Agents
Every agent is a first-class citizen.
Matrix agents are typed actors with persistent memory, permissioned tool access, and a lifecycle managed by the Executor. Build one in minutes, trust it in production.
Personal Agents
Single-user actors with private Cortex memory and full UWAC-vaulted tool access. Runs continuously, learns from every session.
Team Agents
Multi-user shared agents with scoped access control. Useful for dev teams, ops workflows, and knowledge bases.
Task Agents
Ephemeral, single-purpose actors spawned for a specific job and garbage-collected on completion. Ideal for pipelines and automation.
Service Agents
Always-on agents registered in the Deus marketplace, offering discoverable services to other agents via metered invocation.
Agent lifecycle
Every Matrix agent goes through a defined lifecycle managed by the Executor: spawn → authorize → load context → plan → walk → commit or rollback → idle. The Executor handles retries, timeouts, and error recovery at each step.
1. Spawn & Authorize
Agent is instantiated with a typed identity and permission set.
2. Load Context
Cortex memory pages are retrieved and injected into the model context.
3. Plan → Walk → Commit
Intent IR is planned, validated, and walked by the Executor with atomic commit or rollback.