ASQS — Autonomous Software Quality System

ASQS is a CI-native bot that processes small incremental updates to your codebase, finding untested code, writing unit and E2E tests, verifying them in a real sandbox, and opening pull requests — with first-level support for self-hosted, open-source models (Llama, Codestral, Qwen) to deliver maximum privacy, security, and cost reduction.

Self-hostable · Local open-source models · Small incremental updates · Full audit trail

How ASQS Works

Step 01

Index

Parses your codebase into a symbol graph of classes, methods, and their relationships.

Step 02

Plan

Ranks test gaps by risk and centrality so high-value code gets covered first.

Step 03

Retrieve

Fetches the right context: similar tests, dependencies, fixtures, your team's skill files.

Step 04

Generate

An LLM writes the test or doc, informed by your codebase's own conventions.

Step 05

Sandbox

Compiles and runs the generated code in isolation. Repairs failures automatically.

Step 06

Ship

Opens a pull request with green, reviewable tests. Logs everything for audit.

Built for Enterprise. Designed for Engineers.

Graph-based gap planning

Prioritises the riskiest, most-depended-on code using structural graph signals, not random file ordering.

Hybrid retrieval + MMR

Finds the most relevant and diverse context from your codebase so the LLM writes tests that fit your conventions, not generic boilerplate.

Sandbox & auto-repair

Every generated test is compiled and run. Failures are fed back to the LLM fixer and retried. Only green tests ship.

Policy as configuration

Coverage gates, gap caps, commit templates, and quality rules live in version-controlled YAML — not in someone's head.

First-class open-source models

Native, first-level support for self-hosted models (Llama, Codestral, Qwen) via Ollama. Achieve absolute code privacy and slash API costs while running fully offline.

Full audit trail

Every run is recorded: what was attempted, what passed, what was discarded, and the exact prompts sent to the LLM. Downloadable for compliance review.

Bootstrapping

ASQS can cold-start on a legacy codebase with zero existing tests. It builds an initial coverage baseline autonomously, prioritising the highest-risk entry points first.

Small incremental updates

Avoid massive, unreviewable PRs. ASQS generates small, focused test batches and indexes only modified files, keeping changes clean and updates quick.

Pre-generation seam

Before writing a single test, ASQS analyses the target code for testability. It identifies tightly coupled dependencies and proposes interface seams or injection points, making the code itself easier to test.

Quality gates

Define hard thresholds for coverage percentage, cyclomatic complexity, and mutation score in your policy YAML. ASQS enforces them on every run and fails the CI job if they are not met.

Multi-language support

First-class support for Java, C#, and TypeScript/JavaScript out of the box, with language-specific indexers, sandbox runners, and test framework integrations (JUnit, NUnit, Vitest, Playwright).

Per-step LLM routing

Assign a different LLM model to each pipeline step — a fast cheap model for planning, a powerful one for generation, a local model for sensitive sandboxing — and tune cost vs. quality at every stage.