Last updated
What is Together AI used for in a Bolna voice agent?
Together AI gives a Bolna voice agent access to one of the widest catalogs of open-weight models behind a single OpenAI-compatible API. Its serverless tier is per-token with no provisioning, replicas or minimum spend, so you can try several open models on the same agent and keep the one that sounds best. Bolna connects to it as a custom LLM using the Together AI preset.
How do I set up Together AI with Bolna?
- Register at api.together.ai and copy a key from your project's API keys settings page.
- Open Together's serverless models list and copy a model ID from the chat models section. Image, audio and embedding models won't work.
- In Bolna, go to Settings → Integrations → LLM → Custom LLM → Add model, or choose + Add custom LLM from an agent's LLM Provider dropdown. Select Together AI.
- Enter the full model ID, for example
MiniMaxAI/MiniMax-M3, set Endpoint URL tohttps://api.together.ai/v1, paste your key and set Family tominimax,llamaorqwen. - On the Intelligence tab pick custom and the Together model, save, then call the agent.
The Together AI setup guide in the Bolna docs covers rate limits and dedicated options.
What should I know about Together AI models on Bolna?
| Topic | What the Bolna docs say |
|---|---|
| Model ID format | <organisation>/<model_name>, for example MiniMaxAI/MiniMax-M3 |
| Endpoint URL | https://api.together.ai/v1 |
| Model families named | Llama, Qwen, DeepSeek, MiniMax |
| Serverless pricing | Per token, no provisioning, replicas or minimum spend |
| Serverless performance | Best-effort, may be rate-limited in large bursts |
| Guaranteed throughput | Provisioned throughput or dedicated endpoints, with a different model catalog |
What can you build with Together AI and Bolna?
Model A/B tests. Register two or three models as separate custom models, clone the agent, point each copy at a different model and compare call recordings for latency, accuracy and how natural the speech sounds.
Low-commitment pilots. Run a pilot on serverless open models with no minimum spend before you choose a long-term model. Fireworks AI is another serverless option to compare.
Outbound batches. Spread the batch over time instead of launching every call at once, and pick a model that isn't in peak demand. For committed capacity, use provisioned throughput.
Open models without a hosting team. Use Together instead of running your own server on Modal. For one key across closed and open models, see OpenRouter.