What is Meldwork and what does it do?

Meldwork is a local-first multi-Agent collaboration workspace for macOS. It brings your existing Agents into one working group to share context, discuss approaches, carry out authorized work, and check results. You choose the participants, set goals and permissions, and step in when information or key decisions are needed.

How is Meldwork different from tmux or zellij for running multiple AI agents?

Meldwork adds task-level collaboration across selected Agent CLIs: shared context, peer discussion, requests for further contributions, authorized execution, and review of deliverables. This reduces the work of relaying replies and coordinating next steps yourself. Collaboration history and run state stay local so you can revisit progress and blockers.

Which General Agents does Meldwork support?

Thirteen agents today — Codex, Claude Code, Gemini CLI, Qwen Code, Kimi Code, MiMo Code, OpenCode, OpenCodeReview, Hermes, Zcode, Pi Agent, OpenClaw, and WorkBuddy. Approved Agent Connectors can be added through the Agent Connector SDK; custom executables use the desktop custom-Agent path.

Can Meldwork run Codex and Claude Code at the same time on the same task?

Yes. In Concurrent Responses, selected Agents receive the same task snapshot and reply independently. In Auto Discussion, they can build on peer replies, request contributions, carry out authorized work, and check deliverables across rounds. You can mix compatible Agent CLIs in the same working group.

What does Auto Discussion V4 actually do?

Selected Agents start with independent proposals, then discuss differences, request peer contributions, carry out authorized work, and check deliverables. A delivery owner judges progress and whether more work or a user decision is needed. You do not need to direct every round, but missing information, permissions, or key decisions can still require your input.

Do I still need to intervene?

You choose the Agents, goal, context, and permissions. Auto Discussion lets Agents advance the work without you directing each turn, but missing information, permission requests, or key tradeoffs may require your input. Workspace writes are opt-in. Check important deliverables before using them; peer review does not guarantee correctness.

Is Meldwork free? What's the license?

Open source under the Apache License 2.0. Commercial use, private use, modification, and redistribution are permitted under the license terms. No separate commercial authorization or license fee is required. Apache-2.0 does not grant trademark rights.

Does Meldwork work on Intel macOS, Windows, or Linux?

The current V1.0.7 preview targets Apple silicon macOS (ad-hoc signed, not notarized — open Open Anyway in System Settings → Privacy & Security on first launch). Cross-platform builds are on the roadmap; the app is Electron so the same source compiles everywhere once those targets land.

Meldwork — Set the Goal. Let Agents Work Together.

DETECTED ON YOUR MACHINE

Works with the agents you already run.

Meldwork detects a compatible local Agent command and exposes its readiness, capabilities, and session behavior. Approved Agent Connectors can be added through the Agent Connector SDK.

THE SHORT ANSWER

What is Meldwork?

Meldwork is a local-first multi-Agent collaboration workspace for research, analysis, writing, planning, review, and implementation. Selected Agents advance the same task through discussion, authorized execution, and peer review. You set the boundaries and make key decisions when needed.

Supported agents 13+

Codex, Claude Code, Gemini CLI, Qwen Code, Kimi Code, MiMo Code, OpenCode, OpenCodeReview, Hermes, Zcode, Pi Agent, OpenClaw, WorkBuddy — detected natively on your machine.

Modes 3

Direct · Concurrent Responses · Auto Discussion V4

Working together

Discuss → Execute → Check → Refine

Agents build on peer contributions and request the next steps, reducing manual coordination.

Runs on

macOS · Apple silicon · Electron

Local conversations and run history; models may use external providers.

THE WORK CELL

Set up the task. Let the group move it forward.

01

Select

Choose the local Agents and participants. Participants are always selected by you — never auto-assigned from a roster.

02

Scope

Bound the goal, working directory, context, and permissions. Every run starts from a frozen task snapshot.

03

Collaborate

Start Auto Discussion so Agents can discuss, request contributions, carry out authorized work, and check deliverables.

