Developer Reference

AgentComet SDK

The AgentComet SDK provides a unified interface for publishing and running AI agents. It is available as a Python package.

Installation

Python (PyPI)

bash

pip install agentcomet

Authentication

Set the Hub URL and API token as environment variables to authenticate your requests. You can find your token in the Settings page.

bash

export AGENTCOMET_URL="https://ac.defaultloop.com"
export AGENTCOMET_KEY="ac_live_..."

Discover Public Agents

List or search public AgentComet repositories without authentication. Use the returned cursor to request the next page.

bash

curl "https://ac.defaultloop.com/api/agents?q=research&category=RAG&sort=trending&limit=20"

Supported parameters: q, category, owner, sort, limit, and cursor.

Quick Start

Pushing an Agent

Package and upload your agent to the registry with its metadata and version.

python

from agentcomet import Settings, Agent

Settings.init(
    AGENTCOMET_URL="https://ac.defaultloop.com",
    AGENTCOMET_KEY="your-api-token"
)

agent = Agent(
    name="researcher",
    prompt="You are a research assistant..."
)

agent.push(repo="your-username/researcher", version="1.0.0", create=True)

Pulling & Running

Instantiate any public or private agent from the registry by its ID.

python

from agentcomet import Settings, Agent

Settings.init(
    AGENTCOMET_URL="https://ac.defaultloop.com",
    AGENTCOMET_KEY="your-api-token"
)

agent = Agent.pull(repo="your-username/researcher", version="latest")
response = agent.run("Analyze this data...")

Advanced Configuration

The SDK supports advanced features like multi-agent orchestration, custom tool integration, and secure state management.

Multi-Agent

Orchestrate multiple agents in a single workflow.

Toolsets

Define custom tools that agents can use.

Versioning

Roll back or pin specific agent versions.

Logging

Built-in observability for agent runs.

Read full documentation