Six system types.
One engineering standard.
Each type is a different problem shape. Start from the symptom you recognise, it routes you to the system type, the proof, and the case file behind it. Every build ships eval-gated, observable, and owned by you.
Different problem, same spine.
Whatever the system type, the architecture is the same five stages, input passes a typed MCP boundary, specialist agents do the work, and nothing reaches you until it clears the eval gate.
The MCP boundary and eval gate are structural, not policy, an agent can't reach a live system without the typed gateway, and a build can't ship if its eval score drops more than 5pp.
System types
Six standards, whatever
the system type.
These aren't optional extras, they're the engineering baseline under every system we ship.
Eval harness first
100+ scenario stories written and passing before any production deploy. Verified, not just tested.
CI regression gate
Every PR runs the eval. Score drops more than 5pp and the build is blocked. No silent decay.
MCP system boundary
No agent touches a live system without a typed FastMCP gateway. A structural boundary, not a policy.
Full observability
Langfuse traces on every run, Prometheus metrics, inputs, outputs, latency and cost, from day one.
You own everything
100% of the code, infrastructure and data. No licence fees, no hosted dependency. Forkable on day one.
Read-only by default
Systems read; writes need explicit human approval or a named, reviewed exception. Enforced in code.
Not sure which
one fits?
Describe what you want to automate. We'll tell you which system type fits, what the build involves, and what the eval harness looks like. 30 minutes, no commitment.