A shared knowledge layer for agents

One source of truth, every agent.

Every agent gets your project's decisions. Engrym tells you when the code stops honoring them.

Start freeSee how it works

Free for 3 projects · No card · Your source code never leaves your machine

Project

engrym

Decisions · 47

  • DecisionPostgres over MySQLhigh
  • ConstraintAPI responses are envelope-wrappedhigh
  • ConventionRaw SQL only, no ORMhigh
  • ContextPrisma migrations caused two Q2 incidentsmedium

Works with

  • Claude Code
  • Cursor
  • Codex
  • ChatGPT
  • Gemini
  • Windsurf

Switch agents. Keep the context.

Step 01

Bring your Markdown.

Import an Obsidian vault or a folder of docs. Engrym reads it and pulls out the decisions, conventions, and constraints, on your own AI key.

Read the quickstart →

Importing

4 of 6 files
  • architecture.md12 atoms
  • api-conventions.md8 atoms
  • auth-decisions.md6 atoms
  • data-model.mdreading…
  • deploy.md
  • testing.md

Step 02

Every agent starts informed.

Claude Code, Cursor, Codex, Gemini: each one queries Engrym at session start and gets only what applies to the file it's touching.

claude · engrym
$ claude
⟢ engrym: connected to project "engrym"
✓ 3 decisions   apply to packages/api/**
✓ 2 constraints apply to packages/api/**
✓ 1 convention  applies to packages/api/**
2,721 tokens · ready

Step 03

Docs move. Knowledge follows.

Change a document and Engrym re-reads the affected sections. You approve what’s new. Every agent has it on the next call.

auth-decisions.mdedited 2m ago

3 changes detected

  • Token TTL is now 15 minutes
  • Session store decision superseded
  • Legacy cookie note
Approve allReview

Step 04

Then the code drifts. And you find out.

Agents check the implementation on your machine and send back a verdict. Engrym stores the verdict, never your source.

Claim

Sessions are stateless JWTs. No server-side session store.

Status

checking…

Found at

See it

Prose in. Structure out.

Pick an example. Watch what your agents would actually receive.

Example documents. Extraction runs on your own AI provider key.

What you get

Yours, not ours.

Lighter context

16.5×

less context, same answer

The document your agent used to read~0 tokens
What Engrym actually sends it0 tokens

Measured on a live query against our own project.

API Key or subscription

Document extraction and knowledge atomization is unlimited, it runs on your own API key, or on the subscription you already pay for.

Source stays local

Agents check the code on your machine. Engrym keeps the verdict, never the file.

24 MCP tools

Plus a CLI, a REST API, and a TypeScript SDK.

Mirrored to your git

Structured commits to a remote you control. Leave whenever you want.

Three projects, free.

No card. No trial clock. Unlimited extraction on every tier.

Start free

On the way

Two things we're building next.

In development

Engrym, local and open source

The same project-scoped answers, running entirely from a .engrym/ folder on your machine. No account, no server, no network call.

Private beta

Engrym for teams

Intent broadcasting, file-overlap conflict detection, a shared knowledge pool, and cross-member agent sync. €49.99 per seat when it opens.

Tell every agent what must stay true.

Then find out when it doesn't.

Start free

Three projects free. No card.