Signal Triage Agent
- Input
- Recent signed alerts, GPS envelope, alert category, device identity, expiration window.
- Output
- Clustered incident candidates, duplicate suppression notes, anomaly flags, and operator priority.
AI agent model
Looma uses AI to interpret and govern road signals, not to drive the vehicle.
The agent story is strongest when it is operationally specific: agents read signed alerts, cluster incidents, draft briefs, produce pilot evidence, screen counterparties, and enforce approval boundaries through MCP tools and governance logs.
Incident Briefing Agent
- Input
- Alert cluster, route context, event timeline, nearby feed, and operator notes.
- Output
- A short incident brief that explains what changed, where it changed, and why the operator should care.
Pilot Evidence Agent
- Input
- Relay events, uptime, message delivery, coverage zones, customer notes, and demo activity.
- Output
- Investor and customer reports showing product usage, reliability, quality gaps, and pilot readiness.
Integration Agent
- Input
- Fleet/OEM requirements, API docs, sample payloads, data contracts, and LunaForge impact reports.
- Output
- Adapter specs, test payloads, integration checklists, and engineering-risk summaries.
Compliance Agent
- Input
- Customer entity, vendor entity, procurement contact, region, contract notes, and Amliq screening result.
- Output
- Counterparty risk summary, case reference, reviewer recommendation, and audit trail link.
Governance Agent
- Input
- Requested MCP call, user role, customer tenant, approval state, policy, and OpenSyber audit context.
- Output
- Allow, deny, require approval, redact, or escalate with a durable audit record.
Generative AI boundaries
The model writes explanations, reports, specs, and reviews. It does not create unapproved safety outcomes.
Allowed generative work
- Summarize many alerts into one operational incident brief.
- Draft investor-ready pilot reports from telemetry and notes.
- Explain integration risk in plain language for non-technical buyers.
- Generate adapter specs and test payloads for fleet or city integrations.
- Redact sensitive details before evidence leaves a tenant boundary.
Blocked positioning
- No autonomous braking, steering, dispatching, or routing commands.
- No uncertified claim that a generated summary is a road-safety decision.
- No agent writes to customer-facing evidence without approval and audit.
- No compliance screening outside the approved counterparty workflow.
Permission model