Agent Types
and Capabilities
Clear overview of what agent types exist in Ona and what each can do. Use this page as the fastest reference for who handles what, what capabilities are available, and where boundaries apply.
Quick Model
- User talks to Solin
- Solin delegates to specialist agents
- Specialist agents execute tasks and tools
- Oversight and security agents monitor and enforce safety
Core Agent Types
| Agent Type | Primary Purpose | Key Capabilities | Limits |
|---|---|---|---|
| Solin (Squad Lead) | Main orchestrator and user-facing assistant | Understands mission intent, splits tasks, delegates, combines final answer | Does not execute full tool loop directly in full-job mode |
| Task Specialist Agents | Execute domain work | Run tools, use memory, search, call subordinates, produce artifacts | Scoped to assigned tasks and policy |
| Observer (optional) | Mission quality and safety oversight | Watches flow, flags risky outputs, reports to Solin | Does not execute normal mission tasks |
| Guardian (optional) | Security enforcement | Validates tool params/outputs, detects misuse, blocks high-risk actions | Does not execute normal mission tasks |
| Operations Agents (optional) | Operational domain workflows | Lockbox operations, cert lifecycle support, project/task workflows | Limited by approvals, policy, and enabled features |
| Assistant/Subordinate Agents | Extra capacity and specialization | Parallelize subtasks and support specialist workload | Ephemeral by default and bounded to delegated scope |
Specialist Agent Personas
Common persona templates include Writer, Content Creator, Social Content Creator, Researcher, Developer, Designer, SEO, Marketer, Sales Agent, Landing Page, and Project Agent. You can create custom agents and include them in delegation when configured.
| Persona Group | Typical Capabilities |
|---|---|
| Content (Writer/Content/Social/Marketing/Sales) | Drafting, campaign assets, structured messaging, outreach content |
| Research (Researcher/SEO) | Discovery, synthesis, keyword structure, evidence-backed summaries |
| Build (Developer/Designer/Landing Page) | Implementation tasks, UI/design iteration, landing outputs, build/test execution |
| Operations (Project Agent) | Project tracking, status management, execution planning support |
Oversight and Security Agents
Observer Agent
- Focus: mission alignment, quality and safety watch, anomaly signaling.
- Can: read mission context, report concerns, reduce off-track execution.
- Cannot: replace specialist execution or bypass controls.
Guardian Agent
- Focus: policy enforcement, abuse detection, risk blocking.
- Can: inspect risky tool patterns and enforce deny/warn behavior.
- Cannot: replace business specialists or bypass governance boundaries.
Operations-Focused Agents
Lockbox Agent
Manages lockbox lifecycle operations through controlled flows and approval-aware credential workflows.
Limit: should never expose plaintext secrets.
Certificate Manager
Handles certificate and key lifecycle workflows, including status, renewal, and rotation operations.
Limit: follows approval and security policy requirements.
Project Agent
Supports project and task management workflows, including create/list/update and execution tracking.
Limit: not a replacement for research/code/design specialists.
Interaction Modes and Capability Depth
| Interaction Path | What Usually Happens | Capability Depth |
|---|---|---|
| Quick reply | Fast response path for simple interactions | Lightweight, low-latency response |
| Full mission job | Solin orchestrates specialist agents and tools | Full multi-step execution and collaboration |
Use full mission mode when you need deeper output, multi-step work, or specialized execution.
Boundaries That Always Apply
- User-facing conversation goes through Solin.
- Specialist agents execute within assigned mission scope.
- Approvals are required for sensitive actions.
- Oversight and security controls can intervene in high-risk flows.
- Memory and tool behavior follows configured policy and scope.