Harbor stores your knowledge in Markdown and SQLite — readable by humans and AI alike. Every proposed change surfaces as a reviewable diff before it touches your data.
AI changes are shown as paired before/after diffs — never inline strikethrough, never silent. Approve, reject, or undo any hunk. Harbor records who proposed what, when, and why.
Why: Implied from the kickoff conversation. Kari needs the structured block schema before her Thursday design review.
(empty)
:::task id: task_01HY8Z… title: "Send the structured block schema to Kari" due: 2026-05-14 project: ai-knowledge-base :::
One calm place for notes, plans, ideas, people, projects, tasks, and preferences. Written in Markdown, stored as your own files.
Harbor doesn't just store information. It structures it — people, tasks, preferences — so AI tools can search, understand, and update it reliably.
# Kari Nordmann
:::person
id: person_01HY8Z4KTSV4RRFFQ69G5FAV
name: Kari Nordmann
relationship: Friend
birthday: 1990-04-12
:::
## Preferences
| Topic | Value | Source |
|---|---|---|
| Meetings | Async written reviews | User note |Every AI write becomes a reviewable patch. Approve, reject, or undo — per hunk, per agent, per folder. No silent edits, ever.
Plain Markdown on disk. SQLite for structure. Export everything, anytime. Your knowledge stays readable and yours — not trapped inside a model's memory.
Harbor speaks MCP — the open protocol AI tools use to talk to other systems. Connect calendars, mail, and code repos; let ChatGPT, Claude, and other clients search your Harbor directly.
Harbor is in private beta. Free during beta — legacy users keep free hosted access after launch.