We ripped out LiteLLM and wrote our own proxy after the repo became unmaintainable vibecoded slop. This is what happens when you can't audit what you depend on.
LiteLLM HAS BEEN COMPROMISED, DO NOT UPDATE. We just discovered that LiteLLM pypi release 1.82.8. It has been compromised, it contains litellm_init.pth with base64 encoded instructions to send all the credentials it can find to remote server + self-replicate. link below
Claude Code looks for code two ways: grep for a keyword, or spin up subagents that page through files 200 lines at a time.
Rather than burning thousands of unnecessary tokens, Brokk maps your codebase. The AI just looks it up.
@camsoft2000 The issue is that your codebase isn't a bag of text. Most tools grep through it hunting for relevant pieces, which is fine at 10k lines, but at scale the model is reasoning over disconnected fragments with no picture of how anything connects.
Your error rate reviewing diffs is higher than the model's error rate writing them.
@spyced on why the value of an engineer in 2026 is deciding what to build and where to point the agents.
Your AI coding tool works in one IDE.
Brokk works in all of them. IntelliJ @intellijidea. VSCode @code. Zed @zeddotdev. Codex. Our own TUI.
Pick your favorite editor. We're already there.
Most VS Code AI Extensions treat your codebase as a bag of text.
Deep Scan reads code the way a compiler does: symbols, types, call sites across your entire project. That's the context your AI needs to actually be useful.
Brokk for VS Code is live.
You asked for a TUI. We built it.
Stop context switching between your terminal and a browser window. The Brokk TUI brings Plan Mode and Lutz Mode directly into your shell. It is built for speed and keyboard-driven engineering.
ChatGPT @ChatGPTapp Plus users burn message limits on context noise. Brokk is surgical: it sends only what the task needs. Shorter prompts, more tasks per message.