computer-agents 2.6.2

Official Python SDK for the Computer Agents Agentic Compute Platform API. Build with threads, computers, agents, resources, and databases.

Computer Agents Python SDK

The official Python SDK for the Computer Agents Cloud API. Build against the Agentic Compute Platform with threads, computers, resources, databases, skills, and agents.

Installation

pip install computer-agents

Quick Start

from computer_agents import ComputerAgentsClient

client = ComputerAgentsClient(api_key="ca_your_api_key")

# Execute a task
result = client.run(
    "Create a REST API with Flask",
    computer_id="env_xxx",
    on_event=lambda e: print(e["type"]),
)

print(result.content)
print(f"Thread ID: {result.thread_id}")

Multi-Turn Conversations

# Create a thread for persistent conversation
thread = client.threads.create(environment_id="env_xxx")

# First message
result = client.threads.send_message(
    thread["id"],
    content="Create a Python web server",
    on_event=lambda e: print(e),
)

# Follow-up (continues same session)
result2 = client.threads.send_message(
    thread["id"],
    content="Add authentication to the server",
)

Computers

# Create a custom computer
computer = client.computers.create(
    name="data-science",
    internet_access=True,
    runtimes={"python": "3.12"},
)

# Install packages
client.computers.install_packages(computer["id"], "python", ["pandas", "numpy"])

# Add MCP servers
client.computers.update(
    computer["id"],
    mcp_servers=[{
        "type": "stdio",
        "name": "filesystem",
        "command": "npx",
        "args": ["@modelcontextprotocol/server-filesystem", "/workspace"],
    }],
)

Supported Models

Model	ID	Use Case
Claude 4.7 Opus	claude-opus-4-7	Latest Anthropic flagship, hardest coding and reasoning tasks
Claude 4.6 Opus	claude-opus-4-6	Most capable, complex tasks
Claude 4.5 Sonnet	claude-sonnet-4-5	Balanced Claude option
Claude 4.5 Haiku	claude-haiku-4-5	Fast, efficient
GPT-5.5 Pro	gpt-5.5-pro	Highest-accuracy OpenAI model for the hardest professional and agentic work
GPT-5.5	gpt-5.5	OpenAI frontier model for coding, professional work, and long-context agents
GPT-5.4	gpt-5.4	OpenAI flagship for advanced coding and planning
GPT-5.4 mini	gpt-5.4-mini	Fast OpenAI coding, subagents, and computer use
GPT-5.4 nano	gpt-5.4-nano	Lightweight OpenAI model for simple, high-volume workflows
Gemini 3 Flash	gemini-3-flash	Low-latency general workflows
Gemini 3.1 Flash	gemini-3-1-flash	Alias tier that currently routes through the Gemini 3 Flash runtime
Gemini 3.1 Pro	gemini-3-1-pro	Broader reasoning and research tasks
DeepSeek V4 Pro	deepseek-v4-pro	DeepSeek flagship for agentic coding and long-context reasoning
DeepSeek V4 Flash	deepseek-v4-flash	Fast, efficient DeepSeek model for agent execution
Kimi K2.6	kimi-k2.6	Moonshot flagship via Cloudflare Workers AI for long-horizon coding and tool use

Built-in system agents default to DeepSeek V4 Pro, while the Developer system agent defaults to Kimi K2.6.

