Agents
Agents
Neutron AI helps agents remember useful context across sessions without turning your product into an unbounded chat log. Each agent should have a clear purpose, explicit scopes, and predictable memory behavior.
Recommended Agent Contract
Define these fields before connecting an agent to Neutron:
| Field | Purpose |
|---|---|
agentId | Stable identifier for the workflow or assistant. |
| Purpose | What the agent is allowed to help with. |
| Input contract | What the user or system can provide. |
| Output contract | What the agent must return. |
| Scope policy | Which tenant and scope IDs can be used. |
| Memory policy | What can be remembered, recalled, reflected, or forgotten. |
| Safety policy | What should never be stored or revealed. |
Agent Loop
flowchart LR
A["User task"] --> B["Request scoped context"]
B --> C["Call model provider"]
C --> D["Return product response"]
D --> E["Reflect useful outcome"]
E --> F["Store safe memory"]
Keep the current user task as the primary instruction. Treat Neutron memory as supporting context that can improve continuity and personalization.