Provider keys

Talqing is strictly bring-your-own-key: every model an agent runs (LLM, speech-to-text, voice, avatar, noise cancellation) is called with your key and billed to you by the provider. An agent or task cannot be published until every provider it uses has a key. Only an ADMIN can add, replace or remove a key; any member can see which providers have one.

Adding a key

In the dashboard: Build → Provider Keys, pick the provider, paste the key. Over the API, PUT /v1/byok/{provider} with the provider id from the table below.

curl -X PUT https://api.talqing.com/v1/byok/openai \
  -H "Authorization: Bearer $TALQING_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "api_key": "sk-proj-…" }'

const key = await talqing.providerKeys.set({
  provider: "openai",
  setProviderKeyRequest: { api_key: process.env.OPENAI_API_KEY! },
});
console.log(key.account_label, key.api_key_hint);

import os

key = talqing.provider_keys.set("openai", api_key=os.environ["OPENAI_API_KEY"])
print(key["account_label"], key["api_key_hint"])

PUT again with a new value to rotate; agents use the new key from their next run.

Providers

`provider`	Used for	Where the key comes from
`openai`	LLM, realtime, STT, TTS	`platform.openai.com`
`xai`	LLM, realtime, STT, TTS	`console.x.ai`
`gemini`	LLM, realtime	Google AI Studio
`openrouter`	LLM, STT	`openrouter.ai/settings/keys`. Must be an inference key; management and provisioning keys are refused
`deepgram`	STT, TTS	`console.deepgram.com`
`elevenlabs`	STT, TTS	`elevenlabs.io`, under your profile
`sarvam`	STT, TTS	`dashboard.sarvam.ai` (the API subscription key)
`soniox`	STT, TTS	`console.soniox.com`
`raya`	STT, TTS	`getraya.app`
`tts_cache_router`	TTS through a cache. Your upstream TTS keys go in its own dashboard	`ttscacherouter.dev`, under API keys
`anam`	Video avatars	`lab.anam.ai`
`aicoustics`	Noise cancellation	`developers.ai-coustics.com`. Paste the whole signed SDK licence (`<payload>.<signature>`), not an API key

Which models each provider serves is in GET /v1/catalog; see models.

How a key is verified

Talqing calls the provider with the key before storing it. Nothing is written unless the provider accepts it, so a bad paste never replaces a working key.

Outcome	Response	Stored
Provider accepted the key	`200`	The key, encrypted, plus the account name if the provider exposes one
Provider rejected it	`400` `OpenAI rejected this API key (HTTP 401).`	Nothing
Provider unreachable	`502` `Could not reach OpenAI to verify this API key. Try again.`	Nothing

xAI keys that are blocked or disabled are refused even though xAI recognizes them.

Keys are write-only

No endpoint ever returns a stored key. GET /v1/byok returns, per configured provider:

Field	Meaning
`provider`	The provider id
`api_key_hint`	The last 4 characters, e.g. `…8f2a`
`account_label`	The account name for OpenAI, xAI, ElevenLabs and Deepgram; `""` for the rest
`created_at`, `updated_at`	Timestamps

What fails without a key

Publishing an agent or task is refused with one error per missing provider:

no ElevenLabs API key for this workspace - add one under Provider keys

The check runs at publish (and when a draft is run), never on a draft save, so a draft that saves cleanly can still fail to publish. Run validation to find out early.
An agent needs a key for every provider it names, including a fallback model and an explicitly set analysis model. Left unset, analysis runs on the agent’s own LLM.
A task run re-checks at run time: … - add one under Provider keys before running this task.

Deleting a key does not unpublish anything. Agents that need it keep answering and fail as soon as they reach that provider. Remove a key only after stopping those agents. DELETE /v1/byok/{provider} returns 404 if there was no key.

Where Talqing uses its own keys

Never for an agent run, a call or a task. Only for the CoPilots in the editors, and for browsing the voice and avatar galleries before you have stored a key for that provider. Once you have one, the gallery browses your own account, including voices you cloned.

Models

The catalog, and choosing llm, stt and tts.

Pricing and credits

The platform fee on top of your provider spend.

Get started

Build

Guides

Adding a key

Providers

How a key is verified

Keys are write-only

What fails without a key

Where Talqing uses its own keys

Next

Models

Pricing and credits

​Adding a key

​Providers

​How a key is verified

​Keys are write-only

​What fails without a key

​Where Talqing uses its own keys

​Next

Models

Pricing and credits

Adding a key

Providers

How a key is verified

Keys are write-only

What fails without a key

Where Talqing uses its own keys

Next