#1Openopen sourcetechnical preview soon

Coding, rediscovered.

Meshed is an open source desktop app that runs coding agents through workflows — repeatable processes for building features, fixing bugs, keeping docs current and paying down tech debt — with every step in plain sight. A first-class chat sits right alongside, for when you’d rather talk it through.

RunWalks you through this page, one step at a time.

Understand the problem

Agent · Claude Code

01pending

A process beats a lucky prompt. It’s reliable, measurable, and it scales.

Findings– Repeatable: the same steps and checks on every task, not just on a good day.– Measurable: every run is recorded, so you can see what to optimise.– Concurrent: run many workflows at once, each task through the same proven pipeline.– Transparent: follow how the agent got there, then read the diff and review it — if that’s your process.

Agree with the findings?Continue

0:42RememberedResults have to be repeatable, not lucky.

Plan the approach

Agent · Claude Code

02pending

Write the process down once. Run it on every task.

A workflow is the way your team already works — reproduce, fix, test, review — as steps an agent follows.

Approve the plan to continue.Continue

1:18RememberedThis repository reviews every change before it merges.

Pick your agents

Claude Code · Codex · Copilot · Antigravity · OpenCode

03pending

Any agent, any model. Open-weight ones too.

Claude Code, Codex, Copilot, Antigravity and OpenCode all run in Meshed, and each step can use whichever you choose. Open-weight models run on your own machine through Ollama or LM Studio.

Claude CodeAnthropic
CodexOpenAI
CopilotGitHub
AntigravityGoogle
OpenCodeOllama · LM Studio

Happy with this line-up?Continue

0:12RememberedCodex implements, Claude Code reviews.

Run it

Agent · Codex

04pending

Agents do the work. You watch every step land.

Start a run from any task. Claude Code, Codex or Copilot take the steps in turn, on your machine or in a dev container — and when a step needs you, the run waits.

Codex asks to run mise run test.Allow and continue

6:05RememberedTests run with mise run test.Follow-upcheckout.spec.ts is flaky — fix it separately.

Review what happened

Agent · Claude Code

05pending

Every step is on the record. Insights shows you the pattern.

Which agent did what, how long each step took, where runs get stuck. One run tells you what happened; a hundred tell you what to change.

Read the review, then carry on.Continue

2:31Follow-up“Run the test suite” is the slowest step — cache the dependencies.

Capture what matters

Memory · Signals

06pending

Nothing learned gets lost. Memories for later, signals for now.

As agents work, Meshed keeps track of what matters. Memories are facts with their source attached, so later runs can recall them. Signals are what needs you: a step asking for permission right now, or a follow-up the run noticed and keeps in working memory until you get to it.