RepoGuardian RHD v3.1.0 E2E Analysis

RepoGuardian RHD v5.1 end-to-end repository analysis workflow.\n\n- Public chat-first repository analysis via persisted serverless DeploymentJob.\n- GitHub source tree/file sync, code document indexing, architecture inference, and persisted SVG/Mermaid artifacts.\n- Follow-up chat uses selected repository/session context; reload restores persisted architecture.\n- Public validation completed on https://repoguardian-rhd.vercel.app with RepoGuardian-Demo and second-repository isolation checks.

RepoGuardian RHD Conversational v5

RepoGuardian RHD v5 redesigns the product around a chat-first repository intelligence workspace while preserving the existing v4 public deployment, deterministic fallback, MCP, evidence grounding, and human-controlled policy.\n\nHighlights:\n- Home route is now the RHD chat workspace.\n- Light premium visual system replaces the dark/neon dashboard-first shell.\n- Desktop left conversation/repository/tool sidebar and right context panel.\n- Premium bottom composer with repository detection, attachment, repository, voice, model/provider, stop/send controls.\n- New read-only /api/v5 workspace, conversations, capabilities, and architecture endpoints.\n- Deterministic Mermaid/SVG architecture artifacts generated from synchronized repository evidence.\n- Multimodal and voice states are truthful: image understanding requires a configured provider, voice remains optional browser-level UX.\n\nValidation completed:\n- Backend pytest: 74 passed.\n- Frontend lint/typecheck/build: passed, 25 static routes.\n- Frontend Playwright: 51 passed.\n- MCP typecheck and smoke test: passed.\n- Public backend health/readiness/v5 smoke checks: passed.\n- Public homepage desktop/mobile browser checks: passed with zero overflow.

RepoGuardian RHD v4 Intelligence

RepoGuardian RHD v4 adds a read-only autonomous engineering intelligence layer on top of the existing public Vercel + Neon deployment.\n\nHighlights:\n- RHD v4 Mission Control and new intelligence routes.\n- Read-only multi-agent mesh with persisted agent run and step traces.\n- Task-aware model gateway vocabulary with deterministic fallback preserved.\n- Agentic/hybrid RAG pipeline metadata with grounding critic.\n- PR risk, blast-radius, incident, code intelligence, model lab, observatory, and release intelligence surfaces.\n- Additive Neon migration 0007_rhd_v4_intelligence.\n- Public users remain read-only and external actions remain human-gated.\n\nValidation completed:\n- Backend pytest: 71 passed.\n- Frontend lint/typecheck/build: passed, 25 static routes.\n- Frontend Playwright: 51 passed across desktop/mobile.\n- MCP typecheck and smoke test: passed.\n- Neon migration applied.\n- Vercel backend and frontend production deployments ready.\n- Public API and v4 route smoke checks passed.

RepoGuardian RHD Public Web

Public Vercel + Neon deployment.\n\nWeb app: https://repoguardian-rhd.vercel.app\nAPI: https://repoguardian-rhd-api.vercel.app\nAPI docs: https://repoguardian-rhd-api.vercel.app/docs\n\nIncludes Vercel FastAPI serverless entrypoint, Neon PostgreSQL/pgvector migrations, Postgres-backed jobs, public rate limiting, read-only public write guards, persisted anonymous sessions, public architecture page, and deterministic RHD fallback.

RepoGuardian RHD Enterprise Preview

RepoGuardian RHD Enterprise Preview

Implemented

• RHD repository review, issue intelligence, review queue, human approval, and audit trail.
• Agentic RAG v2 foundation with query planning, hybrid retrieval, evidence fusion, critic, and objective retrieval metrics.
• Shared RHD Tool Registry used by REST and MCP.
• MCP stdio server with tools, resources, prompts, and smoke tests.
• Managed-cloud deployment profile for Vercel-compatible frontend and Render-compatible FastAPI backend.
• Provider-neutral PostgreSQL configuration with pgvector readiness checks.
• Model Gateway with validated local Ollama adapter and deterministic fallback.

Validated

• Backend pytest suite.
• Frontend lint, typecheck, production build, and Playwright route checks.
• MCP TypeScript typecheck and stdio client smoke test.
• Terraform syntax validation without apply.

Optional

• Groq, OpenRouter, OpenAI, managed PostgreSQL, pgvector, Redis, and GitHub App credentials.

Experimental / Partial

• Durable Postgres queue worker activation requires a live managed PostgreSQL deployment.
• Code-RAG and Graph-RAG have foundations; production graph/vector infrastructure is not connected.
• Remote MCP over Streamable HTTP remains roadmap until auth and rate limiting are configured.

Not Yet Trained

• ML/DL model training requires a defensible labeled dataset. No fake metrics are claimed.

Known Limits

• Docker is not required for managed-cloud deployment and was not available on the local machine.
• Production is not marked validated until user-owned infrastructure is deployed and tested.