A10fy

What is A10fy?

Ever used ChatGPT/Claude/Gemini/Poe/… web interface? Then you are largely familiar with the chat portion of what A10fy offers. But there is more, read on…

How is A10fy different?

The chat in A10fy is highly tinkerer-friendly:

• You can switch models on the fly: start chatting with a simple (cheap) model and if you are not satisfied, continue the same conversation with a more capable model. Or vary model generation parameters to tweak the creativity of the output.
• You can create a collection of custom system prompts (called personas) tuned for specific creative or information processing tasks. Personas can be as well changed within the same conversation.
• You can edit any message of the chat at any moment of time. E.g. if using LLM for some tasks with similar output format, if LLM does not produce the exact desired output immediately, it is often more effective to correct the output in place than to describe to LLM what was wrong. And later during the conversation that corrected answer serves as a one-shot example!
• You can paste rich text into the chat input area as plain text, HTML or Markdown.
• You can include math notation in the chat using LaTeX syntax, the de-facto standard for math typesetting; formulas are beautifully displayed using KaTeX. Code blocks in LLM output and your messages are automatically highlighted. The chat also supports display of Mermaid diagrams, one of the currently most popular formats of describing diagrams in Markdown ecosystem.
• You can run Python code blocks in the chat messages directly in your browser, this is powered by Pyodide.
• You can define your own “code block players” — sandboxed javascript code that interprets the output of LLM, thus bringing to life static LLM messages.

Beyond the chat. if you go beyond the web app and install Chromium extension, you are also getting:

• AI Assistant: a voice- and text-controlled assistant that can summarise the current page, answer questions grounding on the page content, and perform certain actions on the page DOM.
• Fuzzy Search: use embedding-based technology to perform fuzzy searches through your browsing history for enhanced retrieval.
• Custom LLM-powered actions on the page: add custom context menu entries to process specific portions of the page using LLM with custom prompts.

How do I start using A10fy?

A10fy is a wrapper around Google Gemini API. There is no proxying, the frontend directly communicated with the Google’s servers. Thus, you need a valid Gemini API key to use A10fy (you can get your API key here). You can use multiple API keys if you have many. According to Google’s Terms of Service, you must be at least 18 years old to use the services that power the A10fy extension.

See the Quick Start Guide for an introduction.

Is A10fy free?

A10fy is free to use. The source code is available on GitHub under MIT license.

How does A10fy use my data?

Plain and simple: it doesn’t. A10fy does not have a backend to store your data, so all your conversations are stored on your own computer in your browser’s local storage. However, up to Google Terms of Service, if you are on Free tier of the Google Gemini, the data that comes to Gemini API may be used by Google to improve its services, so it is recommended not to submit sensitive, confidential, or personal information to the chat and not use the A10fy assistant on pages containing confidential or personal information if you are on Free tier of Gemini API.

Chromium Extension vs. Web Version Feature Comparison

Feature	Web Version	Chromium Extension
Browser AI Assistant
Chat with AI with image attachments, Markdown editing, flexible text paste (HTML/Markdown/Plaintext), code block players
Custom Code Players	Sandboxed, can load external libs	Sandboxed, no access to external URLs
Python Code Player: available packages	All Pyodide-compatible packages	Only standard library
Flexible Model and Persona Editing
Fuzzy Search of Browsing History
Customizable processing of page DOM elements

Video demos