Braingent
Integrations
Integrations Overview
One memory, every agent — wire Claude, Codex, ChatGPT, Gemini CLI, and Gather Step to the same Braingent repo.
Braingent is built so that every agent reads from and writes to the same Markdown repo. The integration story is therefore boring on purpose: drop the right entrypoint file, replace placeholders, commit. No plugins, no daemons, no API keys.
This page is the map. Each agent has its own page with the exact paths, quirks, and verification steps.
Supported agents
| Agent | Entrypoint | Where it goes | Status |
|---|---|---|---|
| Claude Code | CLAUDE.md | Repo root or ~/.claude/CLAUDE.md | First-class |
| Codex / Codex CLI | AGENTS.md | Repo root | First-class |
| ChatGPT | CHATGPT_PROJECT_BRIEF.md | Project Instructions | Pasted |
| Gemini CLI | GEMINI.md | Working directory | First-class |
“First-class” means the tool auto-loads the file; “pasted” means you put the contents into the tool’s Instructions field.
Partner
| Partner | What it adds | More |
|---|---|---|
| Gather Step | Native QA evidence for qa-generate | Read why |
Braingent + Gather Step is the recommended pairing for any team that ships software with QA — see the page for the full rationale.
The shared contract
Regardless of agent, every entrypoint encodes the same contract:
- Read pinned context before planning.
- Search Braingent for prior decisions, reviews, and learnings.
- State assumptions before proposing changes.
- Capture meaningful work as durable records.
- Never store secrets, raw transcripts, or sensitive personal data.
The starter pack ships entrypoints that already encode this contract. You only need to personalize a few placeholders.
MCP — the upgrade path
For agents that support the Model Context Protocol, Braingent’s MCP server provides token-efficient retrieval as native tools:
braingent_guide()— discover sections of the memory repo.braingent_find(...)— frontmatter filters.braingent_get(...)— fetch a record at summary or full depth.
MCP is strictly optional. The entrypoint contract works without it. Add MCP when an agent supports it and you find yourself wanting cleaner structured retrieval.
See Installation → MCP Server and MCP Tools Reference.
When you add a new tool
The pattern is the same:
- Find where the tool reads instructions from.
- Drop a Braingent-flavored entrypoint there.
- Replace placeholders.
- Smoke-test by asking the agent for the absolute path to your memory repo.
If a future tool can’t read Markdown, it’s the wrong tool. Markdown is the universal interface.