Now available for Claude Code

Your AI forgets.
ctx doesn't.

Project memory that makes every AI coding session as informed as your last one. No more re-explaining. No more wrong assumptions. No more lost context.

Set up in 30 seconds

Every session starts with amnesia

Your AI coding agent is brilliant — but it doesn't know what you built yesterday, why you chose Supabase over custom auth, or that the middleware approach already broke CORS. So you waste the first 10 minutes of every session re-explaining your project.

“I already told you this last session”

You explained a key decision in a previous session. Next session, the agent has no memory of it. You explain it again. And again.

ctx solves it

Every decision is captured automatically. Next session, the agent already knows.

“It went in completely the wrong direction”

You say “fix the auth” and the agent spends 15 minutes building something using a pattern your codebase doesn’t follow.

ctx solves it

The agent receives your conventions and known pitfalls before it writes a single line.

“It keeps making the same mistake”

Last week, the agent tried an approach that broke something. This week, a different session tries the exact same broken approach.

ctx solves it

Failures are stored and served as warnings. Your AI never repeats the same mistake.

“None of my AI tools talk to each other”

You brainstormed in ChatGPT, coded in Claude Code, and debugged in Cursor. None of them know what the others discussed.

ctx solves it

One project brain for all your tools. Claude knows what you decided in Cursor.

“My CLAUDE.md is 300 lines and half of it is wrong”

You tried the manual approach. It worked for a week. Now it’s bloated with outdated instructions and contradicts itself.

ctx solves it

Context is captured automatically, old info decays, and the agent only sees what’s relevant.

See what your agent sees

At the start of every session, your agent receives this — automatically

ctx context
# Project Context: openalmanac
Branch: main
## Conventions
- Use TypeScript strict mode everywhere
- pytest with minimal mocking for tests
- Conventional commits required
## Decisions
- Chose Supabase over custom JWT for auth
- Research API uses bearer tokens with RLS
## Known Pitfalls
! CORS headers stripped if auth middleware runs first — keep CORS above auth
! Token refresh race condition on parallel requests — use mutex lock
## Recent Activity
- Fixed CORS ordering, added RLS policies
- Added API rate limiting to /api/v1 routes

Three steps. Zero config.

ctx hooks into your existing workflow. Nothing changes except your agent gets smarter.

1

Init

Run ctx init in your project. It reads your git history and learns your codebase conventions automatically.

2

Work normally

Use any AI tool — Claude Code, Cursor, whatever. ctx captures decisions, conventions, and failures in the background. You don’t change how you work.

3

Every session is informed

Next time you start a session, your agent already knows your project. No re-explaining. It just works.

Built for developers who ship

Auto-capture

Learns from every AI session without you lifting a finger. Decisions, conventions, and failures are extracted automatically.

Cross-tool memory

Works with Claude Code today, Cursor and any MCP-compatible tool coming soon. Your memory isn’t locked to one provider.

Smart retrieval

Only serves context relevant to your current task, not everything. Your agent gets signal, not noise.

Self-cleaning

Old context gets summarized. Contradictions auto-resolve. The brain grows smarter, not bigger.

Fully local

Everything lives in a SQLite database on your machine. No cloud, no accounts, no data leaving your laptop. Ever.

One command install

brew install ctx — no runtime, no dependencies, no configuration files to write. Set up in 30 seconds.

You've tried workarounds

ctx isn't a prompt optimizer or a RAG pipeline. It's project memory — the accumulated understanding of why your code is the way it is.

CLAUDE.md / .cursorrulesManual, gets stale, single-tool only
Model memory (Claude, GPT)Provider-locked, session-scoped, no cross-tool
Prompt templatesStill requires human effort, doesn’t learn
RAG over codebaseAnswers “what does this code do” not “why was it built this way”
ctxAutomatic, cross-tool, self-maintaining, serves decisions not just code

Simple CLI, powerful brain

ctx initSet up a project brain in 10 seconds
ctx statusDashboard of what the brain knows
ctx addManually add decisions, conventions, patterns
ctx logTimeline of all captured context
ctx compactSummarize old context to stay lean
ctx contextOutput project context for any AI agent
ctx hooksInstall or update AI tool integrations

Questions

Give your agent a memory

One command. Fully local. Free forever.

Stop re-explaining your architecture. Stop cleaning up wrong assumptions.Just install ctx and let your agent remember.

$ brew install divitsheth/tap/ctx