Canis — Workspace Engine

Canis gives your B2B product AI-built, contract-validated workspaces: your customers describe the screen they need, the model proposes a workspace as data, a validator gates it against your declared data contract, and a deterministic renderer turns it into a live screen — with zero LLM involvement at read time.


prompt ──> LLM ──> WorkspaceSpec (JSON) ──> validateSpec ──> BUILD? ──> renderer
                                              │ your contract │
                                              └── REJECT / CLARIFY (never renders)

Three properties make this safe to ship:

Everything is data. A workspace is a JSON document (Spec v1) — no code, no JSX, no expressions. Every capability it can use is one you declared.
Validation is total. Nothing reaches the renderer, or the Workspace Service’s storage, without a BUILD verdict from the same validator.
Your data stays yours. The SDK calls your fetch with your end user’s auth token. Customer rows never rest with us.

Where to go

Quickstart — a working, validated workspace in your app in under 10 minutes (stopwatch-timed, and the code is compiled in our CI).
Vendor CI guide — ticora contracts diff blocks contract changes that would break your customers’ saved workspaces.
Spec v1 reference — the frozen workspace format.
Error taxonomy — every verdict, error code, and exit code, and what to do about each.
API reference — the public surface of each package.