# Lifeline

> A context runtime for AI-assisted development — "git for reasoning". An append-only,
> content-addressed, anchored ledger of a project's WHY, delivered to any AI over MCP so a
> fresh model connects and already knows what / why / decided / next. Time-to-Context -> 0.

Key facts: open source (MIT); install `pip install lifeline-context`; Python 3.10+; MCP-native;
writes are human-in-the-loop (AI proposes, human approves); every context item is anchored to
the cryptographic hash of its source event (anti-hallucination); append-only with supersession.

## Docs
- [Getting started](https://lifelinecontext.com/docs/getting-started.html): Install Lifeline, run the loop, and bootstrap an existing project. Git for reasoning, in minutes.
- [Concepts & laws](https://lifelinecontext.com/docs/concepts.html): The Entry, content-addressing, the 3 memory layers, supersession, lines as separate reasoning threads (Tree-of-Thoughts), and the 7 laws.
- [Architecture](https://lifelinecontext.com/docs/architecture.html): Lifeline's architecture: the deterministic event model, reducers, recall, projection, and the store-is-source cutover.
- [Integration](https://lifelinecontext.com/docs/integration.html): Wire Lifeline into Claude Code, Cursor, Claude Desktop, and Gemini CLI via MCP — with auto-connect.
- [MCP & remote](https://lifelinecontext.com/docs/mcp.html): The MCP surface: local stdio, remote HTTP/SSE, the OAuth Resource Server, and multi-tenant cloud.
- [CLI reference](https://lifelinecontext.com/docs/cli.html): Every Lifeline command: log, propose/review/approve, context, verify, migrate, init, push/pull/clone.
- [vs alternatives](https://lifelinecontext.com/docs/compare.html): Lifeline vs ADRs, CLAUDE.md, RAG, and memory MCPs (mem0, Letta, Honcho) — honest differences, and when not to use it.
- [Curate well](https://lifelinecontext.com/docs/curate.html): How to keep a Lifeline ledger signal not noise: good entries, granularity, supersession, and avoiding ledger rot.
- [Teams & hosted](https://lifelinecontext.com/docs/teams.html): The hosted Lifeline hub: zero-ops MCP, shared team lines, the GitHub App, and Marketplace billing — in early access.

## Articles
- [How one AI inherits another's context: solving the handoff problem](https://lifelinecontext.com/blog/ai-context-handoff.html): Two AIs on one project share nothing. Here's why the context handoff breaks — and how an anchored ledger over MCP makes context portable between models.
- [Context engineering: the discipline beyond prompt engineering](https://lifelinecontext.com/blog/context-engineering.html): Prompt engineering shapes one request. Context engineering shapes what the model knows across every request — the discipline, and why it lasts.
- [Give Hermes Agent a memory it can trust: Lifeline over MCP](https://lifelinecontext.com/blog/hermes-agent-lifeline.html): Hermes Agent recalls what happened; it doesn't anchor why. Here's how Lifeline plugs in over MCP to give a self-improving agent a verifiable why layer.
- [Share of context: the metric behind Lifeline](https://lifelinecontext.com/blog/share-of-context.html): Share of context is the portion of an AI's reasoning that comes from your curated, anchored record — the AI-era metric Lifeline is built to move.
- [Time-to-context: the hidden tax on AI-assisted development](https://lifelinecontext.com/blog/time-to-context.html): Time-to-context is how long an AI needs before it can act correctly on your project — the real bottleneck in AI-assisted work, and how to drive it to zero.

## Source
- [GitHub repository](https://github.com/lifeline-context/lifeline)
- [PyPI package: lifeline-context](https://pypi.org/project/lifeline-context/)
- [Full text for LLMs](https://lifelinecontext.com/llms-full.txt)