Showcase

Work we're proud to share

The kind of detail most agencies skip — starting with the things we build for ourselves.

2026Desktop app · Our own productLocal-first AI for your Obsidian vault

Chat with your notes — without your notes leaving the building.

PKMA is a desktop AI assistant for Obsidian users. It answers questions from your own notes with real citations, and every byte of it runs on your machine — no accounts, no uploads, no telemetry.

PKMA — research-vault127.0.0.1:8008
What did I actually conclude about spaced repetition?
AI

spaced-repetition.md·exactreading-log.md·ftsactive-recall.md·fuzzy
Deep Research · 3 hops · hybrid0 bytes left this machine

↑ not a screenshot — it streams, same as the real app

Why we built it

We built PKMA because we wanted it to exist: AI that knows your notes but never phones home. It's also our standing proof — when we say we build fast, private, properly-engineered software, this is what we mean.

Hybrid retrieval

Full-text and semantic vector search, fused. It finds the note you only half-remember writing.

Wikilink-aware

Follows your [[links]] up to three hops, so answers carry the context around a note — not just the note.

Three agent modes

Fast, Normal, and Deep Research — each with its own tool budget, so you choose speed or depth.

Citations you can check

Every claim gets a pill — exact, full-text, or fuzzy match — that links back to the source line.

Works air-gapped

Point it at Ollama or LM Studio and unplug the router. Nothing changes.

Zero telemetry, ever

No accounts, no tracking, no phone-home. Your notes stay at 127.0.0.1.

Under the hood

RustTauri v2AxumTokioSQLite + FTS5sqlite-vecNext.js 16React 19

This spot is reserved.

We're taking on a handful of projects this month — one of them could end up here, before/after numbers and all. We only publish the ones we're proud of, which is why we work hard on every single one.

Start the conversation