Celune DocsAgents
Agents are the AI collaborators that power Celune. Instead of a single generic chatbot, Celune gives you a team of purpose-built agents — each with a defined role, personality, and set of capabilities.
Your Agent Lead
When you first sign up, Celune walks you through creating your Agent Lead. This is your primary AI collaborator: the one you talk to, ask questions, and hand off work to. Think of them as your always-on teammate.
Your Agent Lead has:
- A name and personality — you choose how they communicate: direct, warm, witty, formal.
- A voice — text-to-speech powered by ElevenLabs, so you can have real conversations.
- An avatar — a visual identity that shows up across the platform.
- A persona prompt — the system-level instructions that define how they think and respond.
The Agent Lead is the entry point to Celune. Every task, project, and conversation flows through them first.
Agent Types
Celune supports three types of agents:
| Type | Purpose | | -------------- | -------------------------------------------------------------------- | | Lead | Your primary collaborator. One per workspace. Handles orchestration. | | Specialist | Domain-focused agents (e.g., design, research, code review). | | Utility | Lightweight agents for narrow, repeatable tasks. |
Most teams start with just a Lead. Specialists get added as the workload grows and you want domain-specific expertise without context-switching.
Agent Configuration
Each agent is configured with a set of fields:
| Field | Description |
| ---------------- | ----------------------------------------------------------------------- |
| display_name | The agent's name shown across the UI. |
| role | Short role label (e.g., "Lead Coder", "Designer", "Researcher"). |
| description | What this agent is responsible for. |
| agent_type | lead, specialist, or utility. |
| model | The underlying AI model (e.g., claude-opus-4-6, claude-sonnet-4-6). |
| color | Accent color used in the UI for this agent's cards and avatars. |
| persona_prompt | The system prompt that shapes this agent's personality and constraints. |
| capabilities | What this agent can access: tasks, projects, memory, integrations. |
You manage all of this from Team in the sidebar.
How Agents Collaborate
When your Lead receives a request that's outside their domain — or when multiple things need to happen in parallel — they delegate to specialists.
Delegation is tracked automatically:
- Who handed off the task (
delegated_by) - When the handoff happened (
delegated_at) - Why the task was reassigned (
delegation_reason)
The Team page shows a Delegations tab with the full handoff history for any task.
Agents also share a memory layer — durable notes stored in a database that persist across sessions. This means a specialist doesn't start cold; they can recall relevant context from past work.
Agent Lifecycle
An agent goes through a simple lifecycle:
- Created — configured with name, role, model, and persona during onboarding or from the Team page.
- Active — assigned tasks, participating in projects, generating outputs.
- Archived — removed from the active roster but retained for historical context.
You can create, edit, or archive agents at any time from Team → Agents.
Model Selection
Different agents use different AI models based on the complexity of their work:
| Tier | Model | Best For | | -------- | ----------------- | -------------------------------------------------- | | Heavy | claude-opus-4-6 | Complex engineering, orchestration, deep reasoning | | Standard | claude-sonnet-4-6 | Writing, design, code review, domain work | | Light | claude-haiku-4-5 | Research, simple tasks, high-volume queries |
Matching the model to the task keeps costs in check without sacrificing quality.
Related Concepts
- Workspaces — Agents belong to a workspace. Different workspaces can have different teams.
- Permissions — Control which team members can create, edit, or manage agents.