The metaharness for coding agents

One app.
Every harness.

Avalon sits above Codex, Claude Code, OpenCode, and custom agents so you can combine, control, and monitor them from one place.

Avalon
Chats
account usage
⚙ Settings
Avalon AgentsLocal

Avalon Agents

One prompt. Any harness.

▱ Default ProjectApprove for me
Plan & BuildReady
  1. 1Claude plannerQueued
  2. 2Codex builderQueued
CodexClaude CodeOpenCodeACPCustom agents

Your agents become one system.

Every harness has its own sessions, permissions, models, and UI. Avalon gives them a shared control plane without replacing the runtimes you already trust.

01

Switch

Move between Codex, Claude Code, OpenCode, and ACP agents without changing how you work.

02

Orchestrate

Give planning, implementation, review, and testing to different agents in one sequential or parallel run.

03

Control

Set models, workspace access, instructions, approval ceilings, and account boundaries per agent.

04

Monitor

Keep messages, tools, diffs, questions, and handoffs visible in one durable conversation.

Build the agent you actually want.

An Avalon agent can be one harness or a coordinated team. Start with a built-in route, then change its models, roles, access, and handoff order.

A02 steps

Plan & Build

Claude plans, Codex builds

Claude CodeCodex
A03 steps

Full Check

Codex builds, Claude reviews, Gemini tests

CodexClaude CodeGemini
A03 steps

Research First

OpenCode researches, Codex builds, Claude reviews

OpenCodeCodexClaude Code

One control plane. Every client.

The desktop app is the first Avalon client. The server architecture is designed to keep agent sessions, policies, and orchestration available wherever your team works.

Multi-client access is on the roadmap. Today, Avalon runs locally on macOS.

Shared layerAvalon ServerSessions · policies · orchestration
macOS appAvailable
WebComing soon
CLIComing soon
SlackComing soon

One prompt. Any harness.

Use the best agent for each part of the work—and keep it together.

Get Avalon on GitHub ↗