I am building a collaborative coding agent

Hello folks,

I know there are way too many coding agents out there but I believe I am taking a different approach. I am building nocodo for entire teams, particularly for non-technical teams. It is self-hosted and I want to share how I am building it. Happy to hear thoughts and criticisms. At a high level:

  - Runs on a Linux box on your cloud
  - nocodo has a "manager" which is the agent and it runs on Ubuntu Linux
  - Linux makes it easy to setup a full development environment on the cloud so team members do not have to
  - LLMs are OK with multiple languages and frameworks for software development and nocodo does not add any restrictions of its own - no system prompt even at this point
  - nocodo (the manager) is also a headless coding agent, regular HTTP API
  - nocodo does/will (depending on the feature) support managing a Linux box, managing multiple projects, users and teams, permissions, multiple models, coding plans (like GLM 4.6 from zAI)
  - mix and match models for analytical or coding needs, will be mostly done for you
  - Commands, like "build/run [backend, frontend]", "create test DB", etc.
  - Git, GitHub integration, branch/worktree management all through the manager's HTTP API
  - Desktop and mobile (coming) apps for team members to connect to "manager" (let's call them admin apps)
  - Admin apps use SSH and port forwarding to connect to the coding agent
  - Admin apps does/will allow project and prompt management, prompting, collaboration (@ mention team mates into prompts)
  - Admin apps does/will have buttons for the "commands" described earlier, e.g., run the full-stack generated apps, open frontend in browser

I want to add a prompt library, a real-time project context that models can use as a tool to help anyone to improve the technical details of prompts. Or to assist models figure out technical details on their own. Project management, Git management, infra management will also be available as tools to the models. Think `apt install []`. When someone needs full isolation, spinning up another cloud instance should also be possible.

I still have to figure out a bunch of things. I am now showing demos with early prototype. I plan to share a proper "Show HN" in a couple weeks time. I am really focused on non-technical teams and nocodo will have a much heavy server side to get users there. I believe project management will also change as we embrace code generation with LLMs. When features take a couple hours of wrestling with agents instead of days with developers, what will project management look like.

nocodo is itself mostly Rust and that is because I do not want to spend time wrestling with the compiler, but the model can and does it cheap. And the compiler does not allow many types of errors that other languages might. I may nudge nocodo's users, who are backend tech stack agnostic, to also use Go/Rust.

nocodo is itself generated by Claude Code and opencode (using Grok Code Fast 1 and GLM 4.6). I have Claude Pro and a Coding Plan from zAI. Overall about $ 35 / month for about 3 months now.

Personal: I am Sumit. I have been a trying founder and engineer all my adult life. I have been in at least 12 startups across the world. I own nocodo.com since 2013 when I wanted to build a "no code" product. Tried multiple times with ideas around template based code generation. Finally, with LLMs, I can see this may work out at scale. I believe businesses want custom software and they will build (vs buy), if custom software keeps pace with their needs. I live in a little Himalayan village and I have been building LLM enabled products for some time now. nocodo is the one getting real user demand. I have 2 early adopters (founders). I am setting up for 2 small sized businesses now. There is a lot of manual hand-holding I do for them, including jumping into Claude to fix issues nocodo cannot. Please feel free to share your thoughts.

https://github.com/brainless/nocodo

Cheers! Sumit