Loki Mode Architecture

An interactive visual map of how Loki Mode transforms a PRD into a deployed product. Click on any component to learn more.

LOKI MODE SYSTEM ARCHITECTURE PRD Input prd.md Planning Opus Model RARV Cycle Build Sonnet Model Parallel Tasks Deploy CI/CD Pipeline Quality Gates 9-gate pipeline 3-reviewer system PROVIDERS Claude Tier 1 Codex Tier 3 Gemini Tier 3 Memory System Episodic / Semantic Procedural / Vector Progressive Disclosure Dashboard FastAPI + React 117 API endpoints WebSocket real-time RARV Cycle Reason Act Reflect Verify Completion Council 3-vote system MCP Server 15 tools Event Bus Python / TS / Bash run_autonomous() build_prompt() git push + CI run_code_review()
Core Pipeline
Planning
Build / Events
Quality
Memory
Deploy / Monitor

PRD-Driven Development

Start with a Product Requirements Document. Loki Mode parses natural language specifications and converts them into structured development tasks automatically.

Markdown PRD 13 templates Natural language

RARV Iteration Cycle

Every iteration follows Reason, Act, Reflect, Verify. Opus handles planning, Sonnet handles implementation, and Haiku manages unit tests and monitoring.

Opus planning Sonnet coding Haiku testing

9-Gate Quality System

Static analysis, 3-reviewer blind review, anti-sycophancy checks, security scanning, test coverage gates, and backward compatibility validation.

ESLint CodeQL Playwright 3 reviewers

Memory System

Three-layer progressive disclosure: episodic traces from individual tasks, semantic patterns generalized across projects, and procedural learned skills.

Episodic Semantic Vector search Token economics

Live Dashboard

Real-time monitoring with 117 API endpoints, WebSocket updates, agent status tracking, and cost analytics. Built with FastAPI and React.

FastAPI React WebSocket 117 routes

Multi-Provider Support

Run with Claude Code (full features), OpenAI Codex CLI, or Google Gemini CLI. Providers degrade gracefully with feature detection.

Claude (Tier 1) Codex (Tier 3) Gemini (Tier 3)