Test intelligence for software teams · Free and open source · Self-hosted
TestLookup is free, open-source test intelligence for software teams. It collects the results of your automated software tests, explains why tests fail, spots flaky tests, and helps you decide whether a software release is ready to ship. You run it on your own servers, and it works offline by default.
curl -fsSL https://raw.githubusercontent.com/test-intelligence/testlookup/main/install.sh | bashYou need Docker with Compose v2, and Bash. On Windows, run it from Git Bash or WSL. The script downloads the published images, creates random passwords and keys, and shows the admin password once. To build from the source code instead, clone github.com/test-intelligence/testlookup and run make quickstart, which needs Git and GNU Make.
Sample data Screens from TestLookup itself: top failing tests, dashboard figures and a run's recommendation.
Every failed build raises the same questions. TestLookup answers them from your own test history.
TestLookup groups failures that share a cause. It sorts each one by kind: product bug, infrastructure, test data, or automation defect, which is a bug in the test itself. Each failure comes with its evidence and history, so nobody looks into the same failure twice.
A flaky test passes or fails at random, with no change to the code. TestLookup scores each test from four signals and shows how sure it is, so you can tell flaky tests from real bugs. It can suggest setting a flaky test aside (quarantine), but a QA lead must approve that first.
TestLookup gives each software release a GO or NO-GO verdict, with the reasons behind it. It looks at the latest results of the tests linked to the release and at any open defects. Until at least 5 different tests have results, the release shows Not evaluated instead of a verdict.
Upload your test report files, or send them from your continuous integration (CI) jobs with the command-line tool or the API. You can also send results live while the tests run.
Built-in rules check every failure, and a model trained on your team's labels joins once there's enough data. If you turn one on, a local AI model can explain likely root causes. Everything links back to each test's history.
Each person works through failures in their own inbox, tracks flaky tests and defects, and checks whether a release is ready. Do it in the web app, from the command line, or through an AI assistant.
Collect
Send the test reports your tools already write. TestLookup works out the format, keeps the CI details of each run, and links every result to that test's history.
Analyze
See which failures share a cause, which are new and which keep coming back. Each one shows its kind, its history and the recent code changes that may have caused it.
AI you control
Rules and a model trained on your own labels do most of the work. An AI language model is optional, either running locally with Ollama or a hosted one an administrator allows. Every analysis is saved as a version, and a person's correction becomes the next one.
Delegate
After every run, TestLookup's AI agents, which are automated helpers, analyze the failures. The Investigator tests five possible explanations at once. It can propose actions, such as setting a flaky test aside, and they run only after a QA lead approves them.
Stabilize
Find the tests that pass and fail with no change to the code, and rank them by impact. Set them aside with an approval everyone can see, instead of quietly skipping them.
Own
Your ownership rules and your CODEOWNERS file, which says who owns which code, decide who owns each test. Every new failure goes to its owner's inbox. Each test keeps its history run by run, and decisions wait in one approvals list.
Track
Follow pass rate, test executions, failures and durations over time, by suite, branch, environment or release. Insights point out what changed. A number TestLookup couldn't measure shows as blank, with the reason, instead of zero.
Decide
Plan each software release in phases on a shared calendar. Ship on a verdict you can explain, with the blocking reasons and the evidence behind them. Overriding a decision needs a written reason, and every decision is logged.
Manage
Keep your written test cases, suites and plans next to the test results that prove them. Library health shows what is automated, out of date or waiting for review.
Automate
Each run arrives linked to its pull request, CI job and code changes. TestLookup finds the build where a new failure started, and automatic actions wait for a person's approval. If TestLookup can't answer, your pipeline check fails instead of passing.
Govern
An activity log records every change. You also get roles, single sign-on, holds that stop data from being deleted, personal-data requests, and signed evidence an auditor can check offline.
You don't need a new test runner or any rewrite. TestLookup reads the reports your tools already produce.
testlookup release gate "2026.10" --project "Checkout Service" --evaluate --fail-on NO_GOIt records a fresh verdict, then ends with a code any CI system can act on. The code is 0 for GO, 1 for NO-GO, 3 when there's no verdict yet and 2 when TestLookup can't answer. Details: the --evaluate option needs a QA lead or a project API key.
All integrationsSpend your time on new failures instead of re-reading old ones. Your inbox shows what's assigned to you.
See whether your change broke something or the test is just flaky, with the likely causes and the test's history.
Get a clear GO or NO-GO for each release, the evidence behind it, and a record of every override.
Run it on Docker Compose, Kubernetes, K3s or OpenShift, including on networks with no internet access.
AI is optional
Rules check every failure, and a model trained on your team's labels joins once there's enough data. You don't need an AI language model at all.
When you add an AI model, it runs on your own hardware through Ollama, free software for running models. Offline mode is on by default and blocks cloud AI providers.
Every AI report goes to a person for review, and corrections are kept as new versions. By default, reports go out before anyone reviews them. An administrator can make review required.
Private by default
TestLookup runs entirely on servers you control. Offline mode is on by default, and while it's on, TestLookup makes no calls to cloud AI services or outside integrations. It sends notifications only to private addresses or to addresses you have approved.
One command downloads the published images and starts TestLookup. The quick start shows how to add sample data, or how to build from the source code instead.
Click through three short walkthroughs in your browser: a first test run, looking into a failure, and a release review. They use made-up data.
Learn the main ideas, how to send results, the settings, how to install and run TestLookup, and the full API reference.
TestLookup is built in public on GitHub under the Apache-2.0 license. Follow the roadmap, read what changed, and help decide what comes next.
Yes. TestLookup is open source under the Apache-2.0 license. There's no paid edition and no hosted service, so you run it yourself.
Only if you allow it: offline mode is on from the start, so TestLookup makes no calls to cloud AI or outside integrations. Notifications go only to private addresses or to addresses you have approved. The only other traffic is basic housekeeping: downloads of software images and AI models you haven't saved locally, name lookups (DNS) and clock sync.
No. Rules check every failure, and a model trained on your team's labels joins once it has enough data. You can add a local AI model through Ollama for deeper root-cause analysis and the Ask AI chat.
You can use a local model through Ollama, LM Studio, LocalAI or vLLM, or a hosted one from OpenAI, Anthropic, Gemini or OpenRouter. Offline mode, which is on by default, allows only models on your own network. An administrator can limit the choice further.
It reads report files from JUnit, pytest, TestNG, Allure, Playwright, Cypress, Robot Framework, Cucumber, NUnit, TRX and xUnit. It can also receive live results from pytest, Robot Framework, Jest, Mocha, Playwright, Vitest, Cypress, Cucumber JS, TestNG, JUnit 5 and 6, and Go.
No. Your own test frameworks and CI system run the tests. TestLookup collects the results and turns them into answers.
Not yet: TestLookup is still in early development, before version 1.0 and its first release, so expect some bugs and changes. You can run it today with the one-command installer, which uses the newest images from the main branch, or build it from source. Follow the roadmap for news of the first release.
Run TestLookup on your own computer today, or try the demo first.