Reference Architectures
Long-Running Agents
Long-Running Autonomous Agents execute complex tasks over extended periods that may span hours or days. These agents maintain persistent context and operate with limited supervision.
Architecture diagram
Agent (Persistent) ---> Task 1 (Hour 1) ---> Task 2 (Hour 3) ---> ...
A persistent agent executes a sequence of tasks over time. Context accumulates across tasks.
Characteristics
Long-Running Agents suit complex workflows that cannot complete within a single session. Session duration creates unique challenges including credential validity, context accumulation, and oversight across time periods.
Priority controls
| Component | Priority Controls |
|---|---|
| Session Management | AZN-03, IDN-03, OBS-02 |
| Context Accumulation | OBS-05, INT-02, GOV-04 |
| Long-Lived Credentials | IDN-03, IDN-04, AZN-05 |
| Asynchronous Intervention | INT-01, INT-02, INT-04 |
Typical certification profile
Long-Running Agent deployments typically certify at Elevated or Critical profiles. The extended exposure window warrants heightened controls.