GitHub - dunova/ContextGO: Local-first context & memory runtime for multi-agent AI coding teams. MCP-free. Rust/Go accelerated.

Local-First Context & Memory Runtime for Multi-Agent AI Coding Teams
Unified cross-agent memory, hybrid BM25/vector recall, and end-to-end encrypted multi-machine sync.

Key Features • Quick Start • Supported Agents • CLI Reference • Encrypted Sync • Smart Recall • 简体中文

What is ContextGO?

ContextGO is a high-performance, local-first context and memory runtime engineered for modern AI coding workflows. It breaks down memory silos across tools, editors, and machines by indexing conversations, technical decisions, and execution sessions from 15+ AI coding agents into an ultra-fast, local SQLite runtime.

No cloud lock-in, no remote database dependencies, and no silent telemetry uploads. Your project history and architecture memories remain private, durable, and instantly searchable on your machines.

┌──────────────────────────────────────────────────────────────────────────────┐
│                            AI Coding Environments                            │
│  DeepSeek (dsh) │ Reasonix │ Hermes │ Claude Code │ Antigravity │ Copilot ...│
└───────┬─────────────┬───────────┬───────────┬─────────────┬───────────┬──────┘
        │             │           │           │             │           │
        ▼             ▼           ▼           ▼             ▼           ▼
┌──────────────────────────────────────────────────────────────────────────────┐
│                       ContextGO Source Adapters Layer                        │
│  (Streaming zstd decompression, high-SNR event filtering, projection cache)  │
└──────────────────────────────────────┬───────────────────────────────────────┘
                                       │
                                       ▼
┌──────────────────────────────────────────────────────────────────────────────┐
│                         ContextGO Local SQLite Engine                        │
│   • FTS5 Full-Text Search        • BM25S Lexical Ranking                     │
│   • Durable Memory Store         • Optional Model2Vec Embeddings + RRF       │
│   • Time-Decay Scoring           • Low-Latency Local Web Viewer UI           │
└──────────────────────────────────────┬───────────────────────────────────────┘
                                       │
                                       ▼ (Optional Opt-In)
┌──────────────────────────────────────────────────────────────────────────────┐
│                Privacy-First Encrypted GitHub Sync (AES-256-GCM)              │
│       Windows 11 (Task Scheduler) ◄───► macOS (launchd) ◄───► Linux (systemd) │
└──────────────────────────────────────────────────────────────────────────────┘

Key Features

🧠 Universal Multi-Agent Memory: Natively discovers, parses, and harmonizes sessions from DeepSeek Agent (dsh) (with .zstd streaming event decoding), Reasonix, Hermes, Claude Code, Factory Droid, Antigravity (Gemini), GitHub Copilot, Cursor, Windsurf, Kilo, Cline, Roo Code, OpenCode, and shell histories.
🔌 Native Model Context Protocol (MCP) Server: Built-in zero-dependency MCP stdio server (contextgo mcp) supporting standard tools/list and tools/call, enabling out-of-the-box Function Calling in DeepSeek Agent, Claude Code, Cursor, Windsurf, and any MCP client.
⚡ Sub-Second Hybrid Recall: Combines SQLite FTS5 full-text indexing, BM25 lexical ranking, and optional lightweight Model2Vec embeddings with Reciprocal Rank Fusion (RRF) and time-decay prioritization.
🛡️ Zero-Knowledge Encrypted Sync: Synchronize context across Windows, macOS, and Linux using a private GitHub repository with client-side scrypt key derivation and AES-256-GCM encryption.
🎯 Smart Context-First (SCF) Policy: Automatically configures prompt files across all AI tools (contextgo setup) to enforce proactive context retrieval before answering complex architectural or refactoring questions.
💻 Cross-Platform Daemon & Native Services: Native user daemon with automatic restart on Windows (Task Scheduler), macOS (launchd), and Linux (systemd user unit).
🔍 Interactive Memory Viewer: Built-in, zero-dependency local web dashboard (127.0.0.1:37677) for browsing memories, session timelines, and technical decisions.