04

Decide & check

Step in for missing information, permissions, or key decisions. Check the deliverables and request changes when needed.

COLLABORATION MODES

Three ways agents work together.

Work with one Agent, compare independent replies, or let a group advance a task together.

>

Direct

One selected Agent keeps its conversation and native session when supported.

Best for focused work with one Agent.

*

Concurrent Responses

Selected Agents receive the same frozen task snapshot and return independent replies in stable order.

Best for comparing approaches before choosing one.

#

Auto Discussion V4

Agents propose, discuss differences, request peer contributions, carry out authorized work, and check results.

Best for multi-step work with less manual coordination.

WHAT SUPPORTS THE WORK

Less coordinating. More work moving forward.

Shared context and peer contributions help Agents advance the task. You keep control of the boundaries and key decisions.

You Set the Boundaries

Choose the goal and permissions. Workspace writes are opt-in; requests for information or key decisions come back to you.

Decide when needed

Shared Context

Agents start from the same task snapshot and receive peer contributions as discussion progresses.

Less context relaying

Agents Advance the Work

Agents request peer input, continue authorized work, and review deliverables. Revisit replies and run status when you need to.

Contribute, check, refine

Local-first Boundary

Electron work cell on your machine. No remote Agent fleet. Conversations stay local.

Your CLI, your data

INSIDE A REAL RUN

A release review, from proposals to a user decision.

Historical V1.0.3 session excerpts: Codex, Hermes, and OpenCodeReview discussed a private-beta release, with the user steering a key tradeoff. This example shows discussion and human input, rather than an end-to-end execution task.

Auto Discussion V4 Launch Readiness Council

Round 1 · Propose

R1 · 05:34

CodexProposal

Ship a private, invitation-only preview with one promise: keep a real task coherent while users move between Agents. Don't market workflow automation, universal compatibility, or public distribution readiness yet.
R1 · 05:38

HermesEvidence challenge

The beta should test behavior, not enthusiasm. Measure whether users resume the same task, bring in a second Agent for a clear reason, and use the trace before accepting a result. Without those signals, the category story is still unproven.
R2 · 05:44

OpenCodeReviewEngineering review

The MVP is stable enough for a controlled evaluation, but not for a public download. Tests and packaging pass; the blocker is trusted macOS signing and notarization. Keep the first release private and label unsupported platforms explicitly.
R2 · 05:52

YouReconciliation

Round 2: reconcile the disagreement. Keep only decisions we can verify this week and give me one acceptance gate.
Adoption

Human GateDecision

Ship V1.0.3 to the private preview only. No-Go for public release. Reopen after Developer ID signing and notarization pass on a clean Mac.

1 / 5

INSIDE THE WORK CELL

One workspace for the group and its work.

Local Agent discovery

Pick the participants before any work runs. Meldwork detects installed CLIs and shows their adapter status.

Multi-agent review

Agents compare approaches and check each other's contributions in the same task context.

Auto Discussion V4

Agents discuss, request contributions, and advance authorized work across rounds. Step in when a decision needs you.

Direct mode

Work with one Agent and continue its native session when supported. Keep the conversation and results together.

Local Agent discovery

Agents discussing and reviewing a shared task

Multi-agent review

Auto Discussion V4

Direct multimodal session with local files

Direct mode

Try a task such as reviewing a PR, analyzing a bug, or drafting a plan. Choose compatible Agents, give them shared context, and start Auto Discussion. They can compare approaches, request further work, and check deliverables. Authorize workspace writes when implementation is needed, and respond when they need a key decision.

HOW MELDWORK DIFFERS

Spend less time managing the handoffs.

Meldwork brings the Agent tools you already run into a local-first collaboration workspace. Let the group discuss and advance the task, while you focus on the goal, boundaries, and key decisions.

Agents coordinate the next contribution

Agents can request a peer's input or review, and the delivery owner judges whether more work or a user decision is needed.
Authorized execution and peer review

Enable workspace writes for implementation tasks. Agents can carry out work within that scope and ask peers to check the deliverables.
Local-first boundaries

Runs in your Electron work cell with existing CLIs. No cloud Agent fleet, no required server, conversations stay local.
Shared context and local history

Keep the brief, peer replies, run state, and deliverables together so you can revisit progress and continue the work.

