TricorderKit — the local-first second brain for your AI agents

Give your AI agents a second brain — one that can't leak.

TricorderKit is an open-source, local-first knowledge OS for AI agents. Every intention becomes a traceable workflow, every claim is measured, and every push goes through gates that stop secrets and private names before they leave your machine.

★ Star on GitHub Install in one line
100 % vs 33–50 %multi-hop answers found — GraphRAG vs flat RAG, same budgetreproduce
−84 % tokensto answer "what was true at T" — temporal memory, 100 % exactreproduce
−82 % comparisonsto find duplicates, same F1 as exhaustive fuzzy matchingreproduce
4 gateson every push, locally and in CI — secrets, private names, docs, contractssee CI

Every figure comes from a selftest shipped in the repository (re-run on 28 October 2026). No inflated metrics.

See the guardrails work

A private name slips into a note. The pre-push hook blocks it before anything leaves the machine — and because CI logs are public, it reports a digest, never the name. Real output, recorded with the repository's own hook.

Real output: the pre-push gate blocks a private name, then lets the neutral version through

What you get

13 Claude Code plugins

Hybrid RAG (graphify), deep research, workflows on Temporal, memory boot, evaluation lab, security audit — one marketplace.

21 method skills

Model routing and budgets, context compression, agent-config audit, skill evals, development protocol — no data, no Docker.

Local-first

Ollama, Neo4j and Qdrant on your machine — or on a server you own. Cloud models are optional, never required.

Measured

Offline benchmarks with selftests, 579 tests in CI, live badges instead of static claims.

Self-improving, safely

The learning loop only writes proposals. Promotion needs green tests and a human review.

Gated

gitleaks, a private-name boundary gate, docs-sync and plugin contracts — pre-push hook and CI.

How it fits together

Architecture: intention, MainBrain, skills, memory, workflows, self-improving loop, eval-lab, hybrid RAG, research, knowledge base, behind quality gates

Field notes from a real project

Not a demo. TricorderKit runs every day behind a private knowledge-base project. The public engine stays generic; the domain lives in a private linked repository.

The anti-leak gate comes from a real incident. In October 2026, content from that private project slipped into the public repository through a whitelist that exempted whole folders. We fixed the cause, neutralised the engine, and restarted the public history from a clean commit.

Gates find bugs in the gates. Recording the demo above revealed that the pre-push hook was not executable on Linux and macOS. Fixed in the same release.

Install in one line

Inside Claude Code:

/plugin marketplace add GeekFamilyCorp/TricorderKit

Then pick a plugin, for example /plugin install graphify@tricorderkit. The full local engine is one git clone away — see the quick start.

★ Star on GitHub Join the discussion