Open-source AI quality engineering platform
An AI-augmented quality engineering platform where engineers upload artifacts and receive structured, production-quality outputs — test cases, pytest suites, and root-cause analyses. AI powers the analysis; the platform owns the workflow.
The hosted demo is a static UI preview — API features require a running backend, exactly as documented in the repository.
Inputs
Requirement documents
Analyzed into structured test design
OpenAPI specifications
Parsed for endpoints and auth schemes
CI failure artifacts
Logs, stack traces, screenshots, page source
Analysis engine
Pluggable adapters route work to the right model while output contracts stay deterministic — every response validated against typed schemas before it reaches application state.
Outputs
Structured test cases
Steps, boundaries, edge cases, risk ratings
Executable pytest suites
Generated fixtures, ZIP export
Root-cause reports
Severity, confidence, suggested fixes
Verified from the public repository — no invented features
Each module targets repetitive, high-cognitive-load QA work — and each returns structured, validated output rather than chat text.
Requirements → structured test design
Product requirements go in; strongly typed analysis comes out. The platform produces prioritized test cases with steps and expected results, boundary values, edge cases, risk assessments with severity, and automation-feasibility ratings — the output a test lead would spend days writing.
OpenAPI spec → ready-to-run pytest suites
Paste an OpenAPI specification and receive AI-generated pytest API suites packaged with generated conftest fixtures, auth-header handling derived from the spec's security schemes, and ZIP export. Coverage that used to be skipped because it was tedious becomes routine.
CI logs → root causes with suggested fixes
CI/CD logs, stack traces, screenshots, and page source are analyzed into structured root causes — each with severity, confidence, affected components, and suggested fixes mapped to root-cause IDs. Batch mode processes whole folders of failures at once, with Markdown/JSON/CSV export.
A registry maps provider names to adapters behind one protocol. Business modules never know which model serves a request — adding a provider means writing one adapter class. Shipped adapters: OpenRouter (cloud), Ollama (local), Gemini.
Prompts live as versioned Markdown files with Jinja2 templating and few-shot examples — no hardcoded strings in Python. Prompts improve without code changes or redeploys of business logic.
Model output is parsed and validated against Pydantic schemas before it reaches application state. Malformed AI responses become typed errors, not silent corruption.
Retry with backoff and fallback provider chains keep analyses running through provider outages, surfaced through a health dashboard.
JWT-based registration and login with bcrypt hashing and role-based access across admin, user, and viewer roles.
Team workspaces with member roles, session sharing, and per-user plus per-team API rate limiting.
An audit log records platform events into a paginated activity feed with team, user, and action filtering.
HMAC-signed outbound webhooks with retry and backoff, plus in-app and SMTP email notifications with per-user preferences.
Every analysis is a stored, searchable session — comparable side-by-side across runs and exportable as Markdown, JSON, or CSV.
Runs locally via Docker Compose or one-command installers for Windows, Linux, and macOS. Data stays on your machine unless you choose a cloud provider.