Skip to main content

Documentation Index

Fetch the complete documentation index at: https://kupe.in/docs/llms.txt

Use this file to discover all available pages before exploring further.

Use this guide when calling PUT /api/v1/agents/{agent_id} (after POST /api/v1/agents/simple if you start from defaults): pass catalog codes (what you see below or from the live provider APIs), not database row UUIDs, unless you already have the UUID from your environment.

​
Live catalogs (authoritative)

With your x-api-key:
CatalogEndpoint
LLMGET https://api.kupe.in/api/v1/providers/model
TTSGET https://api.kupe.in/api/v1/providers/tts
STTGET https://api.kupe.in/api/v1/providers/stt
Each row includes fields such as model_name, provider_name, supported_languages, and id. Agent update accepts preferred keys agent_model_config.model_name, tts_config.tts_model_name, transcriber_config.transcriber_model_name (legacy *_provider_id fields are equivalent) and resolves them to internal UUIDs before saving. See also: Supported models & providers (static tables and model names) and the Voice Agent API in the API reference.

​
Language codes by provider

Use the convention that matches the STT / TTS row you selected. Wrong codes usually surface as provider errors at call time, not as Kupe validation errors.
Sarvam expects region-qualified tags where the provider documents them, e.g. hi-IN, gu-IN, en-IN. Use the same pattern as in Sarvam’s docs for other locales.Put the value in transcriber_config.language. For TTS you can send tts_config.language (Kupe merges it into voice_parameters.language) or set voice_parameters.language yourself.

​
Logos and quick reference

Logos below match the in-app catalog (app.kupe.in). Model names are examples — always confirm against GET …/providers/model for your tenant.

​
LLM

​
OpenAI / Azure router

Examples: gpt-4.1-mini, gpt-4.1, openai-gpt-4.1-nano

​
Groq

Examples: groq-llama-3.3-70b, groq-gpt-oss-20b

​
DeepInfra

Examples: deepinfra-qwen, deepinfra-llama

​
TTS

Use tts_config.tts_model_name = model_name, provider_name, provider:model_name, or row id. Examples: tts-1, bulbul:v2. Legacy tts_provider_id in JSON means the same thing.

​
STT

Use transcriber_config.transcriber_model_name = STT model_name, transcriber name, or row id (legacy provider_id is equivalent), and transcriber_config.language. The pipeline STT model id is taken from the transcriber_providers row — do not send a duplicate transcriber_config.model_name on create.
Refreshing this page: run the three GET catalog calls and compare model_name / supported_languages to the tables on Supported models & providers; update product copy when vendors add models.