Hacki
About the project
What is Akritas?
Akritas is an autonomous incident response system for production applications. It monitors workloads deployed through Dokploy, detects runtime failures, investigates their root cause using QVAC locally, and creates actionable GitHub Issues and validated Pull Requests.
How it works
- Akritas continuously retrieves logs from a configured Dokploy application.
- A deterministic detection engine identifies errors, crashes, stack traces, HTTP 5xx responses, and project-specific patterns.
- Equivalent events are fingerprinted and grouped into a single incident, preventing duplicate alerts.
- QVAC investigates the incident locally using logs, stack traces, source code, commits, and repository diffs.
- The investigation produces a structured root-cause assessment, confidence score, relevant files, and recommended actions.
- Akritas always creates a GitHub Issue containing the evidence and investigation results.
- If the problem can be safely fixed in code, Akritas creates an isolated branch, proposes the changes, runs validation checks, and opens a Pull Request linked to the Issue.
Why local AI matters
Production logs and source code can contain sensitive operational and business information. Akritas runs all AI inference locally through QVAC, so this context is never sent to an external AI provider.
The local model does not inspect every log line. Conventional rules handle detection and grouping first, allowing QVAC to focus only on relevant, high-quality evidence.
Safety and human control
Akritas never automatically merges Pull Requests or deploys changes. A remediation must pass its configured validations before a Pull Request can be created, and a human remains responsible for reviewing and approving the final change.
What we built
The project includes a Go backend with PostgreSQL, a Next.js dashboard, secure administrator authentication with TOTP, reusable GitHub and Dokploy integrations, project monitoring, incident investigation, evidence visualization, remediation validation, and the complete Issue-to-Pull-Request workflow.