cd /news/ai-agents/show-hn-llx-fast-efficient-gpt-6-2-s… · home › topics › ai-agents › article
[ARTICLE · art-171275] src=github.com ↗ pub= topic=ai-agents verified=true sentiment=↑ positive

Show HN: Llx – Fast, Efficient GPT-6.2 Sol and Astra Coding via ChatGPT

Localix released llx, an open-source coding agent runtime distributed as a single Rust binary under 4 MB that runs at the terminal against 1,000+ models from 200+ providers, including a ChatGPT sign-in, OpenRouter, a user's own API key, or local models via Ollama, LM Studio and llama.cpp. In a benchmark on the evalkit task with 238 tests, llx running Anthropic's Haiku 5.5 solved 238 of 238 tests in 286 seconds across 14 turns at a cost of $0.050, which Localix reports as 60% of the time and a ninth of the cost of Claude Code on the same task, model and key. The agent loops over five tools — bash, apply_patch, ask, agents and jobs — and runs commands in the system sandbox by default, writing only inside the project and opening no network unless the user approves.

read6 min views4 publishedNov 17, 2026
Show HN: Llx – Fast, Efficient GPT-6.2 Sol and Astra Coding via ChatGPT
Image: Michielbdejong (auto-discovered)

An ultrafast, lightweight agent runtime for coding and CI

A coding agent at your terminal, on any model: a ChatGPT sign-in, OpenRouter, your own key or a local model (Ollama, LM Studio, llama.cpp). The same agent runs one task in CI and helps your other agents. One binary under 4 MB.

Quick start · At your terminal · Measured · In CI · In other agents · Docs · cli.localix.ai

<sub>Every step a real run on Claude Haiku 5.5: any model · a fix · answers · agents · a decision · a ship check. It plays on cli.localix.ai too, by chapter, where it can be d and copied from.</sub>

1. Install on macOS or Linux:

curl -fsSL https://raw.githubusercontent.com/localixai/localix-cli/main/scripts/install.sh | sh

2. Run it in your project:

cd your-project
llx

3. Sign in and ask:

/login
the median test fails, fix it

/login takes a ChatGPT sign-in, OpenRouter or a key for any of 200+ providers; a model on your own machine (Ollama, LM Studio) needs none. llx doctor says what is missing. Other ways to install: Installing llx.

Installed with the script? Run llx --update to install the latest stable release. A session only notifies you when one is available.

llx is a model in a loop over five tools: bash, apply_patch, ask, agents and jobs. It reads, searches and runs your checks through the shell, and changes files with patches.

  • Fast and small. One Rust binary under 4 MB. On the larger taskbelow it finished in 60% of the time for a ninth of the cost.
  • Any model. 1,000+ models from 200+ providers, and the ones on your own machine;/model lists every one you are signed in to (models ).
  • Sandboxed by default. Commands run in the system's sandbox, which writes only inside your project (never its.git ) and opens no network. What must reach further is asked first (modes and trust ).
  • Answers you can read. Code under its file's name, a diff in green and red, tables, steps and formulas, drawn in the terminal with no markup left over.
  • Saved. Every run is a session:llx -c goes on with the newest one here,llx sessions lists them andllx resume ID opens one (sessions ).

What finished goes into the terminal's own scrollback, so scrolling, selecting and copying stay the terminal's. Several sessions can work in one project at once, each with agents of its own.

<sub>The renderer's own rows, written as SVG. The Markdown is read by flowdown, the lab's library.</sub>

Key What it does
Enter Sends. While a task works, what you type goes to it, read at its next step
Shift +Tab Goes round the modes: default →plan (read-only) →bypass → your own
/ The commands: /login ,/model ,/mode ,/new ,/rewind ,/diff ,/copy ,/usage and more
@ Mentions a file of the project; a dragged file or picture is attached
← On an empty prompt, lists the sessions in this project
Esc ,Ctrl +C Steps back out of what is open; Ctrl +C also stops work

Every key and command: Using llx.

The same task, the same model, the same Anthropic key, started in the same minute; the tests that score a run were written before it.

Task Model Agent Solved Time Turns Cost
evalkit, 238 tests Haiku 5.5 llx 238 of 238 286 s 14 $0.050
Claude Code 2.1.321 238 of 238 478 s 25 $0.46
shopkit, 69 tests Haiku 5.5 llx 69 of 69 69 s 5 $0.012
Claude Code 2.1.321 69 of 69 67 s 4 $0.011

One run of each agent, on one machine: a measurement, not a law. Every run's log, code and test output are in benchmark/, the slower runs too (history); bun benchmark/run.ts repeats it on your key.

llx exec is the same agent given one task: it ends with an answer, in the shape a JSON Schema defines if you pass one, and an exit code your pipeline can branch on (0 done, 1 no data, 2 out of turns, time or budget, 3 error). It works through bash with gh, git and your project's tools, with no MCP server and no built-in prompt, so the same task runs on GitHub, GitLab or any runner with a shell.

- uses: localixai/localix-cli@v0
  id: llx
  env:
    OPENROUTER_API_KEY: ${{ secrets.OPENROUTER_API_KEY }}
    GH_TOKEN: ${{ github.token }} # read-only in this job; a second job posts the answer
    PR: ${{ github.event.pull_request.number }}
  with:
    model: anthropic/claude-haiku-5.5
    allow: bash
    schema: '{"type":"object","properties":{"summary":{"type":"string"}},"required":["summary"]}'
    prompt: |
      Review pull request #$PR with gh and git. Run something to confirm a problem
      before you report it.

Ready-made workflows, each run for real: a pull request review, @llx in any thread, issue triage, a failed run explained and more. The action, the two-job pattern, keys, limits and GitLab: CI. Exit codes and the --json events: Reference.

Claude Code, Codex or a script of yours can call llx from its shell: llx exec hands a task to a helper on any model, so the agent's own tokens go where they matter, and llx decide picks one of a few options about some text in about a second. A skill tells any agent that takes skills when each pays:

npx skills add localixai/localix-cli

In Claude Code it is also a plugin: /plugin install llx --marketplace localixai/localix-cli. More: In other agents.

All of it is also on cli.localix.ai/docs.

Guide What it holds
Installing llx The install script, other ways, pinning a release, signing in
Using llx The command line, a session key by key, models and local hosts, sessions, project instructions
Modes and trust Modes, rules, approvals, the sandbox and its boundaries
Tools The five tools: bash ,apply_patch ,ask ,agents ,jobs
CI ·Recipes ·Examples The GitHub Action, the two-job pattern, shell recipes, ready-made workflows
In other agents Helpers on any model, llx decide , the skill
Reference ·Compared Exit codes, --json events,llx decide ; the action beside other agents' actions
Architecture ·Models ·Changelog The layers and their rules, the model catalog, what changed

Start with CONTRIBUTING.md and ARCHITECTURE.md. Report vulnerabilities as SECURITY.md describes. llx is open source under the MIT License, made by Localix Lab, built and reviewed with Claude Fable 5.2.

Before Localix CLI, this repository was auto-copilot-cli (2023, TypeScript), its author's first AI command line. llx is its successor, written anew in Rust.

── more in #ai-agents 4 stories · sorted by recency
── more on @localix 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/show-hn-llx-fast-eff…] indexed:0 read:6min 2026-11-17 · —