WHERE MELDWORK FITS

Where task collaboration fits in your workflow.

Meldwork focuses on shared tasks across the local Agent tools you choose. Here is how that focus relates to other ways of organizing Agent work.

What Meldwork is, and isn't.

Is

A local-first desktop workspace for multi-Agent collaboration.
A working group built from the Agent CLIs you select.
Shared context, authorized execution, peer review, and local run history.
Workspace writes as opt-in workflow controls.
Open source under the Apache License 2.0.

Isn't

Not a hosted Agent fleet or general Agent framework.
Not automatic participant selection from a larger roster.
Not a guarantee that every task finishes without your input.
Not an operating-system sandbox — writes happen through your shell.
Not fully offline — a selected Agent may send prompts to its own CLI provider.

ANSWERS

Questions teams ask before adopting.

Short, honest answers — the long versions live in architecture.md ↗.

Meldwork is a local-first multi-Agent collaboration workspace for macOS. It brings your existing Agents into one working group to share context, discuss approaches, carry out authorized work, and check results. You choose the participants, set goals and permissions, and step in when information or key decisions are needed.
Meldwork stores conversations and coordinates work locally. Selected Agents may send prompts, attachments, or Skills to the model provider configured in their own CLI. Local-first describes where Meldwork stores and coordinates the work; it does not mean the model runs offline.
Thirteen today — Codex, Claude Code, Gemini CLI, Qwen Code, Kimi Code, MiMo Code, OpenCode, OpenCodeReview, Hermes, Zcode, Pi Agent, OpenClaw, and WorkBuddy. Approved Agent Connectors can be added through the Agent Connector SDK; custom executables use the desktop custom-Agent path.
Selected Agents start with independent proposals, then discuss differences, request peer contributions, carry out authorized work, and check deliverables. A delivery owner judges progress and whether more work or a user decision is needed. You do not need to direct every round, but missing information, permissions, or key decisions can still require your input.
Meldwork adds task-level collaboration across selected Agent CLIs: shared context, peer discussion, requests for further contributions, authorized execution, and review of deliverables. This reduces the work of relaying replies and coordinating next steps yourself. Collaboration history and run state stay local so you can revisit progress and blockers.
You choose the Agents, goal, context, and permissions. Auto Discussion lets Agents advance the work without you directing each turn, but missing information, permission requests, or key tradeoffs may require your input. Workspace writes are opt-in. Check important deliverables before using them; peer review does not guarantee correctness.
Open source under the Apache License 2.0. Commercial use, private use, modification, and redistribution are permitted under the license terms. No separate commercial authorization or license fee is required. Apache-2.0 does not grant trademark rights — see COMMERCIAL_USE.md ↗.
The current V1.0.7 preview targets Apple silicon macOS (ad-hoc signed, not notarized — open Open Anyway in System Settings → Privacy & Security on first launch). Cross-platform builds are on the roadmap; the app is Electron so the same source compiles everywhere once those targets land.

Set the Goal. Let Agents Work Together.

Works with the agents you already run.

What is Meldwork?

Set up the task. Let the group move it forward.

Select

Scope

Collaborate

Decide & check

Three ways agents work together.

Direct

Concurrent Responses

Auto Discussion V4

Less coordinating. More work moving forward.

You Set the Boundaries

Shared Context

Agents Advance the Work

Local-first Boundary

A release review, from proposals to a user decision.

One workspace for the group and its work.

Local Agent discovery

Multi-agent review

Auto Discussion V4

Direct mode

Spend less time managing the handoffs.

Agents coordinate the next contribution

Authorized execution and peer review

Local-first boundaries

Shared context and local history

Where task collaboration fits in your workflow.

What Meldwork is, and isn't.

Is

Isn't

Questions teams ask before adopting.