Diffblue Cover

Diffblue Cover is a code modifier tool. An autonomous testing agent that generates, verifies and fixes Java unit tests across entire enterprise codebases, working with approved AI coding tools like GitHub Copilot and Claude Code, plus an offline deterministic mode for air gapped environments.

Diffblue Cover is a testing agent built for teams that need serious unit test coverage on large, aging Java codebases. Rather than generating one test at a time from a prompt, it takes a repository as input and works through the whole thing: analyzing existing coverage, repairing build system problems, drafting a test plan, generating tests in parallel, verifying each result, cleaning up the project and preparing a pull request. The pitch is that it finishes the job without a developer watching it. The product comes in two shapes. Batch mode runs from a command line interface against the full project, aimed at hundreds of classes at once. Task mode sits inside the IDE for smaller, on demand requests. It leans on AI coding platforms that enterprises have already approved, with dedicated support for GitHub Copilot and Claude Code, plus Java versions 8, 11, 17, 21 and 25 and Python. Gemini CLI and Codex are listed as coming soon. For regulated environments that cannot send code to a third party model, Diffblue Cover offers a deterministic, offline mode with air gapped deployment and no external dependencies. The audience is clearly enterprise engineering organizations: banks, insurers, defense contractors and healthcare companies that need audit trails and long security reviews. The vendor cites customers including Sun Life, ING, Citi, BNY and MongoDB, and points to an Oxford University spin out heritage in formal methods. Benchmark claims are specific: across eight enterprise Java repositories totaling 31,069 coverable lines, all starting from zero percent coverage, the agent averaged 80.7 percent line coverage versus 32.3 percent for Claude Code with a senior developer, with human intervention limited to setup rather than hours of monitoring. Compared with general coding assistants, Diffblue Cover trades flexibility for reliability. It is not a tool for writing application code or exploring ideas; it is a specialist that generates tests designed to compile, pass and hold up under mutation testing. Pricing is outcome based, charged per verified line of coverage added, which aligns cost with tests that actually work. Teams wanting a fully offline deterministic option should look at Diffblue Cover specifically, while those already standardized on Copilot or Claude Code can layer the agent on top of that stack.