Built for the post-session era
Give every AI coding agent the memory your codebase deserves.
ContextKit turns your repository into a living project brain, so Claude Code, Codex, Cursor, Copilot, and OpenCode stop forgetting architecture, conventions, bugs, and hard-won decisions between sessions.
Live Project Brain
.ctx
Decisions. Patterns. Forbidden moves. Bugs. Context.
Claude
Inject
Codex
Remember
Cursor
Stay Lean
Copilot
Stay Aligned
Works with
Claude Code
Codex
Cursor
GitHub Copilot
OpenCode
Any agent that can read instructions
The friction nobody should accept anymore
Every session reset costs trust, speed, and focus.
The agent forgets your architecture. It repeats bugs you already fixed. It suggests banned patterns. It wastes your time rebuilding context you already taught it yesterday.
Re-explaining drains momentum
Your first 20 minutes go to re-onboarding instead of building.
Knowledge dies between tools
Switch from Codex to Claude or Cursor and the project memory disappears.
Past bugs come back
Without a memory layer, fixes do not become reusable defensive knowledge.
How it works
Small CLI. Massive continuity.
Initialize a brain
`ctx init` creates a `.ctx/` directory in your repo with human-readable memory files plus a fast SQLite index.
Capture what matters
Add memories manually with `ctx add`, or run `ctx watch` to extract durable lessons from git activity and agent sessions.
Inject the right context
`ctx inject codex` or `ctx inject claude` generates clean, agent-specific instruction files with only the memories that matter.
$ pip install ctxmemory
$ ctx init
✓ Initialized ContextKit in `.ctx/`
$ ctx add "We use FastAPI, not Flask." --type forbidden --tags "framework"
✓ Added memory a0b41be4...
$ ctx inject codex
✓ Injected 14 memories into AGENTS.mdA project brain, not another dashboard
Readable in git. Editable by humans. Useful to every agent.
ContextKit stores durable project knowledge where it belongs: inside the repo, in plain markdown files that evolve with your codebase and survive tool changes.
Designed for real development loops
From daily use to team-wide memory.
Start each day with the agent already oriented.
Your conventions, past bugs, and forbidden moves are already loaded before the first prompt.
Switch tools without context collapse.
Use Codex for implementation, Claude for architecture, Cursor for iteration, all backed by the same brain.
Yesterday's lessons become tomorrow's guardrails.
Bug root causes and design reasoning stop being tribal knowledge and start shaping every future session.
Live now on GitHub Pages and PyPI
Stop onboarding your agent from zero.
Install ContextKit, create your first project brain, and turn every coding session into a continuation instead of a reset.