Routes
/agents— list all agents/agents/:id— agent detail page/agents/:id/edit— agent editor/agents/:id/workspace— workspace manager (covered on its own page)
Backing RPC
The list, detail, and editor are powered by:agents.list— list viewagents.get— detail pageagents.create,agents.update,agents.delete— editor and delete actionsagents.suspend,agents.resume— status actionsagent.getOperationModels— shows the resolved primary model and per-operation overrides
persistToConfig (writes agents.<id> into config.yaml, then schedules a daemon restart).
What You See
The Agents view has three sub-views: a list of all agents, a detail page for each agent, and an editor for changing agent settings.Agent List
The main view is a sortable data grid showing every configured agent. Click any column header to sort by that column (the Actions column is not sortable). The columns are:| Column | What It Shows |
|---|---|
| Status | Current state, shown as a colored tag |
| Name | The display name you gave the agent (falls back to the agent ID when no name is set) |
| Model | The AI model the agent uses (such as Claude or GPT) |
| Messages | Messages handled by this agent today |
| Cost | Today’s spend for this agent in USD |
| Budget | A bar showing how much of the agent’s daily token budget has been used |
| Actions | Icon buttons for Configure, Suspend/Resume, and Delete |
- Active (green) — The agent is running and processing messages
- Idle (blue) — The agent is running but has had no recent activity
- Suspended (yellow) — The agent has been paused by a user
- Error (red) — Something went wrong and the agent needs attention
Agent Detail
The detail page gives you a comprehensive look at a single agent. At the top, a breadcrumb trail lets you navigate back to the agent list. Stat cards across the top show at-a-glance numbers:- Messages handled today
- Tokens used (input and output combined)
- Tool calls made
- Configuration summary — The agent’s provider, model, max steps, and other core settings displayed in a two-column grid
- Budget usage — Visual progress bars showing how much of the agent’s per-execution, per-hour, and per-day token budgets have been consumed
- Circuit breaker status — Whether the circuit breaker is closed (normal), open (tripped due to errors), or half-open (testing after a cooldown)
- Safety config — Context guard settings and SDK retry configuration
- Skills list — All skills and tools available to this agent
- Routing bindings — Which channels and routing rules direct messages to this agent
Agent Editor
The editor organizes agent settings into 13 sections using an accordion layout. The first section is always visible; the remaining 12 are collapsible:| # | Section | Collapsible | What It Configures |
|---|---|---|---|
| 1 | Essential | No (always visible) | Agent ID, name, provider, model, system prompt, and core identity settings |
| 2 | Budget | Yes | Per-execution, per-hour, and per-day token budget limits |
| 3 | Session Policy | Yes | Session reset schedule, idle timeout, and session key format |
| 4 | Skills | Yes | Which skills and tools are available to this agent |
| 5 | Heartbeat | Yes | Periodic heartbeat interval and prompt |
| 6 | Advanced | Yes | Max steps, retry behavior, model failover chain, RAG settings |
| 7 | Context Engine | Yes | Context assembly layers, compaction strategy, and token budget allocation |
| 8 | Streaming (System-Wide) | Yes | Streaming response configuration (applies to all agents) |
| 9 | Delivery (System-Wide) | Yes | Message delivery timing and batching (applies to all agents) |
| 10 | Queue / Overflow (System-Wide) | Yes | Queue depth limits and overflow recovery (applies to all agents) |
| 11 | Auto-Reply (System-Wide) | Yes | Automatic reply messages when agents are busy or unavailable (applies to all agents) |
| 12 | Send Policy (System-Wide) | Yes | Message send rate limits and channel-specific policies (applies to all agents) |
| 13 | Log Levels (Runtime) | Yes | Per-module log level overrides for debugging (applies at runtime) |
Sections marked (System-Wide) affect all agents, not just the one being edited. Changes to these sections are applied globally.
Common Tasks
View Agent Details
Suspend an Agent
Find the agent
In the agent list, locate the agent you want to suspend. You can also open the agent detail page.
Click Suspend
Click the Suspend button. The agent immediately stops processing new messages. Its status tag changes to yellow.
Edit Agent Configuration
Make your changes
Expand the accordion sections to find the settings you want to change. Modify the fields as needed.
Delete an Agent
Open the Workspace Manager
About the circuit breaker: If an agent hits too many consecutive errors, the circuit breaker trips to prevent runaway costs. While open, the agent stops making AI provider calls. After a cooldown period, it enters a “half-open” state where it tests with a single request. If the test succeeds, the circuit breaker closes and the agent resumes normal operation.
Related Pages
Chat Console
Start a conversation with any of your agents.
Sessions View
Browse conversation histories for all agents.
Agents Documentation
Learn about agent configuration, routing, and behavior in depth.
Workspace Manager
Browse and edit workspace files, manage git history.