Mihai Builds
open-source self-hosted developer tools
A compounding stack of small, focused products you run on your own hardware. Memory Vault is the substrate that stores and retrieves what your AI assistants need to remember. The Brain is the engine that runs workflows on top of it — Python-defined, four trigger types, four step types, every run persisted. Both Postgres-backed, MIT-licensed, one multi-arch Docker image each. No cloud lock-in.
Memory Vault — what's in v1.0
Open-source · MIT · self-hosted · v1.0 released 2026-05-07
Hybrid search
pgvector HNSW for semantic similarity + tsvector GIN for keyword + Reciprocal Rank Fusion to merge them. Vector-only misses exact terms; keyword-only misses paraphrase. RRF gets both.
MCP server for Claude
Four tools — recall, remember, forget, status — callable from Claude Desktop or Claude Code. Claude reads and writes your memory inside any conversation, no copy-paste.
Knowledge graph
spaCy NER + co-occurrence extracts entities and relationships from every chunk. No LLM, no per-token cost. Cytoscape force-directed graph with pan, zoom, click-through.
Local LLM chat with sources
LM Studio integration with a sources panel showing the retrieved chunks behind every answer. Every response is grounded and the grounding is visible.
REST API
Bearer-auth-protected, OpenAPI-documented at /docs. Every dashboard operation is also a documented endpoint — integrate AI memory into any app, any language.
One-command Docker
Postgres, the app, and the spaCy model bundled into a single image. docker compose up -d, open localhost:8000. 163 tests passing in CI.
Try it now
Memory Vault product page → · GitHub repo · Latest release · Launch post
The Brain — workflow orchestrator for the ecosystem
Open-source · MIT · self-hosted · v1.0 released 2026-06-15
Python-defined workflows
A workflow is a Python file with named steps. Full editor support, full type checking, zero new syntax to learn. The workflow file is the source of truth — not YAML, not a visual builder.
Four step types
Shell subprocess, OpenAI-compatible LLM with per-step overrides, Memory Vault REST, or any MCP tool over stdio. Each step is its own subprocess where it matters; a crashing step never takes the runner down.
Four trigger types
Manual CLI invocation, cron schedules, HMAC-verified webhooks, or filesystem changes via a debounced watcher. The full classical trigger surface — every workflow runs the same way regardless of how it was started.
Ecosystem-aware
Memory Vault is the substrate; The Brain is the engine that composes it. A workflow can call Memory Vault over MCP, pipe the result into an LLM, and write a digest — two products, one Docker network.
Persistent run history
Every run is persisted to a single Postgres database. brain history, brain show <run-id>, structured logs with run_id binding. No more "what ran last Tuesday?"
Multi-arch Docker
linux/amd64 and linux/arm64 images at ghcr.io. docker compose --profile api up -d, the HTTP API on :8001. 363 tests passing in CI.
Try it now
The Brain product page → · GitHub repo · Latest release · Launch post
Other things I've built
Database Toolkit for .NET Developers
NHibernate Code Generator
Generate entities, mappings, and repositories from a JSON configuration file. Full NHibernate project scaffold in seconds.
EUR 10 View on GumroadSQL to C# POCO Generator
Generate clean C# model classes from SQL CREATE TABLE statements. Drop your SQL, get your POCOs.
EUR 10 View on GumroadNHibernate XML to Fluent Converter
Convert NHibernate .hbm.xml mappings to Fluent NHibernate C# classes. Migrate away from XML mappings.
EUR 10 View on GumroadSchema Documentation Generator
Generate complete Markdown documentation from SQL Server schema files. Every table, column, type, and constraint.
EUR 10 View on GumroadSchema Drift Detector
Compare SQL schema against C# models. Find missing columns, type mismatches, and nullable inconsistencies.
EUR 15 View on GumroadChange Script Generator
Compare two SQL schemas and generate ready-to-run ALTER TABLE scripts. Column adds, drops, type changes.
EUR 15 View on GumroadAbout
I'm Mihai. 15+ years writing C#/.NET, Oracle, NHibernate, and industrial automation software. I've spent my career deep in legacy codebases — the kind where XML mappings are everywhere and schema changes break things silently.
The .NET database toolkit came from there — six tools that each removed a friction I kept hitting at work. The AI ecosystem above (Memory Vault, The Brain, and the next products in the stack) is a deliberate move into building open-source self-hosted developer infrastructure: small, focused products that compose, all MIT-licensed, all running on your own hardware.
I build these in public, on evenings and weekends. The blog tracks the work as it happens.