Quickstart

Install#

$ curl -fsSL https://launch80.com/install.sh | sh

macOS and Linux, Intel or ARM. The script verifies a checksum and writes one binary to ~/.local/bin. If that directory is not on your PATH, the script prints the line to add. The whole script is on one page if you want to read it first.

$ L80 update

Get a publish token#

Tokens are issued by hand right now. Ask in Discord and you will usually have one the same day. Nothing publishes without it. When it arrives, save it to disk:

$ mkdir -p ~/.config/L80 && umask 077 && printf '{"token":"%s"}' '<TOKEN>' > ~/.config/L80/credentials.json

$ L80 doctor

Publish#

One command, depending on what you have. Each prints a URL on its last line.

$ L80 betterbench --endpoint http://localhost:8080/v1 --model qwen3-32b

$ L80 betterbench --results results.json

$ L80 publish report.json --template generic.report.v1

A full BetterBench run takes a while: 20 passes per category plus prefill and concurrency sweeps. --quick is a 5-pass smoke run, and the page says so. The results file stays in the working directory. The endpoint URL and hostname are stripped before upload and the server refuses a file that still carries them.

$ L80 betterbench --results results.json --dry-run --payload-out payload.json$ L80 publish report.json --dry-run

BetterBench#

L80 betterbench either runs BetterBench against an endpoint or reads a results.json you already have, then publishes the result. Every flag it accepts is below. Anything after -- is passed to betterbench run untouched.

--title and --summary set the heading and the paragraph under it. --section "Heading=Body" adds a prose section and repeats. --note key=value records a fact about the run and shows it as a chip, for example --note engine="vLLM 0.9.3". It also repeats.

From a coding agent#

Everything above except the token is a shell command, so an agent can do it. Pick the variant that matches what you have, paste it into Claude Code, Cursor, or whatever runs your terminal, and come back for the link. The prompt tells the agent to stop and ask you if the token is missing.

Set up the L80 CLI so you can publish results to launch80, then publish a report.

1. Install it: curl -fsSL https://launch80.com/install.sh | sh
   If ~/.local/bin is not on PATH, add it, then confirm `L80 version` works.
2. If you are Claude Code, install the skill: L80 skills link --target claude-code
   Any other agent: run `L80 skills print` and follow that file instead.
3. Run `L80 auth status`. If it reports no token, STOP and tell me: I will save
   one to ~/.config/L80/credentials.json myself. Never ask me to paste a token
   into this chat, and never write one to a file for me.
4. Run `L80 doctor` and fix anything it flags.
5. Benchmark my server and publish the run:
   L80 betterbench --endpoint <URL> --model <MODEL>
   Ask me for the endpoint URL and model name if I have not given them.
   Use --quick only if I ask for a smoke test.
6. Give me the URL from the `published:` line and repeat every caveat the
   tool printed, especially anything about under-sampled percentiles.

$ L80 skills link --target claude-code

$ L80 skills print

Templates#

Three layouts. The choice matters because the wrong one loses information rather than just looking plainer. Each id links to a real published example.

What a published page is#

• Public. Anyone with the link can read it.
• Immutable. To change something, publish again and share the new link.
• Unindexed. Search engines are told to stay out. Sharing the link is unaffected.
• Self-contained. No scripts, fonts, or trackers load.
• Labelled. Every page carries a visible “self-reported, not verified by launch80” badge.
• Kept 5 days on the free tier, then removed. Republish to keep a result around.

L80 history lists the URLs your key has published, from any machine. With no token or no network it falls back to the local record on this machine.

Commands#

Every network-facing command accepts --json for machine-readable output and --api-base to override the server. Flags may come before or after the file argument. The betterbench flags have their own section.

Environment and files#

Resolution order for the server is flag, then variable, then default. For the token it is variable, then file.

Limits#

Every token is on the free tier today. A paid tier with a higher daily limit and longer retention is coming; the numbers will be posted here and in Discord when it opens. If you need more before then, ask in Discord.

Errors#

Every error prints a code, a sentence, and a remedy naming the next action, and the process exits with a code an agent can branch on.

Help#

Tokens, questions, and new template proposals all go through Discord. The lab notes cover the reasoning behind the templates.