Agents

Agents are autonomous browser-based workers that execute prompts against a target website. Each agent gets its own browser session, follows the instructions in its prompt, and reports back with results.

API Key

Paste your API key to use the Try it panels below. Don't have one? Create an account.

POST /api/v1/agents Create a new agent

Creates a new agent and enqueues it for execution. The agent will be assigned a browser session and begin following the prompt instructions.

Parameter	Type	In	Required	Description
`project_id`	uuid	body	Yes	ID of the project this agent belongs to
`persona_id`	uuid	body	No	Optional persona to use for this agent
`name`	string	body	No	Human-readable name for the agent
`prompt`	string	body	Yes	Instructions for the agent to follow
`model`	string	body	No	LLM model to use (default: gemini-3-flash-preview)
`browser_type`	string	body	No	Browser to use (chrome, firefox, edge) (default: chrome)
`max_tokens`	integer	body	No	Maximum token budget for the agent's LLM calls (default: 2000000)
`max_iterations`	integer	body	No	Maximum number of LLM iterations before the agent stops (default: 100)
`record_video`	boolean	body	No	Whether to record a video of the browser session (default: false)
`clean_session`	boolean	body	No	Clear all browser cookies before starting — prevents stale auth sessions from interfering (default: false)
`files`	array	body	No	Initial files to place in the agent workspace (each with path and base64 content)

project_id *

persona_id

name

prompt *

model

browser_type

max_tokens

max_iterations

record_video

clean_session

files

cURL

Response

GET /api/v1/agents List agents

Parameter	Type	In	Required	Description
`limit`	integer	query	No	Number of results to return (default: 20)
`cursor`	uuid	query	No	Cursor for pagination (agent ID)
`project`	uuid	query	No	Filter by project ID
`experiment`	uuid	query	No	Filter by experiment ID (all runs)
`experiment_run`	uuid	query	No	Filter by specific experiment run ID

limit

cursor

project

experiment

experiment_run

cURL

Response

GET /api/v1/agents/{id} Get agent by ID

Parameter	Type	In	Required	Description
`id`	uuid	path	Yes	Agent ID

id *

cURL

Response

DELETE /api/v1/agents/{id} Delete an agent

Parameter	Type	In	Required	Description
`id`	uuid	path	Yes	Agent ID

id *

cURL

Response

POST /api/v1/agents/{id}/summarize Generate agent summary

Uses an LLM to generate a summary of what the agent did. Only works on terminal agents (completed, failed, captcha).

Parameter	Type	In	Required	Description
`id`	uuid	path	Yes	Agent ID
`force`	boolean	query	No	Force regeneration of existing summary (default: false)

id *

force

cURL

Response

GET /api/v1/agents/{id}/verification Get agent verification result

Parameter	Type	In	Required	Description
`id`	uuid	path	Yes	Agent ID

id *

cURL

Response

POST /api/v1/agents/{id}/verify Verify an agent

Runs verification checks on the agent's outcome.

Parameter	Type	In	Required	Description
`id`	uuid	path	Yes	Agent ID

id *

cURL

Response

POST /api/v1/agents/{id}/retry Retry an agent

Re-enqueues a terminal agent for another execution attempt.

Parameter	Type	In	Required	Description
`id`	uuid	path	Yes	Agent ID

id *

cURL

Response

POST /api/v1/agents/{id}/files Upload files to agent workspace

Upload files via multipart/form-data. Files are placed in the agent's workspace for use with browser_upload_file during execution. Max 50MB per request, 10MB per file.

Parameter	Type	In	Required	Description
`id`	uuid	path	Yes	Agent ID
`path`	string	body	No	Subdirectory to place files in (e.g. 'audio')
`files`	file[]	body	Yes	One or more files (multipart form field name: 'files')

id *

path

files *

cURL

Response