Bring Your Own AI Architecture¶
QLens treats AI as a replaceable execution provider, never as the authority over canonical research.
Normative boundaries¶
- QLens MUST keep canonical research independent of any AI vendor.
- Provider configuration MUST NOT contain API keys, tokens, or secrets.
- Credentials MUST be obtained from named environment variables.
- AI output MUST pass the same narrative traceability and governance checks regardless of provider.
- Changing provider MUST NOT change source OIDs, source digests, publication profiles, or review requirements.
- NUR MUST NOT send credentials to browser JavaScript.
- Public publishing MUST remain an explicit, signed human action.
Supported provider classes¶
| Type | Use |
|---|---|
openai |
Native OpenAI Responses API with strict structured output |
openai_compatible |
OpenRouter, compatible gateways, and self-hosted compatible APIs |
ollama |
Local models through the Ollama HTTP API |
mock |
Deterministic offline testing |
The provider registry is local user configuration. It is not part of the canonical research repository.