Optimized for outcomes. Choose an outcome. Select your code. Get verified tests. Start with the CLI. Add MCP when you want JAIPilot inside your coding agent.
See it run
Apache Kafka
34.3% → 97.1% line coverage for two classes in 1m 53s. Added 13 tests; all 949 module tests passed.
Spring Petclinic — 100% coverage
85.7% → 100% line coverage across the project in 3m 21s. Added 36 tests with coverage_100; all 111 tests passed.
Production code preserved. Actual Terminal recordings; GIFs shorten idle waits. Reported times are full job durations with warm build caches; results vary. Commands, recordings, and full measurements →
How it works
Frontier LLM reasoning. Optimized testing tools. Prompts tailored to each outcome. Parallel workers in the managed service. Independent class tasks can run concurrently; local execution uses your toolchain and supported hardware parallelism.
- Inspect your code, build, and test framework.
- Generate or repair tests for behavior, boundaries, and failure paths, with meaningful assertions.
- Verify with focused tests, the final regression suite, and fresh coverage reports when configured.
90%+ line coverage target for generate_tests, improve_coverage, and lock_behavior, measured in selected production code when coverage is configured. Gaps and blockers are reported. Behavior baselines must pass before edits.
Install
Requires Git, a JDK, and your project's build tools.
macOS / Linux
curl -fsSL https://raw.githubusercontent.com/JAIPilot/jaipilot/main/install.sh | sh
Windows (PowerShell)
irm https://github.com/JAIPilot/jaipilot/releases/latest/download/install.ps1 | iex
npm
npm install -g jaipilot
Homebrew (macOS / Linux)
brew install JAIPilot/tap/jaipilot
Sign in, then run inside your Java repository:
jaipilot auth login
jaipilot run improve_coverage --class com.acme.OrderService
Same JAIPilot account, subscription, and credits. Automatic update checks before workflows.
Install paths, manual updates, and troubleshooting →
Use the CLI
jaipilot run generate_tests --class com.acme.OrderService
jaipilot run test_current_changes --all
jaipilot run lock_behavior --path src/main/java/com/acme
Scope: whole repository, classes, paths, or selected lines. Review the result and git diff before accepting changes.
For a fresh, verified coverage target on a single-module Maven project:
jaipilot run improve_coverage --all --coverage-target 80
The Maven adapter also provides local coverage checks and generation:
./mvnw com.jaipilot:jaipilot-maven-plugin:1.2.1:check -Djaipilot.coverage.line=80
./mvnw com.jaipilot:jaipilot-maven-plugin:1.2.1:run -Djaipilot.coverage.line=80
The Maven adapter 1.2.1 is published on Maven Central. Use mvnw.cmd in Windows
PowerShell. Adapter setup, class/branch targets, and supported JaCoCo rules →
All commands, scopes, JSON output, and exit codes →
Why use JAIPilot if my coding agent already writes tests?
Fast testing. High coverage. Stronger tests. Less work for your coding agent.
- ~5× faster testing than a general coding agent. Optimized prompts, tools, and parallel workers carry the testing workflow. Actual runtime varies by project. CLI examples: 97.1% coverage in 1m 53s on Kafka; 100% in 3m 21s on Petclinic.
- 90%+ line coverage target. Meaningful assertions for selected behavior, boundaries, and failure paths, measured with your configured coverage tools.
- Stronger regression protection. Lock behavior before edits; verify the original tests afterward. Our paired Codex example's saved tests caught 7/10 injected faults vs 3/10 with Codex alone.
- Token efficiency. Offload test planning, generation, and repairs to JAIPilot's optimized service. This can reduce your coding agent's testing context, token use, and costs; JAIPilot credits apply.
Measured example: Codex with JAIPilot
Same refactor. Two fresh Codex sessions. Two Kafka utility classes. The MCP session used JAIPilot automatically, with no repository instruction file or extra testing prompt.
13 more reusable tests. Four more deliberately introduced bugs caught. Same model, settings, initial tests, and warm build caches. One paired example; results vary.
The 90% target applies to selected code. Deadline: 100% (17/17 lines). The combined 65.7% (23/35) includes untargeted LockUtils, unchanged at 6/18 covered lines.
Method, evidence, and reproduction →
MCP for coding agents
Optional: make Codex, Claude Code, or another MCP agent better at Java testing. Give it a tested baseline before edits and real failures to guide repairs.
- Lock:
lock_behaviorcreates and runs characterization tests before production edits, targeting 90%+ line coverage when configured. - Edit: Your coding agent changes production code. Baseline tests stay unchanged.
- Verify:
verify_behaviorreruns the original baseline. Your agent investigates failures and repairs unintended changes before finishing.
Passing tests protect the behavior they exercise. Review old expectations for intentional behavior changes.
Codex
codex mcp add jaipilot -- jaipilot mcp --repo /absolute/path/to/java-repository
Claude Code
claude mcp add --transport stdio jaipilot -- jaipilot mcp --repo /absolute/path/to/java-repository
Start a new agent session. MCP instructions and tool descriptions include the workflow. No repository instruction file required. Tests and commands run locally. Jobs run in the background; 1 hour per local command by default.
Requires a build with the mcp command. Source setup, tool contracts, and timeout options →
Local execution and privacy
Builds and tests run locally. Selected source, project context, and command output go to JAIPilot's managed service and model provider. Sign-in credentials stay local with user-only permissions.
Privacy details · Ask a question · Report an issue · Report a vulnerability privately