Skip to main content

API Reference

Runtime API

The hot path — the four endpoints every integration calls before and after the LLM, plus the bundle-fetch endpoint runtimes poll with ETags.

GET/v1/bundles/latest

Fetch the active bundle for the runtime token's scope

POST/v1/evaluate/agent-plan

Evaluate a proposed agent plan for excessive agency (LLM03)

POST/v1/evaluate/agent-step

Evaluate a single agent-loop iteration for budget/depth caps (LLM06)

POST/v1/evaluate/input

Evaluate user input before calling the LLM

POST/v1/evaluate/memory-write

Evaluate text written to agent memory for persistent injection (LLM01)

POST/v1/evaluate/output

Evaluate an LLM response before returning it to the user

POST/v1/evaluate/retrieval

Evaluate retrieved RAG chunks for indirect prompt injection (LLM01)

POST/v1/evaluate/stream

Evaluate streaming LLM output as it arrives

POST/v1/evaluate/tool

Evaluate a tool / function invocation