Use a terminal based AI assistant to read, edit, and run commands in a coding project.
Let a hosted routing service pick a cheaper or more powerful AI model automatically per request.
Track and cap how much a coding session spends on AI usage in real time.
Reproduce the project's own benchmark to compare its cost and speed against other coding assistants.
| klaatai/klaatcode | moonshotai/kimi-code | s0ld13rr/pentestcode | |
|---|---|---|---|
| Stars | 234 | 234 | 230 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | easy | easy | moderate |
| Complexity | 2/5 | 2/5 | 5/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Model routing depends on a separate hosted service the project runs, not something you self host.
Klaat Code is a terminal based AI coding assistant you install once and use inside any project, similar in spirit to tools like Claude Code, opencode, Codex CLI, and Aider. It reads your code, edits files, runs commands, and checks its own work, asking permission before doing anything risky. What sets it apart, according to the README, is that every message you send is routed through a separate hosted service called Klaatu, which decides how much AI power a given request actually needs and sends it to one of five cost tiers instead of always using the most expensive model available. Because that routing decision happens on the server rather than in the tool itself, the client you install is described as a thin terminal connection to that service, comparable to how a tool like gh connects to GitHub. The project also builds an index of your codebase as a call graph with search built in, which it says lets the assistant look up specific functions and their relationships instead of reading whole files, aiming to use far fewer tokens per task. Other stated differences include unlimited tool use without needing to click continue, real time spend tracking with a spending cap you can set, and a system that tries to preserve important context when older conversation history gets summarized down. The README includes a benchmark comparing this tool against Claude Code, Cursor, and opencode on the same 33 coding tasks, reporting that Klaat Code solved all of them at a lower cost per task than the others in that specific test run, and it provides a command to reproduce that benchmark yourself. You can install it through npm, a one line shell installer, a Windows PowerShell command, or Homebrew, and every method installs a single compiled program that does not require Node or Bun to already be running on your machine. The project is released under the Apache 2.0 license, which allows free use, modification, and commercial use as long as you keep the copyright notice.
A terminal AI coding assistant that routes each request to a cheaper or more powerful AI model automatically, aiming for lower cost than similar tools.
Mainly TypeScript. The stack also includes TypeScript, Node.js, Bun.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.