Integrations & BYO keys
Connect your own LLM, text-to-speech, and speech-to-text providers, and wire up data apps to push your campaign results — all from the Integrations page.
Integrations let you bring your own provider keys and connect the outside apps your agents rely on. You store a key or connect an account once, pick the model or voice you want, and every agent in your workspace can use it. Integration secrets are stored encrypted (AES-256-GCM) and scoped to your workspace.
Connecting your own LLM keys
Callable is provider-flexible: you can bring your own key for the large language model that powers your agents' replies. Supported providers include Anthropic, OpenAI, Google Gemini, Groq, and Mistral (and more).
To connect an LLM provider:
- Open Integrations from the left nav.
- Choose the LLM provider you want to connect (for example, Anthropic or OpenAI).
- Paste your API key for that provider.
- Pick the model you want to use when you connect.
- Save. The key is encrypted and stored against your workspace.
Model precedence: per-agent override vs. default
Once a provider is connected, there are two levels at which a model can be chosen, and they resolve in a clear order:
- Per-agent model override — set on an individual agent (the model preset on its Agent tab, fine-tuned under Settings). This always wins for that agent.
- Owner's default — the workspace default. Used whenever an agent has no override of its own.
In other words, a per-agent model override takes precedence; otherwise the agent falls back to the owner's default model.
TTS providers (text-to-speech)
Text-to-speech turns your agent's reply into the voice the caller hears. Supported providers:
- Cartesia — the platform default (Sonic voices), available without bringing your own key.
- ElevenLabs — bring your own key.
Connect ElevenLabs on the Integrations page the same way as an LLM provider (paste your key, save). You then choose the specific voice, speed, language, and fallback voices on an agent's Voice tab.
STT providers (speech-to-text)
Speech-to-text transcribes the caller's audio so the agent's LLM can respond. Supported providers:
- Deepgram
- AssemblyAI
- Whisper
You configure transcriber settings — such as the model, endpointing, and language — on an agent's Transcriber tab.
Data apps for results
Connect data apps so you can push your campaign results out to the tools your team already uses. Supported apps:
- Google Sheets
- Airtable
- Notion
- Excel
Connect these on the Integrations page, then use them to send campaign results (leads, dispositions, and other outcomes) into your own systems.