André Wolke

Failed hard. Got up. Built different.
Obsessive focus. Crossed 150b Token
| | |

[André Wolke @AndreWolke]

@DavidOndrej1 I build something like that a while back, will release it in a new version of https://t.co/9O4NjojYNZ

Its a agent-mesh network, any provider talks to another to execute tests of your software for example. Included is a new harness that brings the toolbox

[André Wolke @AndreWolke]

Love it, I saved at least 3h. Did not run a single test.

1. I create a PR in xNaut for Codex
2. Codex send a msg to Claude.
3. CC runs all tests for the release. For any issue Claude create a new Ticket and message Codex
4. Report created.
5. Create release when all green.

[André Wolke @AndreWolke]

My favorite #ClaudeCode use cases.

1. I use an isolated mac-mini with a CC instance running

2. I build a network wide agent mesh network to communicate.

3. I create new skills to test for example #xNaut UI

4. CC runs click-tests, and writes a report.

Saves countless hours!

[André Wolke @AndreWolke]

Finally shipped the new Designer feature, it is basically the same as what @Lovable does, only difference you run it with your max plan and it does not cost you anything.

[xnaut.dev @xnautdev]

And we shipped another release:
https://t.co/uzo3iEYbg5

Fixed the Designer that was originally bound to the sandbox (not live yet) and now it also runs locally.
https://t.co/vHEPlQZrBc

[André Wolke @AndreWolke]

xNaut v1.11.0 has arrived, a few bug-fixes and a new Design Feature that lets you build proper Designs into Applications or Websites. The NautFlow has now also a verification process to ensure the concept will work.

Try it out.
https://t.co/QJjjOb4ei2

[André Wolke @AndreWolke]

Last night Kimi K3 dropped, I tried it briefly and planned to do a deep test this morning, unfortunately it's unusable at present, @OpenRouter is overloaded.

Take the stats with a pinch of salt, but I would not be surprised if they are accurate.

Happy testing!

[Arena.ai @arena]

Big news: Kimi-K3 by @Kimi_Moonshot is now #1 in the Frontend Code Arena with 1679 pts, surpassing Claude Fable 5.

This is a 17-place jump from Kimi-k2.6 (#18 -> #1).

In Frontend, Kimi-K3 ranked #1 in 6 of 7 domains: Brand & Marketing, Reference-Based Design, Data & Analytics, Consumer Product, Simulations, and Content Creation Tools, landing #2 only in Gaming behind Fable 5.

The full model weights will be released by July 27.

Congrats to the @Kimi_Moonshot team on this major milestone!

[André Wolke @AndreWolke]

@orca_build You might like what I am building at https://t.co/Gll1BPUmLO - a hosted platform for running swarms of AI agents in isolated microVMs, a bit like e2b or Daytona, but you can use your max-plan with it So instead of killing your laptop the agent runs in a sandbox.

[André Wolke @AndreWolke]

NautLoop is coming alive

- next task: Mobile App for orchestration.

[xnaut.dev @xnautdev]

We just soft-launched https://t.co/UrMk9L3nO8

The first private sandbox for agents that can be run on your existing max-plan.

Your Data, Your Agent, Your Plan!

[André Wolke @AndreWolke]

@mattpocockuk I feel ya, been working on solving that for good. Maybe you are interested to try it out?

Its a privat sandbox, like e2b/daytona you can run with your max-plan, orchestrate agents to loop until task finished, providing evidence (video) and report.

[André Wolke @AndreWolke]

@thdxr All those CLI(s) started to charge money for useless AI tooling, so I build my own. https://t.co/O8ri6FMN63 fully open-source https://t.co/wnqAkFsXBK

And now I am adding the tooling #NautLoop - Agents in a private sandbox you can run from your max-plan.
Want to test?

[André Wolke @AndreWolke]

Its coming together, xnaut-mobile in development.

Your agentic toolbox in your pocket. With NautLoop you deploy a team of agents to solve your tasks, fully isolated in their own sandbox...

And best of it - You can use your own Max-Plan.

Your Plan, Your Agent, Your Sandbox!

[xnaut.dev @xnautdev]

Coming Soon

[André Wolke @AndreWolke]

New experimental feature in xNaut is live
- Added visual node-graph editor to build loops

Also @Zellij_dev is now replacing TMUX and you can start persistent tool sessions (Claude, Codex, Pi) that stay alive even when you close xNaut.

https://t.co/wnqAkFsXBK

[André Wolke @AndreWolke]

Using codebase-memory-mcp, Agents can map architecture, trace dependencies, inspect affected
code, identify tests and assess risk before changing anything.

Install this today and save tokens on your favorite CLI.
https://t.co/eUH5ajV65b

[André Wolke @AndreWolke]

Update!
New release turns xNAUT from a coding workspace into an AI-native development studio:

• Configurable agents
•Project planning
• Project management and tickets
• Improved Vault and Librarian
• macOS and Windows builds

https://t.co/ZWiKHOQljh

[André Wolke @AndreWolke]

New Release of #xNAUT pushed:
https://t.co/PJXG7ohOM0

New 🚀
Markdown Vault
Create your documentation direct in XNaut for your before you start building. As the integrated agent to create the docs, copy from obsidian or do it manual.

An yes it runs on local models

[André Wolke @AndreWolke]

Improve your coding quality today with these simple steps and tools:

1. Write down a short text, markdown file
→ Benefit
→ Background
→ Sources

2. Share with your AI and ask whats not clear

3. Use this tool to finalize the plan: https://t.co/nMwdof4X2x

4. build it

[André Wolke @AndreWolke]

One of my favorite Fable5 Errors...

and best, I spend my weekly limits in just 4 days.. crazy!

[André Wolke @AndreWolke]

Spend around Hours with Fable5 - quite unusable.

Unclear why, and if that is only for users outside the US but it got nerfed and unstable.

- API Error: Unable to connect to API (ConnectionRefused)

- API Error: 529 Overloaded. This is a server-side issue

Back to @OpenAI

[André Wolke @AndreWolke]

I've been struggling the past 2 weeks with a large code-base finding the culprit.

The goal - get benchmarking to +85%. Opus 4.8 did not deliver.

I switched to @OpenAI GPT5.5 and around 9h later I get the happy message!

- 88.8% EBR@5 - 93.8% EBR@10 - Assistant: 100%

🚀