Auditable answers
over operational
APIs.
Prodis is an open-source runtime that turns your Django, FastAPI, or Spring APIs into assistants for factual questions. The runtime plans, executes, and validates answers in code before responding while at the same time you are able audit answer derivation at the most granular level.
Premise.
The demo passes. Production is where it breaks.
Every production AI eventually fails the same way. A confidently wrong answer reaches a user¹. An action runs that should have been blocked². A decision is made from a black-box answer the team cannot explain after the fact³. When evidence is missing, Prodis refuses with a recorded reason⁴. Polished language is not correctness, and the model is not the authority for that distinction.
Method.
The model proposes. The backend executes.
Break question-answering into granular, schema-validated steps, and a non-deterministic task becomes an almost-deterministic one. The model interprets language at the seams; bounded, typed code owns the rest. The answer record shows the explicit inputs from the user, derived inputs from resolvers, implicit inputs from runtime context, and each step used to reach the final answer. Synthesis happens only after evidence is sufficient — and the system can refuse.
Stack.
Today's agent stacks run the loop. They do not own execution.
OpenAI SDK, Anthropic SDK, LangChain, LangGraph, MCP, direct tool calls, in-house loops — useful tools, incomplete execution boundary. They call models, route tools, stream responses, and wire workflows. They do not turn intent into a validated backend execution path over business APIs. In the usual loop, the model still decides what was meant, what to call, whether the result is enough, and how to repair the answer. Prodis moves those decisions into staged contracts and backend code.
- 01user asks
- 02model picks tools
- 03api returns data
- 04model decides if enough
- 05model writes answer
- 01model interprets intent
- 02typed plan validated
- 03backend APIs execute
- 04result derived in code
- 05evidence readiness checked
Prodis owns the execution boundary, not just the final check. Intent, source selection, permissions, evidence, derivation, readiness, and synthesis each have a backend-owned contract.
Runtime.
Install an adapter. Keep your API where it is.
Prodis ships the engine, contracts, local runtime, adapters, eval runner, and examples as open source. Install the adapter in a Django, FastAPI, or Spring service; it introspects the API surface you already have and produces typed contracts the runtime validates execution against. Your APIs stay where they are. No parallel inventory, no rebuild around an AI framework.
A full pilot integration in 2-3 days.
Auditable execution from day one.
Connect a real API surface, keep the model-calling code you already have, and ship an assistant with typed execution paths without restructuring your service.
Apparatus.
Every interpretation becomes inspectable.
The missing control surface is not another prompt. It is the runtime machinery around the model: typed contracts, validated plans, explicit evidence checks, deterministic derivation, and records for each interpretation and execution decision.
-
01
Typed contracts
Capture request and response semantics in machine-usable contracts the runtime validates execution against.
-
02
Validated planning
Reject invalid or unsupported plans before execution instead of repairing bad tool calls after the fact.
-
03
Evidence readiness
Make readiness explicit so the runtime knows when evidence is complete and when the path must stop.
-
04
Deterministic derivation
Use controlled computation for totals, comparisons, rankings, and grouping over API data instead of prompt-side math or guesswork.
-
05
Execution lineage
Inspect the inputs, API reads, evidence, operations, and outcomes behind each answer.
Field.
Already piloted in three industries. Internal workflows and customer-facing commerce. Same hard requirement: execution has to be inspectable.
These pilots were chosen because AI crossed into real API-backed work. Operators and customers acted on the result, so each interpretation, source choice, permission check, derivation, and terminal outcome had to be visible after the fact. FMCG retail, market research, and fintech share little operationally. They hit the same boundary: useful AI is not enough without an auditable execution path.