asklokesh/loki-mode

Multi-agent autonomous SDLC framework. Spec to deployed app. PRD, GitHub issue, OpenAPI/JSON/YAML, or one-line brief. 5 AI providers, 8 quality gates.

What it solves

Loki Mode is an autonomous coding agent that transforms a specification (such as a PRD, GitHub issue, or OpenAPI spec) into a fully functional, verified software project. It solves the problem of "hallucinated completion," where AI agents claim a task is finished but the code is broken or missing. Loki ensures that work is only marked as "done" when it passes deterministic verification gates, including automated tests and build checks.

How it works

Loki operates using a RARV-C cycle (Reason - Act - Reflect - Verify - Close). It classifies the complexity of the request and assembles a team of specialized agent roles across eight domains to execute the build.

Key mechanisms include:

  • Quality Gates: Eight deterministic checks (build, tests, static analysis, secret scans, etc.) that must be passed before completion.
  • Evidence Receipts: A tamper-proof record of facts (git diffs, exit codes) and AI assessments, allowing users to independently verify the build's success.
  • Provider Agnostic: It orchestrates other coding CLIs (like Claude Code, Aider, or Cline) to perform the actual editing.
  • Living Specs: It can lock specifications to detect drift and run "Devil's-Advocate" interrogations (loki grill) to find gaps in the spec before building.

Who it’s for

Loki is designed for developers and teams who need production-ready code from AI agents without having to manually verify every change. It is particularly useful for those building full-stack applications, modernizing legacy systems, or requiring a high-trust, audit-ready development workflow.

Highlights

  • Verified Completion: Refuses to accept "done" if the git diff is empty or tests fail.
  • Evidence Receipts: Generates a machine-readable proof.json and HTML reports to prove work was completed.
  • Full-stack Output: Generates source code, tests, Docker Compose stacks with healthchecks, and CI/CD pipelines.
  • Cross-project Memory: Uses vector search to apply knowledge learned from one project to the next.
  • Standalone Verification: The loki verify command can check any branch or PR diff against its quality gates.
  • MCP Server: Includes a Model Context Protocol server with 34 tools for enhanced agent capabilities.