Open Source · MIT License

Every AI session starts from zero.
Enginehaus changes that.

It's called cognitive drift — decisions evaporate, context scatters, and agents re-discover what you already decided. Enginehaus is the coordination layer that makes context compound across every session and every tool.

$ npx enginehaus init Copied!
Get Started View on GitHub
Works with Claude Desktop · Claude Code · Cursor · Windsurf · Gemini CLI · Kiro · LM Studio · any MCP client

Cognitive drift

What changes

WHAT YOU GET
01

Decisions that persist

Architecture calls, dependency choices, design tradeoffs — captured as a byproduct of working, not a chore. Next week, next month, any agent can retrieve the decision and the reasoning behind it.

02

Quality gates that enforce themselves

Work doesn't close until it's actually done. Uncommitted changes, missing tests, skipped phases — caught structurally by the system. Not by you reviewing at midnight.

03

One memory across every tool

Claude Code, Claude Desktop, Cursor, Windsurf, Gemini CLI, Kiro — they all share the same coordination layer via MCP. One init, every tool remembers.

04

Context that loads itself

Every agent session starts fully briefed — decisions, tasks, project state. No copy-paste, no "here's what we decided" docs. The coordination layer does the briefing.

05

Full decision trail

Every architectural choice, every tradeoff, every "why did we do it this way" — logged automatically with reasoning and context. Searchable. Auditable. Permanent.

A NEW WAY OF WORKING
You think through a problem with one agent. Another picks it up and builds — fully informed. No briefing doc. No re-explaining.
Act 1 The thinking
CLAUDE DESKTOP
Trevor Let's scope the beta to solo devs for now — teams add too much surface area.
Claude Agreed. Tighter feedback loop, simpler onboarding story.
Trevor Auth should be JWT, stateless — we'll want edge deployment eventually.
Claude Makes sense. I'll create a task for the auth layer.
Trevor Rate limiting belongs at the edge too, not the app.
Claude Good call — consistent with the stateless direction.
Enginehaus capturing
decision Beta scoped to solo developers simpler validation path
decision JWT auth, stateless deployment target
task ready
Implement auth layer
decision Rate limiting at edge layer consistent with stateless architecture

No brief written. No tickets filed. The context is already structured for whatever agent picks up the work next.

Three hours later...
Act 2 The building
CLAUDE CODE
> Picking up the auth layer task. Loading context from your conversation earlier...
Found 3 decisions from your planning session:
• JWT auth, stateless — edge deployment target
• Rate limiting at edge layer, not the app
Building based on your decisions...
✓ JWT middleware with stateless token validation
✓ Refresh rotation with edge constraints
✓ Rate limiting middleware at edge layer
Running quality gates...
tests 12 passed
types clean
lint clean
Done. PR opened with full decision trail.

The coding agent didn't ask "what did you decide?" — it already knew. Every decision, every constraint, loaded automatically.

That's the new workflow. You reason with one agent. Another builds — with every decision already loaded.
No handoff doc. No re-explaining. Context doesn't drift — it compounds.

WHO THIS IS FOR

Solo developers

Tired of re-explaining context to every new AI session.

Decision amnesia

Anyone who's lost a good decision because the session ended.

Multi-tool workflows

People running multiple AI tools who want them all to share context.

Technical leaders

Who need to trust what AI agents are doing without watching every step.

GET STARTED

Ship with context that compounds.

Your context stays on your machine. Your workflow stays yours. Open source. Local-first. Free.

Star on GitHub View on GitHub
$ npx enginehaus init Copied!