In progress

Synapse

A from-scratch rebuild of Second Brain OS, a personal productivity app for notes, tasks, projects, and local AI.

  • Notes
  • Tasks
  • Projects
  • Local AI

Started October 2026 · planned through June 2027 · a learning project

Synapse logo: a glowing central node linked to smaller nodes

Local-first

Runs on your machine. Data lives in a folder you control.

Hand-written

Every line of code is typed by me, to learn how it works.

AI as tutor

It explains, reviews and suggests tests. It never writes the code.

Plain SQL

SQLite with no ORM, so every query is visible.

Explore

Start with the plan

A look at the design

The workflow visualiser, sketched

A concept sketch with example data. It shows the planned idea. The real app is still being built.

workflow visualiser · concept sketch

Example data
Overdue
Draft the schemaIn progressHigh
Due today
Write the first migrationIn progressMedium
Due soon
Build the tasks data layerTo doMedium
Add the trash viewTo doLow
On track
Set up CITo doLow

The shape of it

Four parts, one database

Notes, tasks, projects and local AI sit around a single SQLite database. Each part is built by hand, starting with the smallest version that works.

  • NotesPlain Markdown files are the source of truth.
  • TasksStatus, priority and a due date with urgency.
  • ProjectsTasks and notes link to a project. Projects are archived, not deleted.
  • Local AIOllama on localhost. It suggests, and I accept or dismiss.
Synapse architecture Four boxes labelled Notes, Tasks, Projects and Local AI, each connected by a line to a central node labelled SQLite. Notes Tasks Projects Local AI SQLite
Simplified. Notes live as Markdown files, and SQLite holds their metadata and search index.

The plan

Three phases, October 2026 to June 2027

Planned, subject to change.

Why this project

Learning by rebuilding, not by generating

Synapse is a rebuild of a real tool I already use, Second Brain OS (opens in a new tab). I write every line of code myself so I understand how the architecture fits together.

AI acts as a reviewer and tutor: it explains concepts, points out bugs and suggests tests, but it does not write the implementation.

Follow the build

The code is on GitHub, and each session gets an entry in the learning log.