Cratonis
Features

What the substrate actually does

Every claim on this page describes the shipped mechanism, not a roadmap. Where something is scheduled rather than shipped, it says so.

compiler

Eight typed sources

Context records, application identity, technology stack, active modules, prompt history, code-modification history, custom instructions, and external schema snapshots — each compiled by typed SQL in a defined sequence, with hard byte budgets so one source can never crowd out the rest.

records

Scoped, prioritised context

Records carry a type, a priority and a visibility scope. Compilation filters by scope and sorts by priority — a critical decision outranks a low-priority note deterministically, not by embedding distance.

channel one

Hosted MCP endpoint

Agents connect over Streamable HTTP with a bearer credential and call cf_bootstrap for the compiled document, plus focused read tools for search, pages and history.

channel two

Seven instruction files

The same compilation is spliced into the files your tools already read: CLAUDE.md, AGENTS.md, GEMINI.md, and the Cursor, Copilot, Windsurf and Devin conventions — inside managed markers that never touch your own notes, stamped with the artifact's digest so staleness is detectable.

write-back

AI authorship, recorded

Agents persist what they learn through a validated tool that forces origin to ai and scope to shared — an agent cannot label its own guess as a human decision, and every later session can tell the two apart.

governed writes

A gated write path

File writes from agents pass seven ordered gates — credential, scope, path boundary, extension allow-list, content scan, size budgets — land in staging with an append-only ledger entry, and stay staged until a human promotes them.

access

Role-based access control

Roles from super-admin to read-only, permission checks on every panel action, and per-user denial overrides that beat any granting role — with a super-admin that no denial can lock out of the room.

isolation

Your database, your context

One application and one Postgres database per tenant. Your context lives in your application's own schema — the SaaS layer holds subscriptions and plans, never a copy of your memory. Delete the platform; your app keeps working.

scheduled

Modules tier

Agent-built modules with their own schema, previews, and signed content-addressed promotion — the Pro-and-up tier on the roadmap's next phase. Sold as coming, because it is.

The part we refuse to build

No vector database. On purpose.

Similarity search gives you the context that looks like your question. Compilation gives you the context your team decided matters, in the order they ranked it, identically for every agent. That determinism is the patent-pending core of the product — and it is also why there is no embedding bill, no index to rebuild, and no retrieval lottery to debug.