Gigabrain

Gigabrain is the long-term memory layer for OpenClaw agents. It converts conversations and native notes into durable, queryable memory, then injects the right context before each prompt so agents stay consistent across sessions.

It is built for local-first production use: SQLite-backed recall, deterministic dedupe/audit flows, native markdown sync, and an optional FastAPI console for memory operations.

Release references:

• Changelog: CHANGELOG.md
• v0.5.0 release notes: release-notes/v0.5.0.md

What it does

• Capture: Uses a hybrid memory model where explicit remember intent can write concise native markdown and structured registry memories together
• Recall: Before each prompt, the recall orchestrator chooses between quick context, entity briefs, timeline briefs, and verification-oriented recall
• Dedupe: Exact and hybrid semantic deduplication catches duplicates, paraphrases, and malformed near-duplicates with type-aware thresholds
• Native sync: Indexes your workspace MEMORY.md and daily notes alongside the registry for unified recall
• World model: Projects atomic memories into entities, beliefs, episodes, open loops, contradictions, and syntheses that can power better recall and review
• Obsidian surface: Builds a structured vault with native files, active memory nodes, entity pages, briefings, reports, and views so your agent can expose memory clearly in Obsidian and sync it to another machine
• Person service: Tracks entity mentions across memories for person-aware retrieval ordering
• Quality gate: Junk filters, durable-personal retention bias, plausibility heuristics, and optional LLM second opinion keep memory clean without losing important relationship context
• Audit: Nightly maintenance with snapshots, execution artifacts, archive reports, review queue retention, and quality scoring
• Web console: Optional FastAPI dashboard for browsing, editing, and managing memories

Prerequisites

• Node.js >= 22.x (uses node:sqlite experimental API)
• OpenClaw >= 2026.2.15 (gateway + plugin loader)
• Python >= 3.10 (only for the optional web console)
• Ollama (optional, for local LLM-based extraction review and semantic search)
• Obsidian (recommended for the v0.5 memory surface; core capture/recall still works without it)

Installation

Option A: npm install + setup wizard (recommended)

Install:

mkdir -p ~/.openclaw/plugins
cd ~/.openclaw/plugins
npm install @legendaryvibecoder/gigabrain
cd node_modules/@legendaryvibecoder/gigabrain

Run the one-command setup wizard:

npm run setup -- --workspace /path/to/your-openclaw-workspace

What the setup wizard does:

• Ensures plugins.entries.gigabrain exists in ~/.openclaw/openclaw.json
• Sets plugin path and runtime paths (workspaceRoot, memoryRoot, outputDir, registryPath)
• Enables the v0.5 hybr...