Set up TRW for Aider

Keep the minimal Aider terminal loop you already like, and bolt on durable project memory through TRW’s instruction-first surface.

Prerequisites

Aider installed (pip install aider-install && aider-install or your preferred route).
Python 3.10+ (3.11+ recommended) and pip 22+.
A git repository — Aider requires one anyway.

Install

curl -fsSL https://trwframework.com/install.sh | bash

When the installer prompts for target clients, select Aider. The bootstrap installs trw-mcp and writes the Aider surface into the repo.

What the installer writes

AGENTS.md — the shared, durable instruction file Aider reads on startup. Encodes the TRW tool lifecycle so the ceremony loop is obvious from the first turn.
.aider/instructions.md — repo-local supplemental guidance loaded alongside AGENTS.md when your Aider build supports it.
.trw/config.yaml — repo-local TRW config listing Aider in target_platforms.

No hook directory, no MCP config, and no agents directory — Aider’s surface is instruction-first by design.

Run your first session

Start Aider from the project root the way you normally would — e.g. aider --model sonnet or your preferred model flag. Aider reads AGENTS.md and any .aider/instructions.md-compatible file, so the TRW ceremony instructions reach the model on the first turn.

Ask the model to call trw_session_start() — the instructions prompt this on turn one.
Make changes the usual Aider way. Commit frequently.
When you discover a gotcha, ask the model to call trw_learn(summary, detail).
End the session by calling trw_deliver(). Without it, nothing from this session carries forward.

Differences from Claude Code

No hooks. Ceremony is enforced entirely through the instruction file. There is no pre-compact, session-start, or deliver hook to fall back on.
No agent teams. The 12-specialist roster isn’t available — Aider runs a single model loop. You still get the core TRW memory, learnings, and run state.
No skills. TRW’s slash-skill catalog requires a host that supports slash commands. Aider doesn’t, so the functionality moves into the instruction file and MCP tool calls.
Same memory store. Learnings captured here are visible to every other TRW-wired client in the same repo.

Common problems

Next steps

Core concepts

Sessions, runs, learnings, and the ceremony loop that TRW enforces on every client.

See the flagship profile

Claude Code is the reference integration — use it to understand the full surface.

Workflows overview

The six-phase lifecycle your Aider sessions will run through.