> ## Documentation Index
> Fetch the complete documentation index at: https://docs.sureops.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Per-Agent LLM Config

> Configure which LLM model each agent uses, with per-environment overrides.

Each sureops agent can use a different LLM model. By default, all agents inherit the org-level default model. You can override this per agent and per environment — for example, using a more capable model for the Diagnosis Specialist in production while using a faster model in staging.

## Default behavior

When you first set up sureops, all agents use the org-level default model. This model is configured at **Settings → AI → LLM Configuration → Default model**.

Changes to the org-level default immediately apply to all agents that haven't been individually overridden.

***

## Overriding a specific agent's model

<Steps>
  <Step title="Open agent settings">
    Go to **Settings → AI Agents** and select the agent you want to configure.
  </Step>

  <Step title="Toggle custom model">
    On the agent's configuration card, toggle **"Use custom model"** to on. This breaks the inheritance from the org default for this agent.
  </Step>

  <Step title="Select a model">
    Choose from the available models in the dropdown. Available models depend on your LLM configuration (managed or BYOK).
  </Step>

  <Step title="Save">
    Click **Save**. The change takes effect on the next incident that uses this agent.
  </Step>
</Steps>

To revert to the org default, toggle **"Use custom model"** back to off.

***

## Per-environment overrides

An agent can use a different model in different environments. For example:

* Diagnosis Specialist in **production**: a high-capability model for accuracy on critical incidents
* Diagnosis Specialist in **staging**: a faster, lower-cost model for exploratory use

To configure per-environment:

1. Go to **Settings → AI Agents** and select the agent
2. Select the environment from the environment dropdown (top of the page or within the agent card)
3. Toggle **"Use custom model"** and select the model for that environment

The environment-scoped override takes precedence over the agent-level default, which takes precedence over the org default.

**Precedence order (highest wins):**

```
environment-scoped agent override
  → agent-level override
    → org-level default
```

***

## Available agents

| Agent                | When it runs                    | Default recommendation                                  |
| -------------------- | ------------------------------- | ------------------------------------------------------- |
| Incident Commander   | Every stage transition; closure | Org default is usually appropriate                      |
| Diagnosis Specialist | Diagnosis stage                 | Consider a high-capability model for production         |
| Resolution Agent     | Resolution stage                | Org default or same as Diagnosis                        |
| Verification Agent   | Verification stage              | Org default is usually appropriate                      |
| Problem Investigator | Post-closure, async             | High-capability model for deep RCA                      |
| Fix-PR Agent         | Post-closure, async             | Org default is usually appropriate                      |
| Pre-mortem Agent     | Pre-deployment                  | Org default is usually appropriate                      |
| ECHO (chat)          | On-demand chat                  | Org default; consider a faster model for responsiveness |
| SAGE (incident chat) | In-incident stage chat          | Same as Diagnosis Specialist in most cases              |

***

## Available models

Available models depend on your LLM configuration:

* **Managed mode**: sureops selects from a curated set of Anthropic and OpenAI models. The available set is shown in the model dropdown in Settings.
* **BYOK mode**: all models supported by your configured provider are available. The dropdown reflects the models sureops has validated for agent use.

Model availability varies by plan. If you need a specific model that isn't listed, contact [hello@sureops.ai](mailto:hello@sureops.ai).

For details on managed vs. BYOK, see [BYOK vs Managed LLM](/next/ai-agents/byok-vs-managed).

***

## Cost considerations

Different models have different per-token costs. Agents running on high-capability models for every incident can meaningfully increase your LLM spend.

A practical approach:

* Set a cost-effective default for most agents
* Override only the agents where output quality matters most (Diagnosis Specialist, Problem Investigator)
* Use per-environment overrides to use expensive models only in production

In managed mode, sureops applies per-org spend ceilings. Your ceiling is visible at **Settings → AI → LLM Configuration → Budget**. Exceeding the ceiling does not cause incidents to fail — the agent degrades gracefully and notifies you.

***

## Budget and enforcement

Each org has a configurable LLM budget. Budget enforcement has three modes:

| Mode    | Behavior                                                                                                                     |
| ------- | ---------------------------------------------------------------------------------------------------------------------------- |
| Off     | No enforcement. Usage is tracked but not limited.                                                                            |
| Warn    | Sends a notification when usage approaches the ceiling.                                                                      |
| Enforce | Blocks new LLM calls when the ceiling is reached. Agents degrade gracefully (use less expensive paths or escalate to human). |

The enforce mode is designed for cost control, not safety. An enforced org that hits its ceiling has its incidents escalated to human review rather than silently failing.