How to Use ArchonHQ
Day-to-day usage guide for managing agents, tasks, and AI routing in ArchonHQ.
How to Use ArchonHQ
A practical guide to the daily workflow, managing tasks, working with agents, monitoring costs, and keeping work moving.
The dashboard layout
The main dashboard has four areas:
Kanban board (centre): your work, organised into columns: Backlog → In Progress → Review → Done. Tasks move left to right as work progresses.
Activity feed (right panel): a real-time log of every task mutation, agent action, and system event. Collapse it with the arrow if you need more board space.
Agent sidebar: shows connected agents, their current status, and session cost. Appears when agents are active.
Top nav: search, filters, notifications, and settings.
Managing tasks
Creating tasks
Press N or click the + button in any column. Every task has:
- Title: required
- Priority: Critical, High, Medium, Low
- Goal: links the task to a project goal for filtering
- Agent: assigns ownership to a specific agent
- Labels: free-form tags for custom filtering
- Description: markdown-supported notes
Moving tasks
Drag cards between columns. The board enforces WIP limits, if a column is at capacity, it highlights and blocks the drop. Adjust WIP limits in board settings.
The four columns and their meaning:
- Backlog — work not yet started; default landing zone for new tasks
- In Progress — actively being worked on by an agent or human
- Review — card is complete and awaiting human sign-off before closing
- Done — work accepted and closed
You can also change status from the card menu (⋯) without dragging.
Filtering and search
Use the filter bar above the board to narrow by:
- Search text (matches title + description)
- Priority
- Goal
- Assigned agent
- Labels
Filters combine with AND logic. Clear all filters with the × button.
Task detail and timeline
Click any card to open its detail view. You'll see:
- Full description (editable)
- Complete activity timeline, every status change, edit, and comment with timestamps
- Assigned agent and goal
- Creation and last-modified dates
Working with agents
Viewing active agents
The Agents tab shows all agents currently connected via the gateway. For each agent:
- Current status (idle, working, blocked)
- Session start time
- Token usage and estimated cost for the session
- Last action
Agent cost tracking
The Agents tab includes a cost chart. Costs are reported by agents when they complete API calls. If you're using AiPipe, costs are tracked per model and visible in the Router tab.
The Router tab
Shows AiPipe routing statistics:
- Requests per provider (OpenAI, Anthropic, etc.)
- Model distribution, what percentage of requests went to each model
- Total cost and per-model cost breakdown
- Success rate per provider
- Queue depth (real-time)
Use this to understand where your LLM spend is going and whether routing is working as expected.
Monitoring activity
Activity feed
Every event that happens in your workspace is logged in the activity feed:
- Task created / updated / moved / deleted
- Agent connected / disconnected
- Comments added
- API calls made
Events are timestamped and attributed (agent name or "you").
Roadmap view
The Roadmap tab shows your goals and the tasks attached to them. Drag tasks between goals or mark goals as delivered.
Keyboard shortcuts
| Action | Shortcut |
|---|---|
| New task | N |
| Search | / |
| Close modal | Esc |
| Move task left | ← (on focused card) |
| Move task right | → (on focused card) |
Tips
Keep WIP tight. The default WIP limit is 3 per column. Resist the urge to raise it, it's there to prevent context-switching overload.
Use goals for filtering. If you're running multiple projects, assign every task to a goal. The goal filter makes it trivial to switch context between projects.
Let agents create their own tasks. Agents with API access can POST to /api/tasks directly. You review what they've created on the board rather than micromanaging their backlog.
Check the Router tab weekly. If most of your requests are going to gpt-4o for simple tasks, it means some agents aren't sending enough context for the scorer to classify correctly. A short system prompt describing task complexity helps routing accuracy.