overlay: 1.0.0 info: title: API Evangelist enhancements for the Sarj TTS API version: 1.0.0 extends: openapi/sarj-ai-developer-api-tts-openapi.json x-provenance: generated: '2026-08-09' method: generated source: openapi/sarj-ai-developer-api-tts-openapi.json actions: - target: $.info update: x-apievangelist-slug: sarj-ai-developer-api x-apievangelist-reviewed: '2026-08-09' x-apievangelist-docs: https://platform-docs.sarj.ai/text-to-speech x-apievangelist-note: >- Harvested from the API host root at https://sarj-omni-tts.sarj.ai/openapi.json. The spec declares no servers[]; the documented production base URL is https://sarj-omni-tts.sarj.ai/v1. info.version is 0.3.0 — the least mature of the four Sarj specs. - target: $.info update: x-apievangelist-compatibility: family: OpenAI Speech drop_in: true source: https://platform-docs.sarj.ai/text-to-speech - target: $.paths['/v1/audio/speech'].post update: x-apievangelist-note: >- Documented parameters are model (required, "gpt-4o-mini-tts"), voice (required), input (required), response_format (default wav) and num_step (default 16). Published voices are ars_male, ars_fares and ars_female_expressive. - target: $.paths['/v1/audio/speech/clone'].post update: x-apievangelist-note: >- Voice cloning from a reference sample. Provider guidance is a clean 6-10 second recording with matching reference text. x-agentic-access: action-class: acting consequence: physical note: >- Raised above the pipeline default. Cloning a voice from a reference recording is a biometric-adjacent operation with consent implications; it should not be exposed to an agent without an explicit human step. - target: $.paths['/metrics'].get update: x-apievangelist-note: >- An operational metrics endpoint present on the public production spec and absent from the documentation. Flagged for the provider rather than recommended for integrator use.