Symaira Brain

Symaira Brain

[](https://github.com/danieljustus/symaira-brain/actions/workflows/ci.yml) [](https://github.com/danieljustus/symaira-brain/releases/latest) [](https://github.com/danieljustus/symaira-brain/tree/coverage-data) [](LICENSE) [](go.mod)


`symbrain` is the portable agent-context layer for AI coding harnesses. It exposes the three Symaira *state cores* — credentials, memory/entities, and the skill catalog — behind one MCP gateway, with one **profile** per harness connection controlling exactly what that harness is allowed to see.

Memory and skills ship inside this binary (absorbed 2026-08-21); credentials stay behind the separate `symvault` process on purpose.

Point Claude Code, Cursor, Codex, Gemini, or opencode at `symbrain` once, and every one of them talks to the same underlying vault, memory, and skills — each through its own profile, each seeing only what that profile exposes.

**Status:** `v0.8.0` released, in active development. Interfaces may still change before `v1.0.0`.

Why symbrain

• One gateway, three state cores. Credentials, memory, and the skill catalog behind a single MCP surface — one server to wire into a harness, not three.
• Least exposure by profile, not by trust. Each harness connection is bound to a profile that controls what it sees, so a shared or untrusted harness never even sees a tool that could read a secret.
• Portable across every major harness. Register once in Claude Code, Cursor, Codex, Gemini, or opencode and each tal