Running & Debugging

This page covers execution modes, plan progression, token tracking, and debugging techniques in Studio.

Execution Modes

Chat Mode

In Chat, the agent executes interactively. Each message triggers a single agent turn that may include multiple tool calls. The agent runs until it produces a final text response, then waits for your next input.

Flow Mode

Flows execute nodes in sequence following their connections. Execution begins at the Start node and progresses through connections until reaching the End node or encountering an error. Input nodes pause execution and prompt for user input.

Plan Auto-Progression

When running fleet plans, Studio supports auto-progression:

• The hub agent automatically assigns the next task when a spoke completes
• Tasks with satisfied dependencies begin without manual intervention
• Human checkpoints pause progression and prompt for approval in the UI

Toggle auto-progression in the plan settings. When disabled, each task transition requires manual approval.

Token Usage Tracking

Studio tracks token consumption per session:

• Input tokens — Tokens sent to the model (prompts, context)
• Output tokens — Tokens generated by the model
• Total cost — Estimated cost based on provider pricing

View usage in the session header or in Settings for aggregated stats.

Debugging Tips

Inspecting Tool Calls

Click any tool call in the chat to expand its details:

• Full input parameters (formatted JSON)
• Raw output
• Execution duration
• Error messages if failed

Flow Execution

During flow execution, the canvas highlights the currently active node. The chat panel shows:

• Agent responses for each step
• Tool call results
• Input prompts when the flow needs user input
• Error details when a node fails

Daemon Logs

For deeper debugging, check the daemon logs:

# Follow live logs
astonish daemon logs -f

Or run the daemon in foreground mode for direct stdout output:

astonish daemon run

Common Issues

Symptom	Likely Cause
Tool call hangs	External service timeout — check MCP server status
Empty response	Context window exceeded — start a new session
Wrong model	Check the Chat/App Model control (or Settings → Providers for team defaults)
Flow node fails	Inspect node output in the chat panel for error details