Model temperature and max tokens per reply
Tune how creative your agent is and how long each reply can get from the Behavior tab.
Watch this step in the video · Behavior: temperature and reply length
What these two settings control
Inside every agent, on the Behavior tab, the first card is Model & creativity. It holds the two controls that most shape how your AI sounds:
- Temperature: how predictable or creative the agent is.
- Max tokens: the maximum length of each reply.
Getting these right is the difference between a bot that answers short and on-point and one that rambles or makes things up.
How to change them
- Open AI Agents from the side menu and pick the agent you want to edit.
- Go to the Behavior tab.
- In the Model & creativity card you'll find:
- The Temperature slider (0 to 2). The left end reads Deterministic, the right Creative, and the exact value shows in the top-right corner.
- The Max tokens field, a number between 100 and 8192.
- Set the values you want and click Save in the top-right.
Which value to pick
Temperature:
- 0 – 0.3: near-identical replies that stick tightly to your instructions. Great for support, order data, pricing.
- 0.3 – 0.7: the sweet spot for customer service. Sounds natural without drifting. This is the recommended range.
- 0.8 – 2: more varied and creative, but raises the odds the bot improvises or goes off-script. Save it for exploratory sales or brainstorming.
Max tokens:
- A token is roughly 4 characters (about ¾ of a word). 1024 tokens is plenty for short chat conversations.
- Raise it only if your agent needs to return long lists or detailed explanations. Lower it to force shorter replies.
- It's a hard cap: if a reply hits the limit, it gets cut off. Don't set it too low or you'll see messages truncated mid-sentence.
Verify the change
Open the same agent's Sandbox tab and send a couple of messages. With low temperature you'll notice near-identical answers if you repeat a question; with high temperature they'll vary more. If you raised the token cap, try a question that needs a long answer and confirm it no longer gets cut off.
If it doesn't behave as expected
- The agent invents facts: lower the temperature to 0.2–0.4 and tighten your instructions on the Instructions tab.
- Replies are cut mid-sentence: raise Max tokens.
- It sounds robotic or repetitive: nudge the temperature up (try 0.6).
- Token usage spiked: remember more tokens per reply means more cost. See the cost article below.
Tips
- Change one setting at a time and test in the sandbox. Moving both at once makes it hard to tell what caused the change.
- Temperature doesn't affect cost; max tokens does, because it caps how much the model can generate per reply.
Related articles
Related articles
-
Create your first AI agent
Learn what an AI agent does in Omnifox, how to give it instructions, connect it to a channel and test it in the sandbox before it talks to real customers.
-
How to write good instructions for your AI agent
Learn how to write clear instructions so your AI agent replies with the right tone, limits and actions.
-
AI Agents complete guide
AI Agents in depth, in 65 chapters: creating, instructions, tools, knowledge, behavior, tester, inbox, workflows, analytics and usage, with every setting explained.
-
Connect your AI agent to WhatsApp, Instagram, Messenger and web chat
An agent with no channels serves nobody. Connect it to your WhatsApp, Instagram, Messenger or website chat from the Connections section, then switch it on so it starts replying.
-
Start here: AI Agents in 8 minutes
AI Agents essentials in a short video: creating an agent, its instructions, tools, knowledge, channels, the tester, hours and handing over to a person.