context-bridge — Universal Session Memory for AI Agents

Capture. Index. Retrieve. A local-first context layer that works with any AI coding agent.

---

The Problem

Every AI coding agent (Claude Code, Codex, OpenCode, Hermes, Aider) forgets everything between sessions. You re-explain your project, re-state architecture decisions, and re-teach preferences every single time. Existing solutions are locked to one agent or require manual maintenance.

The Solution

context-bridge is a universal, local-first memory layer that:

• Captures sessions from any agent automatically
• Indexes decisions, patterns, lessons in a local SQLite database
• Retrieves relevant context when you start a new session
• Works with Claude Code, Codex, OpenCode, Hermes, Aider — anything that writes logs

No cloud. No API keys. No vendor lock-in.

Quick Start

pip install context-bridge

# Initialize in your project
context-bridge init

# Index existing agent sessions
context-bridge index ~/.claude/projects/

# Query memory
context-bridge recall "How did we handle authentication?"

# Inject context into new session
context-bridge inject --agent claude

Architecture

~/.context-bridge/
├── sessions/           # Captured session transcripts
├── index.db            # SQLite FTS + metadata
├── memories/           # Consolidated memories
└── config.yaml         # Per-project config

Why Now?

• 67k+ MCP servers deployed, AI agents everywhere
• No open-source universal memory layer exists
• Show HN validation: hmem, Engram, Memobase all trending
• Every developer using AI agents has this problem

License

MIT