Agents

Sidebar → Agent Management → New

An agent bundles a model, a system prompt, and the tools / skills / knowledge it can reach. You then assign the agent to chat sessions or channels.

Agent Modes

Single

Pick a model, write a system prompt, optionally attach MCP tools and skills. The default mode for focused single-purpose assistants.

ReAct

Pick a think model, then add sub-agents (each with an id and description for task dispatch). The think model decomposes user requests and dispatches sub-tasks recursively; each sub-agent has read-only access to shared memory.

Use ReAct when:

• The task is open-ended ("plan and execute X end-to-end")
• You want the orchestrator to choose specialists dynamically

Each dispatched sub-task can inherit context from the parent conversation (none — clean start, the default; state — a bounded snapshot of recent parent messages; full — the full forked history). Recursion depth is guarded to prevent runaway nesting.

Generative

Pick a multimodal model for mixed text + image content generation.

Configuration

Section	Purpose
Model	Primary LLM for this agent
System prompt	Persona, capabilities, response style
MCP tools	Per-agent enable list of MCP servers
Skills	Per-agent skill selection (empty = load all)
Notes	Default notes (vector store) for sessions using this agent
Wiki	Default wiki/knowledge base for sessions
Memory	Per-agent long-term memory via background MemoryLLM — see Memory
Agenda	Per-agent reminders / schedules, optionally synced from the conversation — see Agenda
Heartbeat	Periodic self-activation — see Heartbeat

Pre-Built Agents

Want to skip the manual setup? Browse the Agent Store for ready-to-install bundles (model + prompt + tools + skills + MCP servers).