Avalon

The open-source coding agentorchestrator with persistent memory.

Run and compose Codex, Claude Code, OpenCode, and Hermes from one interface. Build chains, parallel teams, and lead-agent workflows with the models and accounts you already use. Let your agents learn as they work, and keep what they know when you switch.

Download for macOSView on GitHub ↗
Avalon
Ship custom 404Morgana · Completed
12 tests passed
You

Add a custom 404 page that matches the marketing site and verify it on mobile.

M
Morgana3 stages · 1m 42s
✓Claude plannedMapped the route and design constraints · 18s
✓Codex builtImplemented the page and responsive styles · 54s
✓Gemini testedVerified desktop, mobile, and the 404 response · 30s
Context carried to CodexPlan · design constraints · 3 target files

Added a responsive Avalon 404 page with a clear return path. The route returns the correct status and passes all web checks.

3 files changed+155 −0Build passed
avalon

Agent work should not be split across apps.

Every coding harness keeps its sessions, models, accounts, permissions, and context in its own interface. Working across harnesses means switching apps and piecing the work back together by hand.

Switch agents without starting over.

Avalon carries an inspectable handoff when work moves between harnesses. The next agent receives the objective, completed work, changed files, command results, open questions, and next steps it missed. Your original message stays unchanged.

Avalon sits above the harness.

Keep the subscriptions, providers, and CLIs you already use. Avalon gives them a shared place to start work, choose a route, inspect progress, and keep the resulting context together.

  • Bring your own Claude Code, Codex, OpenCode, Hermes, or custom agent.
  • Choose the model and account each harness supports.
  • Keep approvals, messages, tools, and handoffs in one conversation.

Use an AI gateway such as Vercel AI Gateway to access supported models with one key, then route that shared gateway across your agents.

Your agents learn. Avalon remembers.

As your agents learn your preferences, projects, and ways of working, Avalon keeps that knowledge with you across chats and agents. Switch from Codex to Claude Code, OpenCode, or Hermes without starting over. Memory stays local, visible, editable, and under your control.

Define multi-harness agents.

Assign each stage to the harness that fits it best, with its own instructions, workspace access, and handoff. Lancelot pairs Claude for planning with Codex for implementation. Morgana adds Claude review and Gemini testing after Codex builds. Use either as-is, customize the stages, or define your own team in source.

More places to dispatch work.

We’re adding support for more harnesses including Cursor, Gemini CLI, and eve, dispatch from web and mobile, and optional cloud support for teams that need shared sessions and policies across devices.