Sensei · the patient observer

A quiet companion
for AI-assisted work.

Sensei watches your sessions with AI assistants — then surfaces the patterns you're too close to see. Not a chatbot. Not a copilot. A patient observer.

Free preview · Local-first · No account required
Sensei · Today
Tuesday, March 12

Good morning, Aiko.

Pattern recurring
You've reached for useEffect three times this week when state could've stayed local. Worth a closer look?
this week
Also worth noticing
Refactor compounding well kazoku-app
Tests skipped 4 sessions shoji-ui
New idiom emerging in shoji-ui shoji-ui
Adopted teachings
· Prefer local state
· Co-locate tests
· Keep async at edges
+ 21 more
0
external requests
<60MB
memory footprint
MCP
open protocol
Preview
free during preview

One desktop app.
One quiet promise.

Sensei runs on your machine and observes your sessions with AI assistants. It logs nothing remotely; it speaks rarely; it remembers what you've actually done.

Over weeks, it begins to recognize your patterns — the idioms you gravitate toward, the workarounds you've adopted, the friction points that keep recurring. When something looks worth noticing, it tells you. The rest of the time, it stays out of the way.

観 · 察 · 覚
Watch, notice, adopt.

Watch

It sits beside you

Sensei sits beside your editor and AI tools, capturing the shape of each session — the prompts, the responses, the corrections.
Local only. Nothing leaves your machine.
Notice

It begins to see

After a few days, patterns surface. Recurring frictions. Idioms forming. Things you taught the assistant once and may want to teach it again.
You decide what's signal and what isn't.
Adopt

It remembers, with consent

Worthy patterns become memories — small, named lessons sensei applies to future sessions on your behalf, with your blessing.
Adopt, refine, or dismiss. Always your call.

The master observes for a long time before teaching.

AI tools are getting louder. More suggestions, more autocompletes, more interrupting. Sensei moves the other way. It speaks rarely, and only when it has something specific to say. Most days it is completely silent — and that is the feature.

The kanji throughout the app are not decoration. Each one names a phase of practice — observation, recognition, adoption, refinement. They are what we ask of the user, and what we ask of ourselves as the people who built this.

Your sessions stay on your machine.

Local storage only
Transcripts, patterns, and memories are stored in a local PostgreSQL database on your machine. Sensei never makes outbound network requests beyond the AI assistant you already use.
No telemetry
We don't track usage. Updates are checked manually from Help → Check for Updates. Local inference via Ollama means even model calls stay on your hardware.
Easy to delete
One data directory. Delete it and sensei forgets everything. Export to JSON anytime.

Free during preview.

Sensei is in early preview — we're learning what works and what doesn't. It's free while we figure that out. If we move to a paid tier, early adopters and supporters get a permanent discount. No surprises.

Download for macOS

The essentials.

What platforms?
Claude Code, Cursor, Windsurf, Copilot, Codex, Aider — anything that speaks MCP.
What's included?
20 commands, 8 agents, skills, hooks, and MCP tools for code search, patterns, and call graph analysis.
Will it slow me down?
Rust daemon, event-driven. Ollama is optional and hardware-aware — it degrades gracefully.
Is it free?
Free during preview. If we move to a paid tier, early adopters get a permanent discount.
All questions & answers →

If sensei has earned a place in your practice, you can help keep it growing.

Sensei is built by a small team. Every coffee buys an hour of focused work.

♥ Buy me a coffee