llm-reviewer 0.2.0

The persistent memory layer for AI coding agents

ProjectMind AI

A persistent memory layer for AI coding agents. ProjectMind analyzes your codebase, builds a structured knowledge base, and generates context-enriched prompts — so every AI interaction starts with full project awareness instead of a blank slate.

What it does

Most AI coding tools start from zero every session. ProjectMind fixes that by:

• Extracting your project's DNA (language, frameworks, architecture, patterns)
• Running static analysis (circular deps, dead code, security issues, duplicates)
• Storing a persistent memory of decisions, errors, and patterns across sessions
• Compressing all of that into a token-efficient context that fits inside any LLM's window
• Generating enriched prompts that give agents everything they need upfront

Features

Feature	Description
Project DNA Engine	Detects language, frameworks, architecture pattern, DB, auth, deployment
Architecture Analyzer	AST-based static analysis — cyclomatic complexity, circular deps, dead code, security scan
Persistent Memory	SQLite-backed store for tasks, decisions, known errors, and patterns
Vector Memory	ChromaDB semantic search across all memory types (no external server)
Token Compression	Converts .projectmind/ files into a compact JSON context — no LLM required
Smart Prompt Generator	Task + compressed context + relevant memories → enriched agent prompt
GitLab MR Reviewer	Two-stage LangChain review chain that posts directly to merge request comments
REST API	FastAPI backend with endpoints for all features
CLI	projectmind command for init, analyze, compress, generate-prompt, memory

Stack

• Python 3.12 + Poetry
• FastAPI — REST API backend
• LangChain — LLM chains and RAG pipeline
• SQLModel — SQLite memory persistence
• ChromaDB — embedded vector store (no server needed)
• NVIDIA NIM / OpenAI / Anthropic / Ollama — LLM providers
• Docker — containerized deployment

Quick Start

Prerequisites

• Python 3.12+
• Poetry

Install

git clone <your-repo-url>
cd projectmind
poetry install

Configure

Copy .env.local and fill in your keys:

# LLM (NVIDIA NIM recommended — access to 50+ models)
API_URL=https://integrate.api.nvidia.com/v1
API_KEY=your-nvidia-api-key
CODE_MODEL=meta/llama-3.1-70b-instruct
CONVERSATION_MODEL=meta/llama-3.1-8b-instruct
LLM_PROVIDER=nvidia

# GitLab (only needed for MR reviewer)
GIT_TOKEN=your-gitlab-personal-access-token
GIT_BASE_URL=https://gitlab.com

Initialize a project

# Analyze your project and build the .projectmind/ memory directory
projectmind init /path/to/your/project

# With LLM-enhanced architectural summary
projectmind init /path/to/your/project --llm --provider nvidia

Run static analysis

projectmind analyze /path/to/your/project

Output: circular dependencies, dead code, duplicate functions, security issues, health score.

Compress context

# Get a token-efficient JSON summary of your project (no LLM needed)
projectmind compress /path/to/your/project

# With budget and text preview
projectmind compress /path/to/your/project --budget 6000 --show-text

Generate an agent prompt

projectmind generate-prompt /path/to/your/project --task "add JWT authentication to the API"

Outputs a context-enriched prompt ready to paste into any AI coding agent.

Manage memory

# List stored memories
projectmind memory list /path/to/your/project

# Record a decision
projectmind memory add-decision /path/to/your/project

# Record a known error + fix
projectmind memory add-error /path/to/your/project

# Semantic search across all memories
projectmind memory search /path/to/your/project --query "authentication pattern"

Start the API server

projectmind serve
# or
poetry run serve

API docs available at http://localhost:8000/docs

API Endpoints

Method	Endpoint	Description
POST	/analyze	Extract project DNA and initialize .projectmind/
POST	/architecture	Run full static analysis
POST	/compress	Compress project context to token-efficient JSON
POST	/prompt/generate	Generate context-enriched agent prompt
GET/POST	/memory/tasks	Task memory CRUD
GET/POST	/memory/errors	Known error memory CRUD
GET/POST	/memory/decisions	Decision memory CRUD
GET/POST	/memory/patterns	Pattern memory CRUD
GET	/memory/search	Semantic search across memories
POST	/review	GitLab MR code review
GET	/health	Health check

Docker

# Start API + Redis
docker compose up --build

# API at http://localhost:8000
# API docs at http://localhost:8000/docs

Project structure

projectmind/
├── backend/
│   ├── api/              # FastAPI app and routes
│   ├── core/
│   │   ├── analyzer/     # AST parser, dep graph, dead code, security
│   │   ├── compression/  # Token budget + compressor
│   │   ├── dna/          # Project DNA extractor + generator
│   │   ├── memory/       # SQLite + ChromaDB memory store
│   │   └── prompt/       # Smart prompt generator
│   ├── git/              # GitLab client
│   ├── llm/              # LLM provider abstraction + prompts
│   └── vector/           # Embeddings + vector store
├── cli/                  # Click CLI (projectmind command)
├── .projectmind/         # Generated per-project memory directory
│   ├── architecture.md
│   ├── coding_style.md
│   ├── decisions.md
│   ├── memory.db
│   └── embeddings/
└── pyproject.toml

LLM Providers

Provider	How to use
NVIDIA NIM	Set LLM_PROVIDER=nvidia, API_KEY=nvapi-... — access to Llama, Mistral, Phi, and more
OpenAI	Set LLM_PROVIDER=openai, API_KEY=sk-...
Anthropic	Set LLM_PROVIDER=anthropic, ANTHROPIC_API_KEY=sk-ant-...
Ollama	Set LLM_PROVIDER=ollama, API_URL=http://localhost:11434 — fully local

Requirements

• No LLM key needed for: init, analyze, compress, generate-prompt (template mode)
• LLM key needed for: --llm flag on init/generate-prompt, GitLab MR reviewer
• GitLab token needed for: MR reviewer only