Feedback

[dskw]

Hey, first of all, great work!

I've tried to build an extensive Web UI to replace my own use of claude code in web and vscode to host CheetahClaws myself and access it through the web on PC and mobile.
But the tool is developing at a pace I can never keep up with, not even with claude.
I was just working on planning support and agent management when I realized you have already added a new agent wizard.
v3.05.72 alone has me scramble part of my development plan as I probably should leverage the sqlite support to enrich my own sqlite storage.

Couple of questions and feedback:

• get rid of the per-language readme docs. Currently only like 10% are actually translated with the rest just being english.
• Is there a roadmap of what you are working on step by step? At the moment each tag contains bugfixes mixed with features and no good overview
• Does the plan mode only use basic tasks, or does it support complex plan documents like claude produces them right now?

Have you already started thinking about a web UI yourself?
I'm wondering if my efforts are in vain as you progress so quickly.
At the current rate of development I might have to switch to building a simpler UI that just hosts a web based shell for direct cheeta access.
My implementation keeps track of sessions and projects, allowing to manage providers, models, agents, tools and git repos and later encapsulate the workspaces in sandboxes to run with authorization unlocked for all tools.

Thanks for the great work and I'd be happy for any feedback to my feedback :)

[chauncygu]

Thanks so much for the detailed feedback — and for building on top of CheetahClaws! Really great to hear someone is pushing it this far.

Let me address each point:

1. Per-language README docs
   You're right, most are only ~10% translated, with the rest still in English. We'll clean these up: either remove the incomplete ones or add a clear "partially translated" disclaimer. Thanks for flagging this.

2. Roadmap
   Fair point, the changelog mixes bugfixes with features and there's no forward-looking view. We'll publish a ROADMAP.md on GitHub with a clear breakdown of what's planned, what's in progress, and what's done. Expect this soon.

3. Plan mode
   Currently plan mode is relatively basic: the AI writes a Markdown plan file in read-only mode, then implements after /plan done. It doesn't produce the structured plan documents with nested checklists and dependency graphs that Claude Code generates. However, you can combine it with TaskCreate / TaskUpdate (which support blocks / blocked_by dependency edges) for more structured tracking. Improving plan mode to support richer plan documents is on our radar.

4. Web UI
   We haven't started a Web UI ourselves yet, but it's clearly a high-demand feature. There are two practical paths:

• Quick path: A web-based terminal (xterm.js + WebSocket proxy to the cheetahclaws process), which gives full access with minimal development effort.
• Full path: A proper Web UI with session/project management, provider config, and visual agent tracking.

Given that you've already built significant infrastructure for this, we'd love to explore collaboration. If you're open to it, we could:

• Provide a stable API/RPC layer that makes it easier for external UIs to interact with CheetahClaws (instead of parsing terminal output)
• Potentially integrate your Web UI as an official frontend (with proper credit)
• At minimum, coordinate on breaking changes so your UI doesn't get caught off guard

5. Your development pace concerns
   Totally understand. The recent burst (v3.05.67 → v3.05.72 in one day). We are working on this project intensively, and we have merged some great PRs from the community. With the ROADMAP we'll publish, you'll have visibility into what's coming so you can plan accordingly rather than react.

The SQLite session store (session_store.py) is designed to be a stable foundation; if you're building on top of it, the schema and API should remain backward-compatible going forward.

We'll be pushing updates soon based on your feedback (roadmap, language docs cleanup). Feel free to reach out directly at shangding.gu@berkeley.edu if you want to discuss the Web UI collaboration; we'd genuinely welcome it.

Thanks again for the thoughtful feedback and for building on CheetahClaws!

[dskw]

Thank you for the response!

I don't think my little project is worth the effort yet.
It is more a testing ground for me to try to build a complex framework with claude code, in the hopes to replace that workflow with a web based cheetahclaws in the long run.

So far I have only achieved some basic implementations and have not even added cheetahclaw execution yet.
I was going to use a custom wrapper (currently a stub for the early development) to wrap the process and parse terminal output.
A real API or RPC layer would be very useful indeed.

You can see the current state of it here:
https://github.com/dskw/clawstation/
UI Mockup used for implementation:
https://htmlpreview.github.io/?https://github.com/dskw/clawstation/blob/main/design/clawstation-ui.html

But honestly, do not waste time on my request. I might fail along the way.
It is a free-time project for me, so I cannot spent a lot of hours per week and I will never keep up with your pace anyway. Others (or yourself) will likely push something out much faster until I have a well working state.

I might reach out to your mail in the coming days to ask about your setup, because I have the impression you have a pretty solid agentic system running to develop the tool and I would be interested learning more.

Thank you once again

[dskw]

I think the best approach for me would be to start from scratch and use xterm.js and make that console directly available in the web application.
The web application would allow configuration management like providers and models and spawn one CC instance for each workspace in a sandboxed environment.
CC features could then be triggered through the web interface for easy access, like wizards for setting up agents or other tools and handle the attachments for prompts with texts and images.

My initial plan with the wrapper around the executable is too risky to get out of sync. Such a web tool would need an extensive API that allows grabbing the current state of CC, read outputs in a processing friendly way and call functions directly instead of console commands. Which is a huge undertaking and maybe not fitting the scope of your tool.

Question:
Is it safe to manage the configuration files of CC manually through the backend of the web application, or are they subject to breaking changes between versions that require CC tool executions to migrate to their latest state?

[chauncygu]

Thanks for sharing the repo and mockup, the ClawStation UI design actually looks really well thought out! The workspace management + sandbox concept is solid.

I think you are right. A web terminal that directly hosts CheetahClaws gives you full feature parity with zero sync risk. We may plan to develop it for CheetahClaws in the future. The web app handles what it's good at (config management, workspace/sandbox orchestration, visual controls), and CheetahClaws handles the AI agent loop.

On your question about config file stability:

The config file (~/.cheetahclaws/config.json) is safe to manage externally. The format is a flat JSON dict with string/number/boolean values, no nested structures, no version-dependent migrations. New versions may add new keys with defaults, but never remove or rename existing ones. So your backend can read/write it freely.

The files that are safe to manage externally:

• ~/.cheetahclaws/config.json: user settings (model, API keys, permissions, etc.)
• ~/.cheetahclaws/mcp.json: MCP server configurations
• ~/.cheetahclaws/plugins.json: installed plugin registry
• ~/.cheetahclaws/sessions.db: SQLite session store (read-only queries are fine)

We'll document this stability guarantee explicitly once we publish the ROADMAP.

Don't underestimate your project, even a "simple" xterm.js + workspace manager would be genuinely useful to the community. No need to race against our pace; different people have different paces and ways. Just take your time. A solid Web UI is a very good benefit for our community.

Thanks again for sharing!