I'm genuinely obsessed with how software is built — not just making it work, but making it clean, principled, and durable. I care deeply about system architecture, sound engineering principles (SOLID, clean boundaries, testability), performance, and developer experience.
I gravitate toward fast, private, no-nonsense tools — the kind that run as a single binary, work offline, and respect your data. Every project I ship is an exercise in getting the fundamentals right.
Architecture · Clean Code · Design Principles · Performance · Local-first · AI Tooling · Rust · Systems Programming
- 🏛️ Architecture first — I design around clear boundaries and stable interfaces (traits/contracts), so systems stay flexible as they grow.
- 🧱 Principles over shortcuts — SOLID, separation of concerns, and dependency inversion aren't buzzwords to me; they're how I keep code maintainable.
- ✅ Tested & verifiable — a clean, fully tested core is non-negotiable. If it can't be verified, it isn't done.
- ⚡ Performance & footprint — single binaries, no bloated runtimes, no unnecessary dependencies.
- 🔒 Privacy by default — your data stays yours; local-first is the baseline, not a feature.
🧠 Featured Project — YourBrain (yb)
A local-first memory engine that gives AI coding agents a persistent, coherent long-term memory — with first-class conflict resolution.
Today's AI coding agents are brilliant but stateless. Every new chat starts from zero — they forget your architecture decisions, burn tokens re-reading pasted context, and confidently contradict facts that changed three refactors ago. Most "memory" add-ons are just an append-only vector store that slowly rots into a pile of conflicting half-truths.
MCP gave agents tools. It did not give them a coherent long-term memory. YourBrain does.
| Without YourBrain | With YourBrain | |
|---|---|---|
| 🔁 Context | Re-explain everything each session | Agent recalls prior decisions instantly |
| ⚔️ Conflicts | Memory silently goes stale | Conflicts detected & resolved on write |
| 💸 Cost | Big prompts, high token spend | Token-budgeted, compressed recall |
| 🎯 Accuracy | Confident hallucinations | Answers fact-checked against your KB |
| 🔒 Privacy | Cloud lock-in, data leaves your machine | 100% local, single binary, your data stays put |
The differentiator is the conflict engine. When new knowledge duplicates, supersedes, or contradicts what was stored before, YourBrain detects it, auto-resolves the safe cases, asks a human for the ambiguous ones, and keeps a full audit timeline + relationship graph so history is never lost.
- 🧩 Automatic conflict resolution — duplicate / supersede / contradiction detection, auto or ask
- 🔍 Hybrid search — SQLite FTS5 keyword + vector similarity fused with Reciprocal Rank Fusion, then BM25 reranked
- 🗜️ Code-safe compression — preserves code, paths, and URLs while shrinking tokens
- 🛡️ Guardrail / fact-check —
yb validateflags unsupported claims before an agent acts - ⚡ KB-grounded semantic cache — reuse grounded answers with provenance-based auto-invalidation
- 🧠 Optional ONNX embeddings — e5 / MiniLM / BGE for true meaning-based recall
- 👥 Team knowledge — endorse / dispute with consensus confidence, share via JSONL
- 🔌 Works in your IDE — MCP server for Cursor, Claude Code, VS Code & Trae
# Build (needs a Rust toolchain 1.75+)
git clone https://github.com/mrapry/YourBrain yourbrain && cd yourbrain
cargo build --release
# Store and recall knowledge
yb remember "Backend API uses Rust with the Axum web framework" --tag backend
yb recall "how is the backend built"
# Wire it into your IDE
yb install --ide cursorGive your AI agent a memory that doesn't rot → github.com/mrapry/YourBrain
🦀 Rust · 🧠 AI memory · 🗄️ SQLite · 🔌 MCP · 🔒 Local-first
If YourBrain sounds useful, a ⭐ helps others find it — and PRs are always welcome.