Configure the agent's LLM model, model routing, fallback providers, heartbeat schedule, voice, and description.
The Configuration tab is where you control the agent's brain. Choose which AI model it uses, how it handles different types of tasks, what happens when a model goes down, how often it checks in proactively, and what voice it uses in meetings. Each setting is organized into its own card with inline help.
Configuration is stored across several config keys: model (provider + modelId + thinkingLevel), modelRouting (per-context overrides), modelFallback (chain + retry settings), heartbeat (interval, active hours, custom prompt), voiceConfig (ElevenLabs voice ID), and description. Each card has independent edit/save state. Saves use hot-update when running, PATCH when stopped. Provider/model lists come from /providers and /providers/:id/models APIs.
The Configuration tab contains 6-7 independent cards, each with its own Edit/Save/Cancel flow. Click "Edit" on any card to modify that section without affecting others. Changes apply immediately if the agent is running (hot-update).
The primary AI model for all conversations and tasks.
| Setting | Description |
|---|---|
| Provider | The AI provider — Anthropic, OpenAI, Google, etc. Must have an API key configured in Settings → Integrations. |
| Model | The specific model (e.g., claude-sonnet-4-20250514, gpt-4o). Auto-populated from the provider's model list. |
| Thinking Level | Extended reasoning capability. Off / Low (2K tokens) / Medium (8K) / High (16K). Higher = better for complex tasks but slower and more expensive. |
Assign different AI models to different types of tasks for cost/quality optimization:
| Context | Description | Recommended |
|---|---|---|
| Chat | Real-time conversations (Slack, Teams, WhatsApp) | Sonnet 4 — fast replies |
| Meetings (Voice) | Google Meet with voice capabilities | Sonnet 4 — low latency |
| Composing and replying to emails | Opus — professional quality | |
| Tasks & Projects | Complex work, analysis, research | Opus — deep reasoning |
| Scheduling | Calendar, reminders | Haiku — quick and cheap |
Leave a route empty to use the Default LLM Model for that task type. Each route has its own provider/model selector.
Automatic failover when the primary model fails:
Periodic check-ins where the agent wakes up and proactively handles pending work.
| Setting | Description |
|---|---|
| Interval | 5 min to once daily. Default: every 30 minutes. |
| Active Hours | Only run heartbeats during these hours (agent's timezone). Set 12AM–12AM for 24/7. |
| Custom Prompt | Override what the agent checks during heartbeats. E.g., "Check emails, calendar events in next 2h." |
| Tokens per Beat | Estimated tokens per cycle — 1K (minimal), 3K (typical), 5K (moderate), 10K (heavy). |
The heartbeat card shows a real-time cost estimate based on your model and interval:
Example: 30-min interval with Haiku ≈ $0.10/day vs Opus ≈ $5/day.
Configure a text-to-speech voice for meetings and voice interactions. Features:
Requires an ElevenLabs API key configured in Settings → Integrations. Without it, voices are shown but won't work.
A human-readable description of what this agent does. Shown to other agents, team members, and included in the agent's system prompt. Be specific about responsibilities.
If the agent was created from a role template (soul), the template name is shown as a badge. Templates pre-configure identity, personality, skills, and permissions for specific roles.
| Problem | Solution |
|---|---|
| "No providers configured" warning | Add API keys in Settings → Integrations for at least one provider. |
| Model dropdown is empty | Select a provider first. If still empty, the provider's API key may be invalid. |
| Heartbeat cost too high | Increase the interval, reduce active hours, or switch to a cheaper model. |
| Voice not working in meetings | Ensure ElevenLabs is connected in Settings → Integrations. Verify the voice ID is valid. |
| Fallback not triggering | Ensure fallback is enabled and models in the chain have valid API keys. |
| Config changes not applying | If the agent is running, hot-update should work instantly. Check for error messages. Try restarting the agent. |