Skip to main content

Live Run Viewer

Route: /projects/:projectId/issues/:issueId/runs/:runId/live

Watch agent runs execute in real time. See the exact moment an agent thinks, calls tools, and produces output.

Overview​

Live Run Viewer shows a live stream of events as an agent processes a card. Use it to:

  • Debug agent behavior
  • See tool invocations in real time
  • Monitor token usage and cost
  • Troubleshoot failures
  • Understand agent reasoning

How to Get Here​

From any of these pages, click a run to open Live Run Viewer:

  • Board → Card Detail → Runs tab — Click a run
  • Now dashboard → Issue Runs — Click a run
  • Ceremonies → Run history — Click a run
  • Dashboard → Recent runs — Click a run

Or navigate directly with the URL above.

Page Sections​

Shows run context:

  • Issue title — The card/work item being processed
  • Agent name — Which agent is running
  • Status badge — Running, completed, failed, etc.
  • Timestamp — When run started
  • Close button — Exit Live Run Viewer

Metrics Row​

Real-time counters (update as run progresses):

  • Elapsed time — How long run has been executing
  • Input tokens — Tokens sent to model
  • Output tokens — Tokens received from model
  • Total tokens — Input + output
  • Cost — USD spent on this run so far
  • Turns — Number of back-and-forth interactions with model

Connection Status​

If run is live:

  • Green indicator: "Connected"
  • Shows real-time events streaming

If connection lost:

  • Yellow indicator: "Reconnecting…"
  • Waiting to resume event stream
  • Auto-retries every few seconds

If run completed or failed:

  • Gray indicator: "Finished"
  • Event stream is complete

Event Stream​

Chronological log of everything that happened during the run:

Event types you'll see:

  1. Agent started — Run initialized
  2. Prompt sent — Full system prompt + user message sent to model
  3. Model response — Model's reply (streamed, shows as it arrives)
  4. Tool invocation — Agent called a tool (name, input params shown)
  5. Tool result — Tool output returned to agent
  6. Agent thinking — Optional reasoning/thought process
  7. Run completed — Success with final output
  8. Run failed — Error occurred; reason shown

Each event shows:

  • Timestamp
  • Event type (icon)
  • Short description
  • Full details (expandable)

Error Recovery Panel​

If run failed:

  • Error type — What went wrong (e.g., "Tool not found", "Model error", "Permission denied")
  • Error message — Human-readable explanation
  • Remediation — Suggested fixes (e.g., "Check tool config" or "Re-check authentication")

Actions:

  • Retrigger run — Click to run the same card again with current agent config
  • Retry stream — Re-attempt to read event log (if connection issue)

Steer Bar (While Live)​

If run is still executing, bottom bar shows:

Send steering message:

  • Text input field
  • Send button
  • Used to guide agent (e.g., "Try a different approach" or "Check the error log")

Available only while run is actively executing. After completion, steer bar disappears.

Actions​

Retrigger run

  • Click "Retrigger" button to run the same card again
  • New run will be created; you'll be redirected to new Live Run Viewer

Retry stream

  • If events stop arriving but run is still processing
  • Click "Retry stream" to reconnect to event feed

Open work item

  • Click "Open work item" breadcrumb
  • Jumps to Board with card detail open

View run history

  • Click "Runs" link
  • Opens Ceremony Runs or Board card runs history

Navigate to board

  • Click "Back to board" or project badge
  • Returns to Board view

Navigate to project

Tips​

Watch for slow steps — If any step takes much longer than expected, that's where to investigate. Click to expand and see full context.

Monitor token usage — Real-time token counter shows efficiency. High tokens = high cost.

Tool invocation details — Expand tool events to see exactly what parameters were passed and what was returned. Useful for debugging.

Error context — If run fails, the error panel provides remediation hints. Most common issues are tool configuration or authentication.

Streaming responses — Model output streams as it's generated; you see it in real time as LLM produces tokens.

Troubleshooting​

"Reconnecting…" banner is stuck

  • Run may have timed out or server connection dropped
  • Click "Retry stream" to attempt reconnection
  • If still stuck, close and refresh the page

Events seem incomplete

  • WebSocket may have dropped mid-run
  • Scroll to end of event log to see latest events
  • Close and reopen to refresh

Metrics not updating

  • Stale data; refresh the page
  • If run is complete, metrics show final totals

Retrigger button disabled

  • Only enabled after run completes or fails
  • While run is executing, retrigger is not available
  • Board — Manage cards and open runs
  • Costs — See aggregated costs across runs
  • Agents — View/configure agent that's running
  • Ceremonies — View ceremony/workflow runs
  • Dashboard — See run metrics aggregated
Last updated on