Agents¶

LLM-backed entities that run the incident pipeline. Each agent has a trust level, a role set, an LLM provider/model binding, and a system prompt.

List¶

Route: /agents Role gating: read for all; create / edit / delete require admin.

Card grid with name, enabled toggle, LLM provider/model, trust level, assigned roles, and a token-usage progress bar against the configured budget.

Below the agent cards the page shows the Auto-execute overrides table — recipe/server-role pairs that bypass the approval gate. The table has Add override / Revoke actions. Overrides can also be granted via Promotion Suggestions on a live-mode server detail page.

Create¶

Modal collects:

Name and description.
LLM provider (deepseek / openai / local) and model. The agent detail page uses the full provider catalog from Settings → LLM (which may include anthropic, kimi, llamacpp, vllm when those providers are configured).
Trust level — autonomous, supervised, manual. Determines the approval gate; see below.
Role checkboxes — which pipeline stages this agent can handle: triage, diagnose, execute, review.
Patrol interval (minutes).
Monthly token budget.

Detail¶

Route: /agents/{id} Role gating: admin.

Configuration panel with four sections.

Configuration¶

Name, LLM provider/model (with a Refresh Models icon button that pulls the live model list from the provider and a free-text fallback input), trust level, role checkboxes (triage, diagnose, execute, review), patrol interval, token budget, custom system prompt. A Save Changes button at the bottom submits the whole section.

LLM Settings¶

Per-agent overrides for temperature, max tokens, and a Reasoning Enabled toggle (chain-of-thought; leave blank to use the global defaults from Settings → LLM).

Skills¶

Markdown knowledge modules currently assigned to the agent. Inline remove button per skill plus a dropdown to add more.

Stats (read-only)¶

Incidents resolved.
Active incidents.
Tokens used this period.
Token budget (or "Unlimited").

Trust gate¶

Trust level interacts with each recipe's risk level via the should_request_approval server-side gate.

Trust × risk	`none`	`low`	`medium`	`high`
`autonomous`	auto	auto	approval	approval
`supervised`	auto	auto	approval	approval
`manual`	auto	approval	approval	approval

The LLM cannot self-approve. Risk classification on a recipe is the operator's responsibility, not the agent's.

skills.md — assignable knowledge modules
incidents.md — agents are assigned to incidents
settings.md — global LLM provider configuration and prompt templates