Quick Start

1. Installation

Install ContextGO globally using pipx to keep your environment isolated:

# Standard installation with lexical hybrid recall
pipx install "contextgo[vector]"

# Include encrypted multi-machine sync dependencies
pipx install "contextgo[sync,vector]"

2. Shell Integration

Add instant aliases (cg for quick recall, cgs for full-text search, cgse for semantic recall):

eval "$(contextgo shell-init)"

Add the line above to your ~/.zshrc, ~/.bashrc, or config file.

3. Verify Health & Detected Sources

# Check runtime health and active index statistics
contextgo health

# View all detected AI coding tools on your machine
contextgo sources

4. Search & Recall Context

# Instant hybrid recall (auto-routes to keyword search or session ID)
contextgo q "how did we resolve the router DNS latency?"

# Full-text exact search across all tool histories
contextgo search "AdGuard Home" --limit 5

# Memory-first semantic recall
contextgo semantic "architecture decisions for sync engine" --limit 3

Supported AI Agents & Tools

ContextGO automatically indexes and standardizes sessions across your environment without manual configuration:

Category	Supported Tools & Platforms	Discovery Mechanism
Autonomous Agents	DeepSeek Agent (`dsh`)	`.dsh/storages/session_projcache.json`, `sessions/*/.zstd`
	Reasonix Agent	`.reasonix/projects/*/sessions`, `events.jsonl`, `turns`
	Hermes Agent	`~/.hermes/sessions/*.jsonl`, sidecar metadata
	Claude Code	`~/.claude/projects/`, `~/.claude/transcripts/`
	Factory Droid	`~/.factory/sessions/*.jsonl`
	OpenClaw & Accio	`~/.openclaw/agents/`, `~/.accio/agents/`
IDEs & Coding Assistants	Antigravity (Gemini)	`~/.gemini/antigravity/brain/*/walkthrough.md`, `logs`
	GitHub Copilot	`~/.copilot/session-state/*/events.jsonl`
	Cursor & Windsurf	Global storage workspaces & vscdb SQLite stores
	Kilo, Cline & Roo Code	VS Code globalStorage task state & transcripts
	OpenCode & Zed	`opencode.db`, `.config/zed/conversations/`
Shell & Manual	Terminal Shells	`~/.zsh_history`, `~/.bash_history`
	Durable Memories	`contextgo save`, exported JSON observation snapshots

Core CLI Reference

usage: contextgo [-h] [--version] <command> ...

Command	Usage Example	Description
`q`	`contextgo q "query"`	Fast hybrid recall: Auto-routes query to BM25S/FTS or session ID lookup.
`search`	`contextgo search "keyword" --limit 10`	Full-text search: Query indexed session logs and tool calls.
`semantic`	`contextgo semantic "topic" --limit 5`	Semantic search: Memory-first retrieval with session fallback.
`save`	`contextgo save --title "..." --content "..."`	Save durable memory: Persist key architectural conclusions to local storage.
`sources`	`contextgo sources`	Inspect adapters: Print detected platforms, session counts, and paths.
`health`	`contextgo health`	Health check: Output JSON status report of DBs and subsystems.
`serve`	`contextgo serve --port 37677`	Web UI: Launch local zero-dependency memory viewer dashboard.
`sync`	`contextgo sync {init,push,pull,status,run}`	Encrypted sync: Manage cross-machine GitHub repository synchronization.
`setup`	`contextgo setup`	One-command setup: Inject Smart Recall (SCF) rules into all AI tools.
`unsetup`	`contextgo unsetup`	Teardown: Safely remove all injected ContextGO prompt rules.
`daemon`	`contextgo daemon {start,stop,status,install}`	Daemon management: Control background capture and OS service unit.
`export`	`contextgo export "" backup.json`	Sanitized export: Export observations with automatic secret redaction.
`import`	`contextgo import backup.json`	Import: Restore observation payloads into local memory store.
`smoke`	`contextgo smoke --sandbox`	Quality gate: Run end-to-end verification in an isolated sandbox.

