it for you. | Read the comparison → |
| Orca vs Orkas: similar names, two different tools — one races coding agents, one ships the whole release. | Read the comparison → |
| Open-source HeyGen alternative: avatar video without a GPU of your own. | Read the guide → |
| OpenMontage vs Orkas: assemble the pipeline yourself, or install one. | Read the guide → |
| Opus Clip vs Orkas: a hosted clipper, or a desktop studio you own. | Read the guide → |
| Remotion vs Orkas: write React for every frame, or describe the video you want. | Read the guide → |
| Local-First AI Agent Tools — Check Files, Models and Shared Data | Read the guide → |
| Best multi-agent AI desktop apps in 2026: a lead agent and a team of specialists | Read the guide → |
| Claude Code alternatives in 2026: eight options, grouped by what you are replacing | Read the guide → |
| Claude Code vs Codex vs OpenCode: what each one lets a host ask of it | Read the guide → |
| Claude Code vs Hermes Agent: one you steer, one you leave running. | Read the guide → |
| Codex vs OpenCode: one is the vendor's own agent, one is yours to fork. | Read the guide → |
| Hermes vs Codex: these are not the same category, and that is the answer. | Read the guide → |
| Claude Code Projects vs Orkas — Cloud Threads or Mixed Local Agents | Read the guide → |
| Claude Cowork Alternatives — Choose by Workflow, Model Access and Deployment | Read the guide → |
| Open-Source Desktop AI Agents — Compare File Work, CLI Sessions and Computer Use | Read the guide → |
| OpenChamber vs OpenCode Desktop vs Orkas — Choose the Right Agent Workspace | Read the guide → |
| video-use vs Orkas VideoStudio — A Skill in Your Assistant or a Desktop Video Agent | Read the guide → |
Blog guides
| Article | Published guide |
|---|---|
| Context Compaction Cuts by Token Count, Not by What Is Safe to Forget | Read the article guide → |
| The Layer That Turns a Model Into a Product: Engineering Orkas's Agent Harness | Read the article guide → |
| Declared Done Is Not Verified Done: Milestone Design for Long-Horizon Agents | Read the article guide → |
| Multi-Agent Orchestration in Practice: How Orkas Runs a Lead Agent and Its Sub-Agents | Read the article guide → |
| Loop Detection Is Not Stall Detection: Catching Agents That Spin Without Repeating | Read the article guide → |
| How to Run Claude Code and Codex Together — One Chat to Orchestrate Both | Read the article guide → |
| Cloud Sync in Practice: How Orkas Syncs Data Across Devices | Read the article guide → |
| Rewriting the Agent's Foundation: A Ground-Up Refactor of Orkas | Read the article guide → |
| How to Get Cited by ChatGPT: What Actually Decides Whether You Get Quoted | Read the article guide → |
| Bigger Was Not the Point: How Kimi K3 Scales Information Flow in Three Directions | Read the article guide → |
| What Is Local-First AI? Your Data, Your Keys, Your Machine | Read the article guide → |
| BEACON: Milestone-Guided Long-Horizon Agents | Read the article guide → |
| An Agent That Gets Better on Its Own: Inside Orkas's Self-Evolution | Read the article guide → |
| Monthly Client Report Template: What to Include and How to Write It | Read the guide → |
| What to Charge as a Freelance Marketer (With a Free Pricing Calculator) | Read the guide → |
| In-House, Agency, or On Your Own: How to Actually Decide | Read the guide → |
| Why Your AI Video Bill Is Mostly Retries | Read the guide → |
| Amazon Seller MCP: What It Gets You, and What It Costs | Read the guide → |
| Connect eBay to Claude: Credentials, RuName, and What the Sandbox Will Not Do | Read the guide → |
| Connect Etsy to Claude: What Each Route Can and Cannot Do | Read the guide → |
| Connect Shopify to Claude: What Works and What Breaks | Read the guide → |
| Connect WooCommerce to Claude: The Store With No App Approval | Read the guide → |
| Before You Give an Agent Write Access to Your Ad Account | Read the article guide → |
| How ChatGPT and Gemini Describe Your Products | Read the article guide → |
| AI Visibility Audit: the $3,000 Agency Quote, Three Tools, and Doing It Yourself | Read the article guide → |
| Reconcile Supplier Tracking Numbers Without Reading Every Order | Read the article guide → |
| Agentic SEO — Build a Repeatable Evidence-to-Fix Workflow | Read the guide → |
| Automate Video Editing — Evidence, Cut Lists and Repeatable Assembly | Read the guide → |
| Batch Video Editing — One Approved Cut and a Checkable Export Matrix | Read the guide → |
| Build AI Employees — Start with One Job, Then Connect the Team | Read the guide → |
| CapCut with ChatGPT or Codex — Official Drafts and the Orkas File Workflow | Read the guide → |
| Claude and DaVinci Resolve — Choose Studio Control or a File Handoff | Read the guide → |
| Claude and Final Cut Pro — Rough Cuts, SRT Captions and Editable Handoffs | Read the guide → |
| Claude and Premiere Pro — Rough-Cut Files, Adobe Tools and Community MCP | Read the guide → |
| cmux on Windows? Choose a Working Multi-Agent CLI Setup | Read the guide → |
| Connect ServiceM8 to Claude or Orkas — Read Jobs and Prepare Follow-up | Read the guide → |
| DaVinci Resolve MCP — Edition Checks, Connection Routes and Verification | Read the guide → |
| Edit Videos with Claude — Tool-Based Cuts and a CapCut File Handoff | Read the guide → |
| GPT-6 Astra Spatial Reasoning — Read the Evidence Behind 3D and Robot Demos | Read the guide → |
| Human–Agent Collaboration — Clean Context, Owned Tasks and File Handoffs | Read the guide → |
| Product Video Prompts — Verified Facts, Timed Scripts and Shot Lists | Read the guide → |
| Shared Projects in Orkas — Invitations, Private Data and Sync Recovery | Read the guide → |
| Vidu S2 vs PixVerse R2 — Compare Real-Time Video Tasks and Technical Trade-offs | Read the guide → |
Frequently asked questions
Do I need my own model key?
Not necessarily. Choose an available managed official model or connect a supported provider. The two paths have different account and billing boundaries.
Does local-first mean fully offline?
No. Check inference, external tools, search and sync separately. A local model does not stop other selected services from making network requests.
Can work run every week automatically?
Auto supports one-time and recurring tasks while the assigned device, access and input data remain available. Review the first result before relying on later runs.
Which platforms are supported?
The website offers macOS and Windows installers. See the main repository and download page for source builds and other platform details.
How should I compare tools?
Read the criteria, evidence, date and scope stated in the website original, then use the additional examples to check your own task. Rankings and measurements depend on their stated conditions; confirm changing features, models and plans against current primary documentation.