New

Agent experiences now coordinate multiple harnesses in one conversation.

See the new composer →
Local-first agent orchestration

One prompt.
The right agents.
Working together.

Avalon is a desktop workspace for orchestrating Codex, Claude Code, OpenCode, and compatible agents. Build your own team or start with a proven experience—then follow every handoff in one calm conversation.

Explore agent experiences Meet the new composer
Avalon playgroundInteractive demo
Choose an experience
Ship the new billing settingsClaude plans, Codex builds, Claude reviews
  1. CL
    PlanClaude planner · read only
    Next
  2. CX
    BuildCodex implementer · write
    Queued
  3. CL
    ReviewClaude reviewer · read only
    Queued
Ready when you areEvery handoff stays visible in one conversation.
0 of 3 handoffsOnly the builder receives write access

A composer built for delegation

Describe the outcome. Avalon assembles the work.

Select an individual agent or a coordinated experience, add files and context, choose how approvals should work, and send one clear goal.

avalon / desktopLocal
New conversation

What should the team accomplish?

avalon / apps / desktop

Add an account usage panel to Settings. Match the existing design system, verify subscription states, and review the final diff for regressions.

＋ Add context▱ SettingsPage.tsx▱ provider-store.ts
3 Codex builds, Claude reviews, Gemini testsApprove for me⌄
3 workers across 3 sequential stepsWrite access only for the builderLocal execution in your workspace

Coordinated by Avalon

Agent experiences for work that deserves more than one pass.

Use built-in routes or create an individual or team agent with up to 32 workers across parallel and sequential steps.

01

Plan, build, review

Claude → Codex → Claude

Turn one request into a deliberate sequence: a focused plan, an implementation pass, and an independent review.

02

Research, build, verify

OpenCode → Codex → OpenCode

Map the codebase with a fast research agent, hand the findings to a builder, then run a read-only verification pass.

03

Build, review, test

Codex → Claude → Gemini

Coordinate three models across implementation, review, and testing while keeping every step visible.

The Avalon model

Unified where it helps. Native where it matters.

Avalon gives every harness one shared operating layer without pretending the agents are identical.

01

One goal, many specialists

Choose a ready-made experience or create your own team with different harnesses, models, roles, and instructions.

02

Every handoff stays visible

Avalon keeps agent turns, tool calls, file changes, questions, and approvals in one ordered conversation.

03

Authority stays bounded

Each worker gets an explicit workspace scope and permission ceiling. Reviewers can stay read-only while builders write.

Avalon desktop

More than a chat. A place for agents to work together.

Follow the build on GitHub
Local-firstNative harnessesBounded permissionsDurable handoffs