Skip to main content
OpenAI powers Lucia’s model-assisted reasoning and language refinement. The model layer must remain upgradeable without breaking Lucia’s response contract.

​
Current Runtime

Lucia Engine v0.1.3.6 Development uses:
The Lucia Guest Agent Netlify production deployment uses:
Both surfaces currently use gpt-5.6-sol, but they own independent configuration, deployment, and verification boundaries. Engine runtime proof does not establish Guest Agent runtime truth, and Guest Agent proof does not establish Engine runtime truth. The Guest Agent does not expose or persist the Engine’s /admin/operator-focus runtime-provenance fields. Its live provider-returned model proof is deployment evidence for the Guest Agent only. The purpose is not novelty. The purpose is:

​
JSON Model Gateway

Lucia’s JSON model gateway has migrated from Chat Completions to the OpenAI Responses API. Current Development truth:
The gateway is infrastructure. It may improve reasoning and language quality, but it must not change Lucia’s public response contract by accident.

​
Critical Constraint

Model upgrades must not break:

​
Model Does Not Own

The model must not own:

​
Model Can Improve

The model may improve:
Model output sits above deterministic truth. Guest identity, booking IDs, stay windows, payment state, task status, route/action safety, and save/completion truth remain system-owned. Guest-facing Lucia follows the same rule: identity_state, verification_state, linkage_state, verification email destination, token validity, and booking-private data access remain system-owned. Workspace context follows the same rule. GPT-5.6 Sol may explain the current surface in natural language, but deterministic context owns the current surface, active booking/task/action, saved DAW workflow state, and whether the underlying issue is still unresolved.

​
Runtime Provenance Contract

The Development /admin/operator-focus route exposes runtime and model evidence in the same response that contains Lucia’s answer. The public top-level contract remains stable:
meta.runtime reports the Engine service_id, environment, version, and commit_sha. meta.model_runtime reports:
The gateway captures the provider-returned response.model as task-level resolved_model. The top-level resolved model is reported only when the attempted model tasks resolve consistently. Per-task evidence records whether a call was attempted, a provider response arrived, parsing succeeded, fallback was used, and the task status. Interpretation rules:
A provider-backed success additionally requires successful task evidence, a provider response, parsing success, a resolved model, and no fallback for that task. A provider failure can therefore report model_invoked=true and fallback_used=true; a client-disabled or no-attempt fallback can report model_invoked=false, deterministic_path=true, and fallback_used=true. A 200 response does not by itself prove model execution or provider success. Consumers must read provenance rather than infer it from the route status or configured model.

​
Current Upgrade Status

The GPT-5.6 Sol default and Responses API gateway are active in Development for Lucia Engine v0.1.3.6. This supersedes the earlier Engine GPT-5.5 Development baseline; the dated GPT-5.5 decision and release entries remain historical evidence. The Guest Agent’s independent GPT-5.6 Sol upgrade was live-verified on 2026-07-12. Its prompts, verification flow, routing, schemas, and core SSE events (start, token, cta, meta, done) were unchanged. Model upgrades are infrastructure changes only when response contract and product behavior remain stable. v0.1.3.6 has not been promoted to Staging yet.

​
See Also