CortexQA — Agentic, Local‑First QEProduction Ready
CortexQA is an agentic Quality Engineering platform that orchestrates the entire STLC using specialist agents, a curated MCP registry, and an ACP knowledge graph. It’s local‑first by default, with enterprise‑grade security, governance, and observability.
Capabilities
Context‑Aware Outputs
Agents emit repo‑native code and assets aligned to your conventions, paths, and CI.
ACP & AFP
Application Context Packs and Automation Framework Profiles formalize domain + testing standards.
Self‑Healing Automation
Locator healing, DOM simplification, and rerun strategies reduce flakes and stabilize suites.
Coverage Intelligence
Vector search + graph traces map requirements → tests → defects; drives TIA and priorities.
Human‑in‑the‑Loop
Approvals, rationale, prompts, and diffs enable transparent, auditable agentic decisions.
Multi‑Runner Federation
Playwright, Selenium, Cypress, Tosca, API, SAP — unified orchestration and reporting.
Core Concepts
- Agentic STLC: Autos orchestrate requirements → design → automation → execution → RCA → reporting.
- MCP Registry: Auto‑discovers Jira/ADO/Git/CI, Playwright/Tosca/API runners, Vector/KG tools.
- ACP Knowledge Graph: Domain, APIs, data contracts, workflows, and reusable assets.
- Framework‑Aware Codegen: Repo‑ready components for Playwright/Selenium/Cypress/Tosca/SAP/API.
- Vector Intelligence: Qdrant/pgvector for semantic search, dedup, impact & traceability.
- Human‑in‑the‑Loop: Approvals, rationale reveal, prompt lineage, and audit trails.
Architecture
Orchestrator
ASP.NET/FastAPI with DAG planner, retries/rollbacks, idempotency, and real‑time UI via SSE/WebSocket.
Model Router
Policy‑driven routing across Ollama, OpenAI, Gemini, Claude with cost/latency/PII guardrails and fallbacks.
Execution Grid
Local/K8s runners, sharding, test‑impact analysis, smart retry matrices, artifact retention.
Observability
Agent timelines, coverage heatmaps, failure clustering, flakiness radar, performance budgets.
Deployment & Security
- Deploy on‑prem/private/hybrid; local‑first keeps data on your machine by default.
- SSO/SAML/OIDC, RBAC/ABAC, tenant isolation, secrets vaults, least‑privilege runners.
- Policy‑as‑Code (OPA), audit logs, PII redaction, license guard, WCAG/axe compliance gates.
Data & Storage
- Vector & KG: Qdrant / pgvector powering semantic search, dedup & ACP graph queries.
- Relational: Postgres/Neon for system‑of‑record; SQLite for local jobs/runs/tenants.
- Artifacts: local disk / R2 / S3 with checksums, retention, and encrypted at rest.
Operations & Observability
- Dashboards: agent timelines, coverage maps, defect clusters, SLA adherence.
- Self‑healing: locator healing, DOM simplification reruns, failure bucketing, hypothesis suggestions.
- Budgets: cost ceilings, latency SLOs, flaky/quality gates per pipeline.
Extensibility & MCP
- Add custom agents, tools, and evaluators; publish to the community registry.
- MCP servers for runners (Playwright/Tosca/API), issue trackers, CI/CD, and data sources.
- Recipes: versioned inputs/outputs; templatized prompts and guardrails for reuse.
Advanced Scenarios
Zero‑Touch Onboarding
Fingerprint repos, generate ACP/AFP, and emit native tests with PRs and guardrails.
Regression Triage
Cluster failures, propose fixes, and auto‑open issues with repro steps and artifacts.
Cross‑Repo Impact
Graph analysis to select smoke/regression suites across services after risky changes.
Scenario Authoring
Generate multi‑step flows spanning UI/API/data with seeded fixtures and auth state.
Sandboxed Evaluators
Deterministic evals for locators, assertions, prompts, and agents with benchmarks.
Cost & Latency Budgets
Enforce ceilings per job/model; route for SLAs with fallbacks and retries.
FAQ
Is CortexQA a replacement for my automation framework?
No. It augments existing investments (Playwright, Selenium, Cypress, Tosca, SAP/API) with framework‑aware codegen and reusable assets.
Can we deploy on‑prem?
Yes. On‑prem, private cloud, or hybrid. Local‑first by default. Integrates with SSO/SAML/OIDC and enforces zero‑trust with policy‑as‑code.
Does CortexQA self‑optimize over time?
Yes. Agentic flows learn from execution history, defect patterns, and coverage gaps; they refine locators, prompts, and data to improve each cycle.