Installation

Three steps: install the panel, choose an AI when the panel first opens, press Run a first request.

Pellaeon comes in two editions. Pick the one for the program you use.

Both use the same AI providers, the same panel and the same safety rules.

ChimeraX edition

Pellaeon is on the ChimeraX Toolshed, the bundle catalogue built into ChimeraX, so ChimeraX installs and updates it for you.

1. Install UCSF ChimeraX 1.9 or newer.
2. Start ChimeraX and install Pellaeon, from the menu or from the command line. From the menu: Tools ▸ More Tools… opens the Toolshed; find Pellaeon (category General) and press Install. From the command line: paste this into the Command: line at the bottom of the window and press Enter:

toolshed install ChimeraX-Pellaeon

3. Open Tools ▸ General ▸ Pellaeon. The panel opens as a floating window (drag it onto an edge to dock it) and asks you to choose an AI. The pellaeon command works on the ChimeraX command line too.

No terminal, no Python installation, no administrator rights. It works the same on Windows, macOS and Linux.

Straight from GitHub instead: this one line fetches the newest release from the releases page, installs it and opens the panel. Use it to pick up a release before it reaches the Toolshed.

open https://github.com/tggr-lab/pellaeon/releases/latest/download/install_pellaeon.py

Offline install: get the .whl file from the releases page and type (quotes needed when the path has spaces):

toolshed install "C:\Users\you\Downloads\chimerax_pellaeon-<version>-py3-none-any.whl"

<version> is a placeholder: type the name of the file you actually downloaded.

Uninstall: toolshed uninstall ChimeraX-Pellaeon in ChimeraX.

Classic edition (Chimera 1.x)

1. Install Python 3.9 or newer. On Windows tick Add Python to PATH in the installer.
2. Download pellaeon-classic.zip from the releases page and unzip it anywhere.
3. Windows: double-click Start Pellaeon Classic.cmd. macOS: double-click Start Pellaeon Classic.command. Linux: ./start.sh. A small launcher window opens and your browser shows the Pellaeon panel.
4. In the launcher press Launch Chimera (it starts Chimera with its REST server and connects by itself), or start Chimera yourself, open Tools ▸ Utilities ▸ RESTServer, type the port shown in the Reply Log into the launcher and press Test.
5. Optional: python install.py puts a Pellaeon Classic shortcut on your desktop and Start menu.

Choosing an AI

The first time the panel opens it shows provider cards. There is one decision to make: local or cloud. Local (Ollama) needs no account and no key, but needs a machine with a reasonable GPU and a one-time model download of a few GB. Cloud needs a key, or a ChatGPT Plus/Pro account.

If you have a ChatGPT Plus or Pro account, press Sign in with ChatGPT: no key, and its gpt-6-sol has the highest score in the feature comparison on Tested models. Otherwise start with Mistral: the free key takes two minutes at console.mistral.ai/api-keys, needs no credit card, and Pellaeon uses ministral-8b-latest on it. If you would rather not send anything to a provider, install Ollama and pull qwen3:8b instead.

Press Test connection, then Save & use or Run a first request (it saves the settings and has the AI open ubiquitin and color it by chain). Keys are stored privately on your computer (system keyring or a private file), never in ChimeraX sessions or files you share. Environment variables (MISTRAL_API_KEY, GOOGLE_API_KEY, ANTHROPIC_API_KEY, OPENAI_API_KEY) are picked up automatically.

Which model to run on the provider you picked, and how each one did on Pellaeon's own request sets, is on Tested models.

When a limit is hit Pellaeon waits and retries rather than failing, and on the tightest tiers it automatically shortens its prompt to fit. A quota that resets tomorrow is reported plainly instead of being retried.

Privacy

With a local Ollama server, model requests are processed on your computer. Fetching structures and annotations still contacts external databases (PDB, AlphaFold DB, UniProt, ClinVar, ConSurf-DB). With a cloud provider, your request, the list of open models, the current selection, the commands you ran by hand in ChimeraX since your previous request (with their output) and any table you loaded are sent to that provider. Screenshots are shared only when the review-the-view option is enabled in Settings.

The free tiers of Mistral and Gemini are evaluation tiers, and both may use your requests to improve their models. For unpublished work, use a local model through Ollama, or check the provider's data-handling terms before you paste a key: Mistral, Google.

Reporting a problem

The ? button in the panel has Report a problem…: it opens a new GitHub issue in your browser, prefilled with the Pellaeon and ChimeraX versions, the provider and model, and the last two turns of the chat with the commands that ran and their errors. Paths and web addresses are replaced; keys from Settings are not part of the report. Read the text on GitHub before you post it, since the chat itself may contain things you typed. Copy report puts the same text on the clipboard for an email, and Copy chat copies the whole conversation.

Updating

ChimeraX edition: ChimeraX notes available bundle updates in Tools ▸ More Tools…; toolshed update ChimeraX-Pellaeon on the command line does the same thing. If you installed from GitHub, run that install line again and it fetches the newest release. Classic edition: unzip the new zip over the old folder; your settings and chats live elsewhere and are kept.

Where things are stored

Chats are saved there as JSON, and learned.json holds the command corrections Pellaeon learns (Settings ▸ Remember command fixes; view lists them). Delete the folder to start clean.