Docs

Install

Install the uac core and connect it to Claude Code and Codex.

Prerequisites

Python 3.11+ and uv.
At least one of the agents, installed and logged in with your subscription:

npm install -g @anthropic-ai/claude-code
npm install -g @openai/codex
claude   # log in
codex    # log in

uac never calls a model API — it only drives these.

Install the core

Install the uac command:
```
uv tool install unified-agent-context
```
Set it up in a project you work in:
```
cd your-project
uac init
```
This scaffolds AGENTS.md + .agents/, creates the memory database, generates CLAUDE.md and the native Claude skill mirror, and registers the MCP server + hooks in both agents.

Then: trust the workspace (required)

Both agents ignore what uac init wrote until you trust the project. This is a one-time, per-project step — and it’s a security decision, so uac won’t do it for you.

In Claude Code — run claude in the project once and accept the trust prompt. Otherwise it drops the permission rule and silently refuses every memory_write.

In Codex — run codex once and accept the directory-trust prompt. Then, inside Codex, run /hooks and trust the uac hooks — Codex ignores wrapper-installed hooks until you approve them there, and without the session-start hook it never learns the shared memory exists.

See Gotchas for why, and what it looks like when you skip it.

Optional front ends

VS Code extension — a chat panel + memory view.
Terminal app — the uac-chat full-screen TUI.

Both are separate installs; the core above is all you need for the shared layer to work inside claude and codex directly.