Qodo Cover generates unit tests to raise a project's measured code coverage. It runs from a terminal or inside a GitHub CI workflow, taking a source file, a test file, a test command, and an existing coverage report, then iterating until a target coverage percentage or an iteration cap is reached. A repository mode scans a whole codebase, identifies the test files, collects context for each, and extends the existing suites.

Four components make up the loop: a test runner that executes the suite and produces the coverage report, a coverage parser that verifies each added test actually increases coverage, a prompt builder that gathers code context, and an AI caller that requests the tests. Worked examples cover Python with pytest-cov, Go with gocov and gocov-xml, and Java with Gradle and JaCoCo; coverage is read from Cobertura XML or JaCoCo CSV reports.

Model access goes through LiteLLM, so any model it supports can be selected with a --model flag — OpenAI by default, Vertex AI, Azure deployments, and OpenAI-compatible endpoints via --api-base. A record-and-replay mode stores model responses under a hash of the source and test files so repeated runs reuse them instead of calling the model again. Each run writes a log and an HTML results table listing every generated test's status, failure reason, exit code, stdout, and stderr, and setting a Weights & Biases key additionally logs prompts and responses there.

As of June 2025 the repository is marked as no longer maintained, with forking suggested for anyone continuing development.

Features

Iterative generation: tests are generated, executed, and kept only when they pass and increase measured coverage
Coverage targets: a desired coverage percentage and a maximum-iteration cap bound the loop
Language examples: Python with pytest-cov, Go with gocov and gocov-xml, and Java with Gradle and JaCoCo
Coverage formats: Cobertura XML and JaCoCo CSV, with a documented path for adding new coverage types
Repository mode: scans a codebase, auto-identifies test files, and extends every existing suite
Flakiness check: generated tests can be re-run several times to detect flaky results
Model choice: LiteLLM routing to OpenAI, Vertex AI, Azure, and OpenAI-compatible endpoints
Record and replay: model responses cached per source-and-test hash and replayed on later runs to save credits
Run outputs: a run log and an HTML table of per-test status, failure reason, exit code, and output, suppressible with a flag
Experiment logging: optional Weights & Biases logging of prompts, responses, and run metadata

Discussing Qodo-Cover· 0

0 threads · General