Introducing CodeAssist, your personal coding assistant that learns as you work. Every edit becomes training data. Every session makes it better. The more you code, the more it understands you.

New podcast recording with Gensyn CEO, @fenbielding, coming soon.

5/ This report marks the start of a living research paper - updated after each major release. We’re building CodeAssist in the open to advance AI research through transparency, reproducibility, and shared progress. Get involved discord.gg/gensyn

4/ The report outlines a three-phase roadmap: Phase 1: LeetCode-style tasks for measurable research progress Phase 2: open-ended, single-file coding Phase 3: multi-file assistance Each phase expands the assistant’s ability to generalize and collaborate across real workflows.

3/ In CodeAssist, this plays out as a pair-programming environment. Your assistant learns patterns in how you code, test, and reason - deciding when to edit, comment, or hold back. Small local models handle policy decisions, while frozen coding models like Qwen 2.5 via Ollama execute them.

2/ Tool-mediation games extend assistance games to real-world settings. Instead of acting directly, the assistant decides which tool (e.g. an LLM) to call, how to parametrize it (e.g. the prompt), and when to intervene. This separation of decision-making and action-taking keeps training modular, auditable, and personal.

We’ve released the Day 0 research report for CodeAssist, outlining a new approach to training personalized AI coding assistants with tool-mediation games. 📄 github.com/gensyn-ai/codeass…

This walkthrough took place yesterday in the Gensyn Discord. For additional support or questions, please join here discord.gg/gensyn

Need help with CodeAssist? Here’s a live walkthrough of CodeAssist from @gab_p_andrade, the research lead who helped bring the project to life.

Announcing CodeAssist CodeAssist allows you to continually train a local, sovereign AI coding assistant on your own device, just by coding yourself using the @gensynai stack, any app can passively train from direct user interactions - the highest quality data for AI

Train your assistant when to take action, what action to take, and instill your taste and style into it.

Introducing CodeAssist 🧑‍🤝‍🧑 A personalized coding agent built on @gensynai. It learns directly from your keystrokes, adapting to your personal style. Run it today and train the best agent.

so proud of the entire team behind CodeAssist - a true labor of love to bring something this innovative to life huge shoutout to the gensyn community for showing up with 30 minutes’ notice for the live walkthrough. 1.1k of you joined.. unreal

We’re thrilled to release CodeAssist today - a novel research prototype exploring whether AI can truly learn to cooperate with you through interaction alone. Unlike traditional coding tools that autocomplete or take over, CodeAssist learns how you work in order to work with you. Every edit, deletion, and test run becomes training data - all processed locally, on your device. Blog 🔗 blog.gensyn.ai/introducing-c… Repo 🔗 github.com/gensyn-ai/codeass… How-to 🔗 docs.gensyn.ai/testnet/~/cha…

Our relationship with machines is increasingly bidirectional. This switch began with IDE autocomplete and accelerated with coding assistants like Claude Code. Today we're pleased to push the envelope even further with CodeAssist - an assistant that requires no direction. 👇

Today we’re launching CodeAssist: a research prototype with a privacy-preserving, on-device coding assistant. The model runs and trains locally, learning alongside you as you solve LeetCode-style coding challenges. Try it now and help us build the next generation of AI assistants: docs.gensyn.ai/testnet/codea…

Was going to wait a few days before saying more, but am too hype about what we're building so here we go... In this 🧵 I'll drop my (biased) thoughts on what I think are the most important and exciting points about CodeAssist.

Live CodeAssist demo and FAQ happening in the Gensyn Discord right now 👋 discord.gg/gensyn

8/ Join us on Discord to share feedback, connect with other builders, and help improve CodeAssist: 💬 discord.gg/gensyn

7/ This is the first step toward a new kind of coding assistant - one that learns with you, not just for you*.* We’ll keep expanding CodeAssist to new environments, stronger models, and open research updates. Learn tips, tricks, and best practices for using CodeAssist 🔗 github.com/gensyn-ai/codeass…

6/ Built on assistance games, an alternative paradigm in cooperative AI, CodeAssist treats collaboration itself as the reward signal. Your preferences are learned directly through interaction - no manual feedback or prompt tuning required. It’s alignment through participation. For background, check out this paper that inspired us arxiv.org/abs/2504.07091