🌠 AgentComet
Build once. Run anywhere. Resume anytime.
Stateful. Portable. Composable AI agents.
Official Website: ac.defaultloop.com
⚡ The Problem
Every AI team ends up rebuilding the same agents.
Same logic. Same tools. Different environment. New framework.
And the worst part? All the memory, context, and progress is lost. Agents today are stateless, fragile, and non-portable.
🚀 Introducing AgentComet
AgentComet is a modern agent framework designed to make agents:
- Stateful by default
- Portable across environments
- Resumable at any point in time
Instead of treating agents like scripts, AgentComet treats them like artifacts.
🎯 What makes AgentComet different?
AgentComet introduces a new primitive: Portable, Stateful Agent Artifacts (.uaf)
You don't just write agents. You package, move, version, and resume them.
🚀 Quick Start
Installation
pip install agentcomet
Build an Agent
Create an agent, assign it custom tools, and let it handle the heavy lifting.
from agentcomet import Agent, tool
from agentcomet.models import Ollama
@tool
def greet(name: str) -> str:
return f"Hello, {name}!"
class GreeterAgent(Agent):
def setup(self):
self.name = "greeter"
self.add_tools(greet)
# Instantiate and chat
agent = GreeterAgent(llm=Ollama(model="llama3"))
print(agent.run("Say hello to Alice!"))
✨ Core Capabilities
🧠 Persistent Memory
Every agent comes with a highly flexible key-value memory store (self.memory). Conversational history is auto-saved without any manual boilerplate.
agent.memory.save("user_preferences", {"theme": "dark"})
print(agent.memory.get("messages"))
💾 State Checkpoints
Save and restore exact agent timelines. Need to rollback a prompt mistake? Just load the state.
agent.save_state("checkpoint-1")
agent.load_state("checkpoint-1")
📦 Portable Agents with .uaf
Stop writing boilerplate to serialize agent code. Export your entire agent—along with its memory, history, LLM config, and tools—into a portable .uaf archive.
agent.export("assistant.uaf")
from agentcomet import load_agent
agent = load_agent("assistant.uaf")
☁️ Registry & Hub (Push & Pull)
Sync your agents with either a locally hosted AgentComet Studio registry or the public cloud registry AgentComet Hub.
🏢 1. AgentComet Studio (Local Registry)
Sync your agents locally across your team.
A. Start AgentComet Studio
docker pull vaibhavhaswani/agentcomet-studio:latest
docker run -p 3451:3451 \
-v $(pwd)/data:/app/data \
vaibhavhaswani/agentcomet-studio:latest
Windows (PowerShell):
docker run -p 3451:3451 -v ${PWD}/data:/app/data vaibhavhaswani/agentcomet-studio:latest
B. Configure and Push/Pull
from agentcomet import Settings, Agent
# Configure SDK
Settings.init(
AGENTCOMET_URL="http://localhost:3451",
AGENTCOMET_KEY="your-local-key"
)
# Push — auto-increments version
agent.push(repo="assistant")
# Pull — latest or specific version
agent = Agent.pull(repo="assistant", version="latest")
🌌 2. AgentComet Hub (Cloud Registry)
Publish and distribute your agents on the official public registry at ac.defaultloop.com.
A. Configure the SDK
from agentcomet import Settings, Agent
# Configure SDK to target the cloud registry
Settings.init(
AGENTCOMET_URL="https://ac.defaultloop.com",
AGENTCOMET_KEY="your-hub-api-token"
)
B. Push & Pull
# Push to the cloud registry — auto-increments version and creates if it doesn't exist
agent.push(repo="username/assistant", create=True)
# Pull from the cloud registry
agent = Agent.pull(repo="username/assistant", version="latest")
🔌 Multi-Provider LLM Support
Switching your backend is as simple as defining a new class.
from agentcomet.models import OpenAIChat, Gemini, Anthropic, Ollama
agent.use_llm(OpenAIChat(model="gpt-4o"))
🧠 Why this matters
We believe the future of AI is not just models. It's agents that evolve over time.
Agents that:
- Learn
- Persist
- Move
- Continue
AgentComet is a step toward that future.
🛣️ Vision
This is just the beginning.
AgentComet is part of a larger vision to build:
- A universal agent format
- A global agent registry
- A composable ecosystem of agents
Where agents are not rewritten, but shared, reused, and extended.
📚 Documentation & Website
- Official Website: https://ac.defaultloop.com
- Comprehensive Docs: Refer to the detailed Usage Documentation for full API schemas, setup guides, and config settings.
🤝 Contributing
- ⭐ Star the repo
- 🛠️ Break things
- 🚀 Open PRs
All contributions are welcome.
📄 License & Legal
- License: Apache License 2.0
- Patent: Indian Provisional Patent Application No. 202611013684
- Branding: "AgentComet" is a trademark. See BRANDING.md.
Metadata
Release files for agentcomet 0.6.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| agentcomet-0.6.0.tar.gz | 29.2 kB | Without provenance | Details for agentcomet-0.6.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| agentcomet-0.6.0-py3-none-any.whl | Python 3 | none | any | Details for agentcomet-0.6.0-py3-none-any.whl |
Total release size: 58.4 kB
Release files / agentcomet-0.6.0.tar.gz
| Download URL | agentcomet-0.6.0.tar.gz |
|---|---|
| Size | 29.2 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
2b8ad1cf4873703f0d1238308feb32d7553fe9cff0ca426e6703d145f3a3fa9c
|
|
BLAKE2b-256 checksum How to use checksums |
243b8b562eae25e320d6b250f8f135af69f931dfb474d87173d07346df25c4e6
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.4
|
Release files / agentcomet-0.6.0-py3-none-any.whl
| Download URL | agentcomet-0.6.0-py3-none-any.whl |
|---|---|
| Size | 29.3 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
7b9f0376df51c7fa84d09c82e0cc8423cc8f744d90aaf2ffdafffa19e33b7299
|
|
BLAKE2b-256 checksum How to use checksums |
957cd8c4e5376077889702cd00a78555576870d247723986706bd1ddb3c51008
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.12.4
|
Security for agentcomet 0.6.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.