Choosing Your Agent's AI Model
The AI Model is the language model that decides what your agent says. Token Limit caps how long each reply can be. Both are on the agent's page in My Agents.
Changing the model
- Open the agent from My Agents.
- Find AI Model and open the list. Models are grouped by provider (Google, OpenAI, Anthropic and others).
- Select a model. It’s saved straight away and used from the next call; a call in progress finishes on the old model.
The list comes from what’s currently available on the platform, so it changes as providers release and retire models. If a model you used before is missing, it has been retired. Pick the closest replacement.
Choosing a model
There’s no single best model. What matters for a phone agent:
- Speed. Callers notice pauses. Smaller, faster models keep the conversation moving and suit booking, message-taking and FAQs.
- Judgement. Larger models follow long, detailed instructions more reliably and handle unusual calls better, but they can be slower to reply.
- Consistency. After switching models, re-test the calls that matter to you. The same instructions can behave a little differently.
A good approach is to start with the default, make a few test calls, and change model only if you can hear a problem.
Token limit
A token is roughly three quarters of a word. Token Limit sets the longest reply the agent can give in one turn:
- A number (1 to 200,000) sets a maximum per reply.
- default uses the model’s own limit.
- unlimited removes the cap.
For phone calls, short replies sound more natural, so most agents are best left on default or given a modest limit. Set it too low and the agent may stop mid-sentence.
Troubleshooting
The agent is slow to respond. Try a faster model, and shorten the instructions in the agent’s behavior.
The agent ignores parts of its instructions. Try a larger model, or make the instructions more specific. See How to Set Agent Behavior.
Replies get cut off. Raise the token limit, or set it to default.