pantheon-agents 0.6.4

A framework for building distributed evolvable multi-agent system.

Official Site · Online App · Desktop App · Documents · Feedback

1 What is PantheonOS?

PantheonOS is an evolvable and privacy-preserving multi-agent framework designed to reconcile generality with domain specificity. PantheonOS automates, scales and accelerates data sciences, with a particular focus on end-to-end single cell biology analyses. Cruicially, PantheonOS leverages agentic code evolution to optimize state-of-the-art and nascent algorithms, achieving super-human performance on specialized scientific tasks.

Key Highlights

• Evolvable — Pantheon-Evolve module enables agents to improve algorithms and code through genetic-algorithm-driven agentic code evolution
• Multi-Agent Teams — PantheonTeam, Sequential, Swarm, Mixture-of-Agents (MoA), and AgentAsTool team patterns for flexible orchestration
• Distributed Architecture — NATS-based messaging for scalable, fault-tolerant deployments across machines
• Friendly Interfaces — Interactive CLI (pantheon cli) and Chatroom UI (pantheon ui)
• Pantheon Store — A community marketplace with 1,000+ curated agents, teams, and skills for biomedical AI, installable via UI or CLI

2 Quick Start & Community

	Play with our Web UI!
	Download the latest desktop app release.
	Browse 1,000+ agents, teams, and skills!
	From source: pip install git+https://github.com/aristoteleo/PantheonOS.git (why not PyPI?)
	Join our Slack community!
	Join our Discord community!

3 Installation

[!WARNING] Security notice (June 2026) — install from source, not PyPI. The PyPI account that published pantheon-agents was compromised in the "Hades" supply-chain attack, and the PyPI releases pantheon-agents 0.6.1 and 0.6.2 are trojanized (a *-setup.pth hook downloads the Bun runtime and runs a credential stealer). Do not pip install pantheon-agents from PyPI. Install from the GitHub source as shown below — the source repository is clean. If you previously installed 0.6.1 or 0.6.2, uninstall it immediately and rotate any credentials that were present on that machine. PyPI distribution will resume once the account is restored.

Desktop App

Download the latest Pantheon Desktop app:

• Desktop release page

Open the release page to download the latest version for your platform.

Using uv (Recommended)

uv is a fast Python package manager that handles dependencies efficiently.

# Install uv (if not already installed)
# macOS/Linux
curl -LsSf https://astral.sh/uv/install.sh | sh
# Windows
powershell -c "irm https://astral.sh/uv/install.ps1 | iex"

# Clone and install
git clone https://github.com/aristoteleo/PantheonOS.git
cd PantheonOS
uv sync

# With optional dependencies
uv sync --extra knowledge  # RAG/vector search support
uv sync --extra claw       # PantheonClaw mobile gateway channels
uv sync --extra r          # R language support (requires R installed)

Using pip (from GitHub source)

PyPI installs are temporarily disabled (see the security notice above). Install directly from the clean GitHub source instead:

# Basic installation (from source)
pip install "git+https://github.com/aristoteleo/PantheonOS.git"

# With optional dependencies (e.g. RAG / vector search)
pip install "pantheon-agents[knowledge] @ git+https://github.com/aristoteleo/PantheonOS.git"

Development Installation

git clone https://github.com/aristoteleo/PantheonOS.git
cd PantheonOS
uv sync --extra dev --extra knowledge

# Run tests
uv run pytest tests/

Using Docker

Run Pantheon in a containerized environment with everything pre-configured:

# Pull the image
docker pull nanguage/pantheon-agents:latest

# Run in standalone mode (for local use)
docker run -it --rm \
  -e PANTHEON_MODE=standalone \
  -v $(pwd)/workspace:/workspace \
  -p 8080:8080 \
  nanguage/pantheon-agents:latest

After startup, copy the displayed connection URL to your browser and start using Pantheon. The Docker image reads factory templates directly from the running image, so startup does not copy templates into /workspace or container home. User-created project overrides still persist under the mounted /workspace/.pantheon/ directory.

For detailed Docker usage, deployment modes, and configuration options, see the Docker documentation.

4 Usage

CLI Mode

# Start the interactive REPL
pantheon cli

Chatroom UI

# Start the multi-agent chatroom
pantheon ui --auto-start-nats --auto-ui

API Usage

Please refer to our Documents for detailed API usage, including creating agents, using toolsets, and building teams.

5 Replayable Trajectories

Interactive replays of real multi-agent Pantheon sessions. Click any card to open the session step-by-step in the Pantheon UI — see every message, every file, every tool call.

🐭 E6 Mouse Embryo Tangram + 3D Visualization Run Tangram deconvolution on E6.0 mouse embryo binned spatial transcriptomics (51,711 spots across 6 embryos, 16,732 shared genes) using the TOME atlas reference subset to timepoints 5.5/6.25 and 5 cell types (Epi, ExE, EmVE, PE, ExVE).	🧬 Immune Oncology Gene Panel Design Design a 1000-plex human immune oncology gene profiling panel using multi-agent collaboration.
🫀 3D Human Fetal Heart Disease Gene Pattern Map congenital heart disease genes onto a 3D PCW12 human fetal heart using MERFISH spatial transcriptomics and scRNA-seq.	🗺️ ATAC Spatial Mapping via MOSCOT Spatially map fetal heart snATAC-seq onto MERFISH 3D coordinates using MOSCOT optimal transport.
🧬 Spatial Ligand-Receptor Disease Analysis Run spatial ligand-receptor (CCI) interaction analysis on MOSCOT-imputed disease gene data using the Spateo spatial_cci skill.	🫀 Heart MERFISH Data Overview Exploratory data overview of fetal heart MERFISH spatial transcriptomics.
🧬 Multi-Channel Cell Segmentation Compare cell segmentation tools on a 3-channel embryo image (membrane / DAPI / brightfield).	More trajectories coming soon — browse the full dataset on HuggingFace →

6 Contributing

Contributions of all types are more than welcome! Whether it's publishing skills to the Pantheon Store or contributing code, feel free to check out our GitHub Issues to dive in.

License

This project is BSD 2-Clause licensed.

Copyright © 2026 Qiu Lab.