Thread Feedback and Permissions

summary = client.threads.set_feedback(thread["id"], "up")

client.threads.report_issue(
    thread["id"],
    report_type="bug",
    message="The thread missed the requested test run.",
)

requests = client.threads.list_permission_requests(thread["id"])
if requests:
    client.threads.approve_permission_request(thread["id"], requests[0]["requestId"])

Tasks

task = client.tasks.create(
    "Add login audit logs",
    project_id="proj_xxx",
    status="todo",
    priority="high",
)

client.tasks.create_comment(
    task["id"],
    body="Include success and failure cases.",
)

started = client.tasks.start_thread(
    task["id"],
    environment_id="env_xxx",
)

result = client.tasks.run_thread(
    task["id"],
    environment_id="env_xxx",
    message="Implement the ticket and summarize changed files.",
)

Computer Change History

history = client.computers.list_changes(
    computer["id"],
    limit=25,
    project_id="proj_123",
    operation=["created", "modified"],
)

latest = history["data"][0]
latest_file = latest["files"][0]

diff = client.computers.get_change_diff(
    computer["id"],
    latest["id"],
    path=latest_file["path"],
)

file_state = client.computers.get_change_file(
    computer["id"],
    latest["id"],
    path=latest_file["path"],
)

fork = client.computers.fork_from_change(
    computer["id"],
    latest["id"],
    name="history-branch",
)

Agents

# Create a custom agent
agent = client.agents.create(
    name="Code Reviewer",
    model="claude-sonnet-4-5",
    instructions="You are a thorough code reviewer.",
    enabled_skills=["web_search"],
)

# Use the agent
thread = client.threads.create(
    computer_id="env_xxx",
    agent_id=agent["id"],
)

Files

# Upload a file
client.files.upload_file(
    "env_xxx",
    filename="app.py",
    content='print("hello")',
    path="src",
)

# Download a file
content = client.files.get_file("env_xxx", "src/app.py")

# Download a folder as a zip archive
zip_bytes = client.files.download_directory("env_xxx", "src")

# List files
files = client.files.list_files("env_xxx")

Git Operations

diff = client.git.diff("env_xxx")
client.git.commit("env_xxx", message="Update feature")
client.git.push("env_xxx")

Schedules

schedule = client.schedules.create(
    name="Daily Report",
    agent_id="agent_xxx",
    agent_name="Reporter",
    task="Generate daily report",
    schedule_type="recurring",
    cron_expression="0 9 * * *",
)

Resources

resource = client.resources.create(
    name="crm-web",
    kind="web_app",
    auth_mode="public",
)

client.resources.deploy(resource["id"])
analytics = client.resources.get_analytics(resource["id"])

Databases

database = client.databases.create(name="crm-data")

client.databases.create_collection(database["id"], name="leads")
client.databases.create_document(
    database["id"],
    "leads",
    data={"company": "Acme", "stage": "new"},
)

Skills

skills = client.skills.list()
print([skill["name"] for skill in skills])

Configuration

The API key can be provided via:

1. Constructor argument: ComputerAgentsClient(api_key="ca_...")
2. Environment variable: COMPUTER_AGENTS_API_KEY

# Custom base URL and timeout
client = ComputerAgentsClient(
    api_key="ca_...",
    base_url="https://custom-api.example.com",
    timeout=120.0,
    debug=True,
)

Context Manager

with ComputerAgentsClient(api_key="ca_...") as client:
    result = client.run("Hello world", computer_id="env_xxx")
    print(result.content)
# Client is automatically closed

API Resources

Resource	Description
client.threads	Conversation management with SSE streaming
client.environments / client.computers	Computer configuration and lifecycle
client.agents	Agent configuration (model, instructions, skills)
client.resources	Web apps, functions, auth modules, runtimes, and secret vaults
client.web_apps / client.functions / client.auth / client.runtimes / client.secrets	Product-shaped server resource managers
client.databases	Managed database surfaces
client.skills	Custom ACP skills
client.files	File management in computer workspaces
client.tasks	Planning tasks, comments, releases, sprints, and task-linked threads
client.schedules	Scheduled task management
client.triggers	Event-driven triggers
client.orchestrations	Agent-to-agent orchestration
client.budget	Budget and usage tracking
client.billing	Billing records and statistics
client.git	Git operations on computer workspaces
client.notifications	In-app notifications and push token registration

Requirements

• Python >= 3.9
• httpx >= 0.25.0

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

MIT - see LICENSE for details.