projmind

Overview

projmind is a background daemon that watches your project directories, analyzes source code with local LLMs via Ollama, and builds a structured knowledge base capturing each project's intent, architecture, current state, remaining work, and future directions. Think of it as an always-current project brief that understands your code.

Key Features

• Continuous Scanning: Watches project directories for changes and triggers incremental re-analysis via git diff tracking.
• Deep LLM Analysis: Uses tree-sitter for AST parsing combined with Ollama-powered LLM analysis to understand code structure, patterns, and intent.
• Knowledge Branching: Maintains versioned knowledge snapshots so you can compare how a project's understanding evolved over time.
• Cross-Project Synergies: Identifies shared patterns, potential code reuse, and dependency relationships across your entire project portfolio.
• Digest Generation: Produces human-readable project summaries and status reports on demand.
• Git-Aware: Tracks commits and branches to understand what changed and why, linking knowledge updates to specific code changes.

Technical Architecture

The daemon runs as a background process, periodically scanning configured project directories. For each project, it builds an AST using tree-sitter, extracts structural information, and feeds it through the LLM-Pipeline crate for semantic analysis. Results are stored in a SQLite knowledge base with full-text search.

Core components:

• Scanner: File system watcher with git-aware change detection (gix).
• Analyzer: tree-sitter AST parsing + LLM-Pipeline for semantic understanding.
• Knowledge Store: SQLite database with versioned knowledge entries and FTS5 search.
• Digest Engine: Generates formatted project summaries from the knowledge base.

Technology Stack

• Language: Rust 2021
• CLI: Clap for command-line interface
• Analysis: tree-sitter (AST), LLM-Pipeline, agent-graph (from Rust Libraries)
• Git: gix for native git operations
• Storage: SQLite with FTS5, job-queue for background processing
• LLM: Ollama for local inference

Current Status

Active development with scanning, analysis, knowledge storage, and digest generation all functional. Knowledge branching and cross-project synergy detection implemented. Currently refining LLM prompts for deeper architectural understanding.