Trajectories

The Trajectories tab provides a detailed viewer for your agent's LLM call history. Every call the agent makes to a language model is recorded as a trajectory, giving you full visibility into prompts, responses, token usage, and performance.

Overview

Access Trajectories from the Advanced section of the dashboard at /trajectories. The viewer displays a reverse-chronological list of all LLM interactions.

What's Recorded

Each trajectory entry captures:

Field	Description
Timestamp	When the LLM call was made
Model	Which model was used (e.g., GPT-4o, Claude Sonnet 4.6)
Prompt	The full prompt sent to the model, including system message and conversation history
Response	The model's complete response
Token usage	Input tokens, output tokens, and total
Latency	Time taken for the call to complete
Status	Success or error state

Using the Viewer

The trajectory list shows a summary card for each call. Click any entry to open the detail view, which displays the full prompt and response in a readable format.

Filtering

Filter trajectories by:

• Model — narrow down to calls made to a specific model
• Time range — view calls from a specific period
• Status — filter by success or error

Detail View

The detail view for an individual trajectory shows:

• Full prompt with system message, conversation history, and any tool calls
• Complete model response
• Token breakdown (input/output/total)
• Latency and timing information
• Error details if the call failed

API Reference

Method	Endpoint	Description
GET	/api/trajectories	List and search trajectory entries with filters
GET	/api/trajectories/:id	Get trajectory details with LLM calls and provider accesses
GET	/api/trajectories/stats	Get aggregate trajectory statistics
GET	/api/trajectories/config	Get trajectory logging configuration (enabled/disabled)
PUT	/api/trajectories/config	Enable or disable trajectory logging
POST	/api/trajectories/export	Export trajectories (JSON, CSV, ART, or ZIP)
DELETE	/api/trajectories	Delete trajectories (by IDs or all)

See the REST API Reference for full endpoint documentation.