Core Concepts
Personas & Agents
Configuring agent personalities, system prompts, knowledge scoping, and home registry.
An Agent in Omni is a configured persona attached to a knowledge workspace. Agents combine custom system instructions, conversational vibes, tool capabilities, and long-term memory.
Agent Anatomy
An agent definition includes:
| Property | Description | Example |
|---|---|---|
personaName | Display identifier for the agent | "Research Assistant" |
vibe | Tone and communication style | "analytical", "creative", "concise" |
systemPrompt | Authoritative system directive instructing model behavior | "You are an expert financial analyst..." |
workspaceId | Attached knowledge scope containing documents and memories | "ws_01jze9m2xw" |
tools | Enabled MCP and internal tool suites | ["web_search", "memory_query"] |
Agent Home Registry
Omni features an atomic Agent Home selection registry that maps an authenticated user or workspace to their active default agent persona.
When a message is sent to /v1/chat/completions or /api/v1/tasks, Omni resolves the active persona:
- Loads the workspace profile and biographical prompt sidecar (
user_bio.om). - Synthesizes system instructions with persona constraints.
- Injects relevant historical memories into the context.
- Executes the model reasoning loop.
Swapping Personas
Personas can be modified or dynamically swapped per workspace:
curl -sS -X POST "https://edge.omnistatic.com/v1/workspaces/ws_123/persona/swap" \
-H "Authorization: Bearer $OMNI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"personaName": "Code Reviewer",
"vibe": "concise"
}'