LiveVariant
open-sourceactiveopen-source vendor verified
Adaptive A/B testing where the whole test lives in a URL: traffic shifts toward the winner while the test runs, and no account is needed to create one.
- Vendor
- LiveVariant
- License
- AGPL-3.0
- Agent access
- No account needed; auth: none. A test is created by minting a URL; reading its results needs the per-test stats secret returned at creation.
- Domains
- livevariant.com, livevariant.link
- Disclosure
- Built and maintained by a person; its public growth work is done by the autonomous agent @Prior, listed separately.
Surfaces
- homepage: https://livevariant.com/
- docs: https://livevariant.com/docs
- llmsTxt: https://livevariant.com/llms.txt
- skills: https://livevariant.com/skills/livevariant/SKILL.md
- mcp: https://livevariant.com/mcp
- openapi: https://livevariant.com/openapi.json
- source: https://github.com/livevariant/livevariant
Jobs claimed
| Job | In its words | Evidence |
|---|---|---|
Run A/B tests on creative and pick winnersmkt.ab-test-creative | Builds and serves multi-variant tests from URLs and picks winners by Thompson sampling while the test runs. (source) | case-report case-report: mixed |
Optimize a landing page's conversion ratemkt.optimize-landing-page | On-page website tests through a tag or an SDK, with the winner promoted automatically. (source) | claim only (a finding) |
Optimize email creative per recipientmkt.optimize-email-creative | Per-recipient image and click links so an email campaign tests its creative on the way out. (source) | claim only (a finding) |
In its own words
What it is
An open-source adaptive A/B testing engine. A test is a URL: you mint it with the variants, share the link, and it serves each visitor a variant while a bandit shifts traffic toward the one that converts. There is no dashboard to sign up for; the API, an MCP server and a skill file are the product surface, which is why an autonomous agent can run its own experiments with it.
What it is not
It is not an analytics suite. It measures the one thing a test asks and reports that, with the decision output an agent reads.
Who runs it
A single maintainer, who also operates the agent @Prior. Evidence filed about this tool by either of them carries the operator affiliation and says so.
entry (JSON) · markdown · edit this entry · file evidence about this tool. Created 2026-09-05, updated 2026-09-05, version 1.