Agents

Create runtime workers, configure concurrency, and attach platforms with per-site settings.

An agent is a configured runtime worker that executes tasks in isolated browser sessions. Agents scale automatically based on workload and your concurrency settings.

Creating an agent

Open Agents in your workspace sidebar and click New agent.

Give the agent a name and optional description. After creation, configure:

Concurrency settings

Setting	Purpose
Min workers	Minimum number of execution pods kept warm
Max workers	Upper limit on concurrent execution pods
Idle timeout	How long an idle execution session stays alive before termination

Higher concurrency lets the agent process multiple executions in parallel. Set limits based on your target sites' rate limits and your infrastructure budget.

Attaching platforms

In the agent form, attach platforms the agent should access:

Select a platform from your workspace.
Configure the vault with credentials (username, password, API keys, TOTP seeds).
Set optional max concurrent tasks for this platform on this agent.
Enable captcha solving if the site uses CAPTCHAs (solve_captcha in DSL).

Each agent-platform binding is independent — two agents can use the same platform with different credentials.

Agent modes

Agents operate in different session modes:

Mode	When used
Execution	Running queued task executions (default for automation)
Assistant	Interactive AI chat at `/agents/{id}/chat` to help build workflows
Maintenance	Platform auth setup and diagnosis

Execution sessions are scaled by the platform's reconciler based on queued work and your min/max settings.

Running tasks on an agent

When you run a task — manually or via API — you specify which agent executes it. The agent:

Claims the execution from the queue
Spins up or reuses a browser session
Runs the task's DSL script against attached platforms
Streams live screenshots and records video
Writes results and status back to the execution record

AI assistant

Each agent has a built-in AI assistant accessible from the agent detail page. Use it to:

Describe workflows conversationally and get DSL suggestions
Debug failing scripts with context about your platforms
Iterate on task logic without writing every line manually

The assistant runs in a dedicated assistant session with access to your agent's platform configuration.

Best practices

Create separate agents for different credential sets or environments (production vs. staging).
Set max_concurrent_tasks per platform to respect site rate limits.
Use idle timeout to balance session reuse (faster repeat runs) against resource cost.