AI Agents
AI Agent devices paired with your facility connect to AI Models automatically — no API key is ever created, copied, or entered on the device. Each agent's access is secured per device and can be individually configured and bounded.
"Devices paired with this facility connect to AI Models automatically. Set per-agent spend caps, restrict models and override guardrails where required."
Connected AI Agents
Open Settings (cog) → Agent Access & API Keys. The Connected AI Agents card lists every paired agent:
Figure 1: The Connected AI Agents card — agent name and location, online status, spend and limits, and last used
| Column | Meaning |
|---|---|
| Agent | The agent's name and its location at your facility. |
| Status | Online or Offline — live device connectivity. |
| Spend / limits | The agent's spend, and progress toward any per-agent limits (Today, This month, Lifetime, with a Limit reached badge when hit). |
| Last used | The agent's most recent request. |
| Actions | Configure — opens the agent configuration drawer. |
If no agents are paired yet: "No AI Agents are connected yet. Pair an AI Agent device to this facility to see it here." (Pairing devices is done in the AI Agent module, not here.)
Configuring an agent
Click the Configure action to open the agent configuration drawer. The options mirror API-key configuration, scoped to the agent:
| Setting | Behaviour |
|---|---|
| Spend limits (optional) | "Set any combination of daily, monthly and lifetime caps for this agent." Daily resets at the start of each day and monthly on the 1st of each month, in the facility timezone; the lifetime cap never resets. |
| Rate limits (optional) | "Cap requests and tokens per minute for this agent. Leave blank for unlimited." |
| Restrict models for this agent | "Only the selected models will be usable by this agent." |
| Override facility guardrails for this agent | "Select none to explicitly disable guardrails for this agent." |
When an agent hits one of its spend limits, only that agent is blocked (with a clear budget exceeded message stating when the window resets); other agents and API keys keep working. Account-wide limits still apply on top — see Budgets & top-ups.
You can also set per-agent spend limits from Budgets & Limits → Spend Limits → Per-agent spend limits — it edits the same configuration.
Agents and credits
Agents draw from the same prepaid balance as API keys. If the module is not enabled for the facility, or the balance is depleted, agents cannot run inference — the agent's chat shows a credits-required state until the facility is enabled and topped up. Keeping auto top-up on is the easiest way to make sure agents never stop mid-conversation; see Budgets & top-ups.
Agents in analytics
Agents appear alongside API keys everywhere usage is shown: the Overview tab's Top agents & keys, the Activity tab's Top API keys & agents and Key / agent filter, and the Logs tab's Agent / key column. This makes it easy to see exactly what each device is spending and on which models. See Activity & logs.
Security model (in brief)
Each agent authenticates with short-lived, per-device credentials that Performance Hub manages and rotates automatically. There is no static API key on the device to leak, and decommissioning a device revokes its access.