Fleet SDK

The Fleet Python SDK provides programmatic access to Fleet's environment infrastructure.

Installation

pip install fleet-python

API Key Setup

Get your API key from the Fleet Dashboard, then set it as an environment variable:

export FLEET_API_KEY="sk_your_key_here"

Tasks

A Task represents a unit of work for an agent to complete within an environment. Each task combines:

• Prompt: Instructions describing what needs to be done
• Environment: The environment configuration (env_key, data_key, env_variables)
• Verifier: Code that validates task completion and returns a score (0.0 to 1.0)

Tasks provide all the configuration needed to spin up an environment and verify an agent's work.

Task Properties

Property	Description
key	Unique task identifier
prompt	Instructions for the agent
env_key	Environment identifier (e.g., "hubspot:v1.2")
data_key	Data configuration identifier
env_variables	Environment variables for the task
metadata	Additional info (e.g., avg_steps for task difficulty)

Quick Start

import fleet

async def main():
    # Load a task
    tasks = await fleet.load_tasks_async(
        keys=["task_abcdef"]
    )
    task = tasks[0]

    # Create an environment from the task
    env = await fleet.env.make_async(
        env_key=task.env_key,
        data_key=task.data_key,
        env_variables=task.env_variables,
        ttl_seconds=7200,
        run_id="run-123",
    )

    # Access the environment URL
    print(env.urls.app[0])

    # ... interact with the environment ...

    # Verify task completion
    result = await task.verify_detailed_async(env.instance_id)
    print(result)

    # Clean up
    await env.close()

Loading Tasks

By Task Keys

tasks = await fleet.load_tasks_async(
    keys=["task_abcdef"]
)

By Project Key

tasks = await fleet.load_tasks_async(project_key="my-project")

Creating Environments

env = await fleet.env.make_async(
    env_key=task.env_key,
    data_key=task.data_key,
    env_variables=task.env_variables,
    ttl_seconds=7200,
    run_id="run-123",
)

With Heartbeats (Optional)

Optionally enable heartbeats to keep environments alive during long-running operations:

env = await fleet.env.make_async(
    env_key=task.env_key,
    data_key=task.data_key,
    env_variables=task.env_variables,
    ttl_seconds=10800,
    heartbeat_interval=30,  # seconds
)

Send heartbeats to keep the environment alive:

# Via the environment object
await env.heartbeat()

# Or via instance ID
await fleet.env.heartbeat_async(instance_id)

Heartbeats are optional. If heartbeat_interval is not set, the instance lifetime is controlled solely by ttl_seconds. If heartbeats are enabled and missed 3 consecutive times, the instance will be terminated. Heartbeats take precedence over the TTL.

Instance Management

List Instances

# List all instances for a run
instances = await fleet.env.list_instances_async(run_id="run-123")

# List all instances for your profile
instances = await fleet.env.list_instances_async(profile_id="self")

Close Instances

# Close all instances for a run
await fleet.env.close_all_async(run_id="run-123")

# Close all instances for your profile
await fleet.env.close_all_async(profile_id="self")

# Close a specific instance by ID
await fleet.env.close_async("bc8954c2")

"self" is an alias for the profile associated with your FLEET_API_KEY.

Account Information

View your current account details including team info, instance limits, and profile ID:

account = await fleet.env.account_async()

Returns:

{
  "team_id": "aaaaaaaa-bbbb-cccc-dddd-eeeeeeeeeeee",
  "team_name": "My Team",
  "instance_limit": 32000,
  "instance_count": 924,
  "profile_id": "11111111-2222-3333-4444-555555555555",
  "profile_name": "Jane Doe"
}

Run Tracking

Track active and past runs:

# List active runs
runs = await fleet.env.list_runs_async()

# List all runs (active and inactive)
runs = await fleet.env.list_runs_async(status="all")

# Filter by profile
runs = await fleet.env.list_runs_async(profile_id="self")

Returns:

[
  {
    "run_id": "run-123",
    "running_count": 0,
    "total_count": 4,
    "first_created_at": "2025-10-24T09:48:47.152387",
    "last_created_at": "2025-10-24T09:55:19.284294",
    "profile_id": "11111111-2222-3333-4444-555555555555"
  }
]

Task Verification

Verify task completion and get detailed results:

result = await task.verify_detailed_async(env.instance_id)
print(result)

Returns:

{
  "key": "task_abcdef",
  "version": 4,
  "success": true,
  "result": 1.0,
  "error": null,
  "execution_time_ms": 2291,
  "stdout": ""
}

On failure, stdout contains detailed verification errors:

{
  "key": "task_abcdef",
  "version": 4,
  "success": true,
  "result": 0,
  "error": null,
  "execution_time_ms": 2291,
  "stdout": "Verification errors: [\"Expected field to be 'value', got None\", \"Form not marked as complete\"]"
}

Complete Example

import fleet
import asyncio

async def main():
    # Load tasks from a project
    tasks = await fleet.load_tasks_async(project_key="my-project")
    
    for task in tasks:
        # Create environment
        env = await fleet.env.make_async(
            env_key=task.env_key,
            data_key=task.data_key,
            env_variables=task.env_variables,
            ttl_seconds=7200,
            run_id="my-evaluation-run",
        )
        
        try:
            # Access the environment URL
            print(env.urls.app[0])
            
            # ... run your agent ...
            
            # Verify task completion
            result = await task.verify_detailed_async(env.instance_id)
            print(f"Task {task.key}: score={result['result']}")
            
        finally:
            await env.close()
    
    # Clean up all instances from this run
    await fleet.env.close_all_async(run_id="my-evaluation-run")

if __name__ == "__main__":
    asyncio.run(main())