Braingent

Get Started

Wire Up Your Agents

One-time setup so every agent reads Braingent before planning and writes to it after work.

Braingent’s contract is simple: every agent should read your memory repo before planning, and write a record after meaningful work. That contract lives in entrypoint files — short, agent-specific Markdown files that the tool already loads automatically.

This page covers the smallest setup that gets each agent on the contract. Detailed integration guides live under Integrations.

The pattern

For every agent you use, you do exactly two things:

  1. Drop the agent’s entrypoint into the right location.
  2. Personalize a small handful of placeholders inside it.

That’s it. There’s no daemon, no service, no plugin to install at this layer. The agent reads Markdown; that’s the whole interface.

Claude Code

Where it goes: the root of any working repo (or ~/.claude/CLAUDE.md for global rules).

What it does: Claude reads CLAUDE.md automatically at session start and treats it as system-level instructions.

cp ~/Documents/repos/braingent/CLAUDE.md ./CLAUDE.md

The starter CLAUDE.md already includes the Engineering Memory section that points Claude at your memory repo and tells it when to search and when to capture. Open it and replace the path placeholder with the absolute path to your memory repo.

Tip: If you want the same memory across every project, put the Braingent block in ~/.claude/CLAUDE.md (your user-level file) so it applies everywhere.

See the full guide: Claude Integration.

Codex / Codex CLI

Where it goes: the root of the working repo as AGENTS.md.

What it does: Codex automatically picks up AGENTS.md for context.

cp ~/Documents/repos/braingent/AGENTS.md ./AGENTS.md

Same pattern as Claude — replace placeholders, commit. See Codex Integration.

ChatGPT (Project / Custom GPT)

ChatGPT doesn’t auto-read files from your machine, so the entrypoint goes into a Project’s “Instructions” field (or a custom GPT’s system prompt).

  1. Open the contents of ~/Documents/repos/braingent/CHATGPT_PROJECT_BRIEF.md.
  2. Paste it into your ChatGPT Project → Instructions.
  3. Optionally, attach the few pinned files (preferences/*.md, the most relevant repo profile) to the Project’s knowledge.

ChatGPT will treat the brief as standing instructions, including the search-before-plan and capture-after-work rules. See ChatGPT Integration.

Gemini CLI

Where it goes: the working directory as GEMINI.md.

What it does: Gemini CLI reads GEMINI.md for project-specific context.

cp ~/Documents/repos/braingent/GEMINI.md ./GEMINI.md

Same placeholder pass, same commit. See Gemini CLI Integration.

Make sure it actually loads

After wiring an agent, run a tiny smoke test. Open a session and ask:

What’s the path to my Braingent memory repo, and what’s the first thing you’ll do before planning a non-trivial task?

A correctly wired agent answers with the absolute path you set and restates the search-before-plan, capture-after-work contract. If it doesn’t, the entrypoint isn’t loading — check the filename, location, and that you saved.

What about MCP?

MCP is the upgrade path. Once an agent supports the Model Context Protocol, you can connect Braingent’s MCP server and the agent gets dedicated retrieval tools (braingent_guide, braingent_find, braingent_get) instead of having to search by hand.

MCP is strictly optional. The entrypoint contract works without it. See Installation → MCP Server and MCP Tools Reference.

When you add a new tool

Pattern repeats:

  1. Find out where that tool reads instructions from.
  2. Drop a Braingent-flavored entrypoint there.
  3. Replace placeholders.
  4. Smoke test.

Markdown is universal; the only thing that changes is the filename.

Where to go next