LaunchFlag / Changelog
Changelog
Everything shipped in LaunchFlag, newest first. New features, improvements and fixes, one entry each. RSS.
9 September 2026
#- New
API tokens, one per thing that uses the API
There was one token per account, so revoking the leaked one broke your agent, your CI job and your status page together. Create as many as you like, name each, and revoke one without touching the rest. Each carries a scope: Full does everything you can, CI changes flags in dev and staging and cannot reach prod or Emergency off, Read only changes nothing. The token you already have keeps working, now called Default.
- launchflag 0.5.0New
One flag can require another
Pick a flag under Depends on, and this one stays off for a user while the flag it requires is off for them, whatever its own rollout says. The chain resolves inside your process in the same pass, so it costs no extra request, and why() reports prerequisite_off rather than leaving you staring at a rollout that looks like it should be serving. Agents get it as the depend_flag tool. A flag published to browsers cannot use it: the public document holds no unexposed flags, so the answer would be wrong rather than merely missing.
- launchflag 0.5.0New
An OpenFeature provider
import { launchFlagProvider } from "launchflag/openfeature" and your code talks to the standard API instead of ours, which means leaving is one line. Evaluation still happens in your process: same config document, same cache, same fail modes, no network call per evaluation. Booleans are flags and strings are variant arms; there are no numeric or object flags here, so those hand back the default you passed rather than pretending.
- Improved
Free takes teammates, and staging is on Pro
Free was one person with no percentage rollouts, which taught anyone who tried it that we could not do rollouts. It now carries unlimited teammates and every targeting mode, so the thing you evaluate is the actual product. Pro gained the third environment: staging is a $12 feature, not a $29 one.
8 September 2026
#- Improved
The trial is 7 days
It was 14. With a free plan underneath it, the end of a trial is a downgrade rather than a lockout, so a shorter one costs you nothing: every flag, environment and teammate stays, and you keep the kill switch. The warning email moves to two days out to match.
- launchflag 0.2.5Fixed
npx launchflag init installs the newest SDK
init ran a bare `pnpm add launchflag`, which resolves against your lockfile and local store, so a machine that had seen an older version installed that one again and printed "0.2.4 is available" while doing it. It now pins @latest. If you already have an old copy, `npm install launchflag@latest` moves you.
7 September 2026
#- New
A free plan, forever
One person, one project, two environments and three days of audit history, at no cost and with no card. Nothing about a billing state can lock your account any more: a spent trial, a failed card or a cancellation lands you on Free with every flag intact, and the kill switch keeps working. We will never freeze the switch over money.
- Improved
No seat limits on any paid plan
Pro and Launch take as many teammates as you like and the price never moves when you add someone. The product is a switch you pull from your phone at 2am, so charging for the people allowed to pull it would be a safety problem before it was a pricing one. Free is the exception, and it is a limit rather than a charge: Free is for one person.
- New
Find the flags whose code can go
Every flag is temporary, and the last step is the one everyone skips. LaunchFlag now spots settled flags, on for everyone in every environment and unchanged long enough that the rollout is clearly over, and hands your coding agent the exact edit to delete the flag, the old branch and the check. Available on the dashboard and as the cleanup_flags MCP tool.
- Fixed
Archived flags are a real state
Archiving used to mean "off in every environment", which is also what a brand-new flag looks like, so hundreds of live flags read as archived and archiving one changed nothing you could see. It is now its own state: an archived flag says so in the list, its configuration is shown rather than editable, and it stays off the stale worklist.
- New
A status page
launchflag.dev/status runs the checks live when you load it, so it is never reporting a cached answer: the database, flag delivery and the evaluation engine, each with the milliseconds it took. If the page is red, your app is still fine, because flags evaluate inside your own process against config the SDK already holds.
5 September 2026
#- launchflag 0.1.3Improved
One command sets up the SDK, the skill and the MCP server
npx launchflag init now does the whole setup in one pass: installs the package, writes your dev key into .env.local, downloads the agent skill, and points Claude Code and Cursor at the MCP server. Pass --key, --token and --url or let it ask. Run it twice and nothing changes, because an env value that already exists is never overwritten.
- New
LaunchFlag runs on LaunchFlag
The site and the dashboard now read their own flags through the published SDK. Four of them: the landing headline is a 50/50 test, and the founding offer, Insights and the docs code tabs each sit behind a kill switch we can flip without a deploy. Every fallback is the page you already know, so a dead API changes nothing. The public footer prints how many flags that page evaluated.
4 September 2026
#- launchflag 0.1.0Improved
Flagship is now LaunchFlag
The product, the SDK and the MCP server all carry the LaunchFlag name, so the package you install is launchflag and the setup command is npx launchflag init. Nothing about your keys or your flags changed.
- launchflag 0.1.0New
Local evaluation and the config endpoint
The SDK now fetches one rule document from /api/config and decides every flag inside your process, so 200 users across 40 flags cost a single request per cache window instead of one per user. The document carries no user data, caches at the edge, and revalidates with an ETag, so an unchanged config costs nothing.
- New
Three environments with promotion
Every project has dev, staging and prod, each with its own key, so the same code returns different answers on your laptop and in production. Promote copies one environment forward, dev to staging and staging to prod, and a human always does the last step.
- launchflag 0.1.0New
Multivariate flags
A flag can now carry 2 to 5 arms with integer weights summing to 100, read with variant() instead of flag(). Each user keeps the same arm forever because the bucket is hashed from the user id, so the split is stable across servers and deploys.
- launchflag 0.1.0New
First N targeting
Turn a flag on for the first N people who reach it and it closes itself once the seats are gone, which is the shape most early access actually has. Admission is a row in the database rather than arithmetic, so those flags call /api/eval and the answer is cached for 60 seconds per user.
- New
Project members
Invite someone to a project and they can create and flip flags while the owner keeps billing, deletion and key regeneration. Every flip records who did it, so the audit log names a person rather than an account.
- New
Trial pricing and the founding offer
LaunchFlag is a 14-day trial with no card, then Pro at $12 a month or Launch at $29, priced per project rather than per seat. The first 10 customers keep 50% off for life with the code LAUNCH50.
- launchflag 0.1.0, launchflag-mcp 0.1.0New
Hosted MCP server and npx launchflag init
Claude Code, Cursor and Codex can create, read and flip flags through the hosted MCP server using a management token, so an agent can wrap its own risky change in a switch. One command, npx launchflag init, writes your dev key to .env.local, downloads the agent skill and configures the MCP server.
- New
Docs
Two pages that hold everything: the SDK quickstart with the raw API and rate limits, and a page on agents and MCP. Both have a contents rail that follows your scroll, deep links on every heading, and a filter on the / key.
- Improved
New interface across the whole product
The dashboard, the landing page and the docs now run on one design system: off-white, hairline rules, Archivo headlines, and red reserved for the things that actually do something. The flag list is a table on desktop and cards on mobile, and the kill switch is a full-width control you can hit with a thumb.