Pick the Right LLM
How to choose a model for your agent without overthinking it.
A quick decision tree
- Does the agent use any tools?
- Yes → you need a model with the Tools capability. Most modern models have it.
- Does the agent process images or multimodal input?
- Yes → you also need the Vision capability.
- Does the agent answer questions requiring careful multi-step reasoning?
- Yes → consider a model with Thinking and set a thinking budget.
- Is voice involved?
- Yes → set a Speech Model in addition to the Text Model.
After that, pick the cheapest model in the right tier.
Where to set it
Agent editor → Basic Info section.
Two model selectors:
- Text Model — required.
- Speech Model — optional. Default is "No speech model selected" for text-only agents.
Comparing models
Both Text Model and Speech Model headings have a Compare button. Click it to open a comparison table of available models — useful for evaluating options on cost, speed, and capabilities before committing.
The model dropdown
Click the model selector button to open the dropdown. It provides:
- A search box ("Search text models…") at the top — type to filter quickly.
- Models grouped by provider (Anthropic, Amazon Bedrock, Google, OpenAI, etc.).
- Provider headers marked "API key not set" when credentials haven't been configured — those models are greyed out and won't work until a super-admin adds the key on the Deployment Settings → LLM Provider API Keys page.
- A model details button (info icon) on each row — click it to see pricing per million tokens and capabilities.
Capability indicators
The dropdown flags mismatches:
| Visual state | Meaning |
|---|---|
| Greyed out | Provider API key not configured. |
| Red | Model is missing a capability the agent needs. |
You can still select a red model, but the agent won't work correctly. Pick a compatible one instead.
Thinking budget
Below the Text Model selector (when a thinking-capable model like Claude is selected), a four-button picker appears:
| Preset | Tokens reserved | When to use |
|---|---|---|
| Off | 0 | Default. Use for most agents. |
| Low | 2,048 | Light extra reasoning. |
| Medium | 4,096 | Multi-step questions. |
| High | 8,192 | Hard problems. |
This section only appears for thinking-capable models. If you don't see it, your selected model doesn't support thinking.
Start with Off. Bump up only if the agent gives wrong answers on complex, multi-step questions.
Cost vs. speed vs. quality
A rough guide:
| Goal | Pick |
|---|---|
| Cheapest, fastest | A small/fast model (e.g., a Lite or Flash variant). |
| Balanced quality | A mid-tier model (e.g., Sonnet-class, Flash-class). |
| Best quality | A frontier model with thinking enabled. |
Use the Compare button or click the info icon on any model row to see pricing.
Switching models on a live agent
Open the agent → change the Text Model selector → click Update Agent.
Existing conversations keep the old model; new conversations use the updated one. There's no migration cost — it's just a selector change and a save.
Tips
- Don't pre-optimize. Ship with the cheap model. Upgrade only when you see a quality problem.
- Use the Compare button. It's the fastest way to evaluate options side by side.
- Test on real conversations. Synthetic prompts can mislead. Run the same question on both candidates and compare.
- Watch the bill. Switching from a small to a frontier model can multiply cost 10× or more. Check Analytics regularly.
- Voice and text are independent. A powerful text model doesn't make voice better. Set the Speech Model separately.