Kanban Dashboard

The web UI includes two workflow-oriented areas:

• Kanban for managing queued agent tasks across stages.
• Dashboard for observing completed and running agent runs across sessions and tasks.

Kanban overview

Open Kanban from the web UI header. The board stores task cards for the current workspace and displays them by stage.

Each task contains:

Field	Purpose
Title	Short task label shown on the card.
Prompt	The user prompt sent when the task runs.
Image attachments	Optional PNG, JPEG, WEBP, HEIC, or HEIF images sent with the initial task prompt.
Stage	Current board stage.
Profile override	Optional model profile for this task. If empty, the stage/default runtime is used.

You can add, edit, delete, drag, and run tasks from the board.

Creating tasks from the CLI

Use pbi-agent kanban create when you want to add a task from automation, scripts, or a terminal workflow:

pbi-agent kanban create \
  --title "Refactor API endpoint" \
  --desc "Improve and optimize the main API endpoint for faster response time." \
  --lane "In Progress"

The command creates a card on the current workspace board and exits without starting an agent run. Omit --lane to use the first board stage, normally Backlog. When provided, --lane (also available as --stage or --state) must match an existing stage ID, stage name, or slugified stage name. Configure custom stages in the web UI before targeting them from scripts.

For automation, add --json to print the created task ID, stage, timestamps, and related fields as JSON.

Use pbi-agent kanban list to inspect board tasks without opening the web UI:

pbi-agent kanban list
pbi-agent kanban list --stage "In Progress"
pbi-agent kanban list --json

The list command defaults to all stages. Its --stage filter (also available as --lane or --state) must match an existing stage ID, stage name, or slugified stage name. Human output includes each task's ID, title, full prompt, stage, position, project directory, associated session/profile IDs, run status, timestamps, last result summary, and image attachment count. JSON output returns an array with the same task fields and image attachment metadata.

Board stages

Use Edit Stages to configure the workflow. The board always has two fixed stages:

Stage	Behavior
Backlog	Stays first and never runs directly unless there is a runnable stage to move into.
Done	Stays last and is archive-only.

Custom stages can be placed between Backlog and Done. Each custom stage can define:

Stage option	Purpose
Name	Label shown as the column title.
Profile	Default model profile for tasks in that stage.
Command	Default project command to apply when tasks enter or run in that stage.
Auto-start	Automatically starts a task when it moves into the stage.

Stage profile and command references are validated against the current saved model profiles and project commands.

Running tasks

Click Run on a task card to start an agent run for that task.

Profile selection follows this order:

1. Task profile override.
2. Stage profile.
3. Active default model profile.
4. CLI/env/provider defaults if no saved profile applies.

Command selection follows the stage configuration. The first runnable stage sends the slash command plus task details; later stages can run command-only handoffs when configured. Project commands come from .agents/commands/*.md and are documented in Session Commands.

Task image attachments are included only with the initial full task prompt. Later continuation or command-only stage handoffs do not resend the original images. The selected task/stage/default runtime must support image inputs for attached-image tasks to run.

If the board only has Backlog and Done, pbi-agent prompts you to create a runnable stage before starting backlog tasks.

Moving tasks

Drag tasks between stages to update their status. If a destination stage has Auto-start enabled, pbi-agent starts the task automatically after it enters that stage.

The Done column sorts newest-first so recently completed tasks stay visible near the top.

Dashboard overview

Open Dashboard from the web UI header to inspect observability data. The dashboard includes:

• Date-range filters.
• Workspace/global scope toggle.
• KPI cards for run totals, token usage, cost, duration, and errors.
• Provider/model breakdown table.
• All Runs table with status, provider, model, duration, cost, token counts, and errors.
• Run detail modal with recorded request/tool events.

Workspace vs global scope

Scope	Meaning
Workspace	Shows runs associated with the current workspace.
Global	Shows runs across all stored pbi-agent workspaces.

Use global scope when comparing cost or reliability across projects. Use workspace scope when debugging the current repository.

Run details

The All Runs table opens a detail modal for a selected run. Use it to inspect:

• Run status and duration.
• Provider, model, profile, and session metadata.
• Request/response observability events.
• Tool calls, inputs, outputs, and failures.
• Token and estimated cost accounting.

Session pages also show a per-session run history, which is useful when you want the run list next to the conversation that produced it.

Recommended setup

1. Create at least one model profile.
2. Add project commands under .agents/commands/, such as /plan, /execute, or /review.
3. Open Kanban and create custom stages like Ready, Implement, and Review.
4. Assign a profile and command to each runnable stage.
5. Enable Auto-start only for stages where automatic handoff is safe.
6. Use Dashboard to review cost, failures, and tool behavior after runs complete.

Related pages

• Web UI
• Model Profiles
• Session Commands
• Project Commands