GigaBrain本地记忆层

Gigabrain

<p align="center"> <strong>Local-first memory control plane for agents.</strong> </p>

<p align="center"> <a href="https://github.com/legendaryvibecoder/gigabrain/releases"><img src="https://img.shields.io/github/v/release/legendaryvibecoder/gigabrain?include_prereleases&style=for-the-badge" alt="GitHub release"></a> <a href="https://www.npmjs.com/package/@legendaryvibecoder/gigabrain"><img src="https://img.shields.io/npm/v/@legendaryvibecoder/gigabrain?style=for-the-badge" alt="npm version"></a> <a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-blue.svg?style=for-the-badge" alt="MIT License"></a> <a href="https://github.com/legendaryvibecoder/gigabrain/stargazers"><img src="https://img.shields.io/github/stars/legendaryvibecoder/gigabrain?style=for-the-badge" alt="GitHub Stars"></a> <img src="https://img.shields.io/badge/node-%3E%3D22-brightgreen?style=for-the-badge" alt="Node >=22"> </p>

<p align="center"> <a href="docs/configuration.md">Configuration</a> · <a href="CHANGELOG.md">Changelog</a> · <a href="CONTRIBUTING.md">Contributing</a> · <a href="SECURITY.md">Security</a> · <a href="https://github.com/legendaryvibecoder/gigabrain/discussions">Discussions</a> </p>

---

Gigabrain is the local-first Memory Passport and control plane for AI agents. It gives you one inspectable place to inventory, audit, deduplicate, and safely hand off memory across OpenClaw, Codex App/CLI, Claude Code/Desktop, Hermes-style MCP/HTTP bridges, Cursor/Windsurf, and explicit manual imports from cloud products.

The product face is Memory Passport + Auditor + Handoff Layer. The cross-memory bus is the architecture underneath: SQLite-backed recall, deterministic dedupe/audit flows, host-memory adapters, native markdown sync, world model, Obsidian surface, and an optional web console for memory operations.

Gigabrain does not replace native memories. It connects, checks, deduplicates, exports, and audits them where the host exposes local files or the user provides an explicit export. Closed cloud memories such as ChatGPT, Claude.ai, Gemini, and Microsoft Copilot are manual import/export flows only.

Highlights

• Capture — hybrid model: explicit remember intent writes durable memory, Codex checkpoints write episodic session logs
• Recall — orchestrator chooses between quick context, entity briefs, timeline briefs, and verification-oriented recall automatically
• MCP tools — gigabrain_recall, gigabrain_remember, gigabrain_checkpoint, gigabrain_provenance, gigabrain_recent, gigabrain_sources, gigabrain_sync_status, gigabrain_export_brief, gigabrain_doctor, gigabrain_entity, gigabrain_contradictions, gigabrain_relationships
• Memory Passport — static Markdown/HTML source inventory, readiness verdict, dedupe audit, contradiction audit, stale-memory audit, provenance gaps, secret-risk flags, and safe handoff briefs that omit secret-risk rows entirely
• Host memory bus — read-only local sync for Codex, Claude Code, Cursor/Windsurf, OpenClaw native memory, and bridge-friendly Hermes integrations
• Dedupe — exact + hybrid semantic deduplication with type-aware thresholds
• Native sync — indexes MEMORY.md and daily notes alongside the registry for unified recall
• World model — entities, beliefs, episodes, open loops, contradictions, and syntheses
• Obsidian surface — structured vault with native files, entity pages, briefings, and views
• Quality gate — junk filters, plausibility heuristics, and LLM second opinion
• Audit — nightly maintenance with snapshots, archive reports, and quality scoring
• Eval — in-process recall eval harness, nightly quality history, and latency summaries
• Web console — optional FastAPI dashboard for browsing and managing memories
• Person service — entity mention tracking for person-aware retrieval ordering