# Tribal

> Semantic memory served over MCP. Your team's tribal knowledge, made queryable.
> Tacit engineering knowledge captured once and recalled by your coding agents the moment
> it's relevant. Self-hosted on your own Postgres. Source-available under the Elastic License 2.0.

Tribal is a human-knowledge layer that coding agents read from. It stores the reasoning
behind a load-bearing decision, the heuristic someone keeps reaching for, the breakthrough
that closed a gnarly bug, captured as portable principles indexed by the situation they
apply to. Not what your agent did. What your team knows.

## Install

The intended setup is two steps. Add the skills, then tell your agent.

- [Step 1: add the skills](https://tribal.build/#get-started): Run `npx skills add tribal-memory/skills` to install the install-and-use skills into your harness.
- [Step 2: tell your agent](https://tribal.build/#get-started): Say "Set up Tribal for me" to your coding agent. The skill installs Tribal, registers your project, mints a token, and wires your harness's MCP config. You watch.

## What it does

- [Captured as portable principles](https://tribal.build/#how-it-works): Tribal distils a session into a reusable principle: the decision, when it applies, and the trap it avoids. It stores the insight you would want next time, not a summary of what happened.
- [Retrieved by the situation, not the keyword](https://tribal.build/#how-it-works): Items are indexed by when they apply. Ask for help with a problem worded nothing like where the lesson was first learned, and Tribal still surfaces it.
- [Curated, not hoarded](https://tribal.build/#how-it-works): Every ingest is checked against what is already there. Near-duplicates are merged, and items are linked by what they support, contradict, or derive from.

## What it does not do

- It is not autonomous agent memory. It does not record everything an agent does, and it does not edit itself mid-task. You decide what gets ingested.
- It is not a code index. It does not retrieve files, symbols, or call sites; it holds the reasoning behind the code, not the code itself.
- It is not a hosted service. It runs on your own Postgres.
- It is tested today with Claude Code, Codex, and OpenCode. Other MCP-capable harnesses are expected to work and are confirmed as they land.

## Where the knowledge lives

- [Make it yours](https://tribal.build/#licence): Tribal's source is public under the Elastic License 2.0. Your memories are stored in your Postgres and never sent to us. Ingest and retrieval call the model and embedding provider you configure, so the text you ingest is processed by that provider unless you run it all locally with Ollama. Tribal emits OpenTelemetry only to a collector you connect, or nowhere at all, and sends nothing to us. The setup skills are CC BY 4.0 licensed and call nothing home.

## Project

- [GitHub repository](https://github.com/tribal-memory/tribal): Source, releases, issues.
- [README](https://github.com/tribal-memory/tribal#readme): Full project documentation.
- [Latest release](https://github.com/tribal-memory/tribal/releases/latest): Current version and changelog.
- [Licence](https://github.com/tribal-memory/tribal/blob/main/LICENSE): Elastic License 2.0.

## Optional

- [Landing page](https://tribal.build): The full marketing page, with real recall transcripts and the install flow.
- [Troubleshooting](https://github.com/tribal-memory/tribal#troubleshooting): Common setup issues and recovery patterns.
- [Open an issue](https://github.com/tribal-memory/tribal/issues/new): For bug reports and feature requests.