Encrypted Multi-Machine Sync

ContextGO provides private, zero-knowledge multi-machine synchronization backed by any private GitHub repository:

# 1. Initialize sync on your primary machine (e.g. Windows 11)
contextgo sync init --repo your-user/my-contextgo-sync --device-id desktop-win11

# 2. Push encrypted shards
contextgo sync push

# 3. Pull and merge on your secondary machine (e.g. macOS)
pipx install "contextgo[sync,vector]"
contextgo sync init --repo your-user/my-contextgo-sync --device-id macbook-m4
contextgo sync pull

Security Architecture

Client-Side AES-256-GCM: Payloads are compressed and encrypted locally before leaving your machine.
Passphrase-Derived Key: Key derived with scrypt using per-repository salt. Your password never leaves your device.
Independent Device Shards: Each machine commits to its own encrypted partition (shards/<device-id>.enc.json) to eliminate merge conflicts.
Pre-Upload Sanitization: Absolute paths, API keys, tokens, and secret patterns are automatically redacted before encryption.

AI Agent Smart Recall (SCF)

Configure all your AI assistants to read project context automatically:

contextgo setup

This injects the Smart Context-First (SCF) policy into active agent instruction files (e.g., ~/.gemini/GEMINI.md, ~/.reasonix/AGENTS.md, ~/.dsh/AGENTS.md, ~/.hermes/SOUL.md, ~/.claude/CLAUDE.md, .github/copilot-instructions.md, .cursorrules).

Proactive Recall Policy Table

Trigger	Agent Action
Continuation task (`接着做` / `continue`)	Run `contextgo semantic "<topic>" --limit 3`
Uncertainty regarding past decisions	Run `contextgo search "<keyword>" --limit 5`
Before major architectural refactoring	Query previous design decisions before altering code
Root cause confirmed	Save durable memory via `contextgo save --title "..."`

Daemon & Background Services

Run ContextGO as a native, background-indexed service:

# Manage daemon state
contextgo daemon start
contextgo daemon status
contextgo daemon stop

# Install native OS service unit
contextgo daemon install

Operating System	Service Backend
macOS	Native user `launchd` service (`~/Library/LaunchAgents/`)
Linux	Native `systemd` user service (`~/.config/systemd/user/`)
Windows	Native Windows Task Scheduler user task

Development & Verification

ContextGO maintains rigorous release gates with 86%+ test coverage and extensive platform matrices:

# Clone repository
git clone https://github.com/dunova/ContextGO.git
cd ContextGO

# Sync dependencies with uv
uv sync --extra dev --extra sync --extra vector

# Run code style & security gates
uv run ruff check src/contextgo tests
uv run ruff format --check src/contextgo tests
uv run mypy src/contextgo --ignore-missing-imports
uv run bandit -r src/contextgo -c pyproject.toml --quiet

# Run test suite
uv run pytest
uv run python -m contextgo smoke --sandbox

License

ContextGO is licensed under the AGPL-3.0-only license.

Name		Name	Last commit message	Last commit date
.github		.github
docs		docs
native		native
packages/dsh-plugin-contextgo		packages/dsh-plugin-contextgo
scripts		scripts
src/contextgo		src/contextgo
tests		tests
.cursorrules		.cursorrules
.gitignore		.gitignore
AGENTS.md		AGENTS.md
LICENSE		LICENSE
Makefile		Makefile
README.md		README.md
README.zh.md		README.zh.md
VERSION		VERSION
pyproject.toml		pyproject.toml
uv.lock		uv.lock

Navigation Menu

Repository files navigation

What is ContextGO?

Key Features

Quick Start

1. Installation

2. Shell Integration

3. Verify Health & Detected Sources

4. Search & Recall Context

Supported AI Agents & Tools

Core CLI Reference

Encrypted Multi-Machine Sync

Security Architecture

AI Agent Smart Recall (SCF)

Proactive Recall Policy Table

Daemon & Background Services

Development & Verification

License

About

Releases

Packages

Contributors

Languages

Latest commit

History

Folders and files

Repository files navigation

What is ContextGO?

Key Features

Quick Start

1. Installation

2. Shell Integration

3. Verify Health & Detected Sources

4. Search & Recall Context

Supported AI Agents & Tools

Core CLI Reference

Encrypted Multi-Machine Sync

Security Architecture

AI Agent Smart Recall (SCF)

Proactive Recall Policy Table

Daemon & Background Services

Development & Verification

License

About

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages