Advanced integrations

Most applications should use the OpenAI wrapper with the default HTTP transport.

Custom storage and application hooks

Use a durable store when guidance must survive restarts. Shared deployments also need application-owned conversation coordination. Sync and async engine/store contracts are documented in the application integration guide.

persist_transcript stores participant messages; persist_result stores completed analysis results. These are independent of HTTP persist. Use on_provider_request with failure_policy="raise" when audit storage must succeed before generation. on_provider_outcome reports response and stream results; callback failures are reported without replacing the provider result.

A different model framework

caesura-io-core provides sync and async engines, stores, typed analysis requests, and rendering helpers. Use it directly when the OpenAI wrapper does not fit your model framework. Keep explicit customer turn identity and application dialogue separate from tool or retrieval input.