HAIMIDB
Getting started
HaimiDB is a shared memory interface, not a hosted database service. The human table and local AI tools work with the same model and records. GitHub supplies storage and access control; local clones support offline work.
1. Connect GitHub
Open the application, choose Connect with GitHub and approve HaimiDB. Manual personal access tokens remain an advanced alternative.
2. Create and authorize a memory repository
Create a private, empty GitHub repository without a README, license or gitignore. In the GitHub App installation settings, add it under Only select repositories and save. Refresh alone cannot grant access.
3. Initialize and choose a notebook
Return to the app, refresh the repository list, select the empty repository and confirm Initialize memory repository. Name your first notebook; there is no default notebook.
4. Connect your local AI
Install the release archive with npm install --global https://github.com/olerhan/haimidb/releases/latest/download/haimidb-engine.tgz, run haimidb login, then copy the app's MCP configuration into your desktop client. Browser and local sessions are independent.
5. Push and keep history
Push browser changes before closing the tab. Local clones can commit offline and synchronize when connected. If an offline edit conflicts with a deletion, deletion wins and the dropped change is reported.
A table for you
Browse models, search records and edit scalar or structured values in your browser. Changes stay in tab memory until you push them.
Tools for your AI
Eleven MCP tools run locally over stdio. Your AI can inspect models, find records, save changes and work with the same notebook you see.
Readable Git history
Each notebook is a deterministic memory.v2.jsonl file. Commits show what changed; Git history is the recovery source after hard deletion.
Models that explain themselves
Everything is a node. Models carry example values and descriptions; records and linked fields share the same structure.
Repository protection
Memory writes and notebook deletion require a valid .haimidb.json marker. An empty repository is initialized only with your confirmation; an existing project is not silently adopted.
Local-first processing
The browser or local engine processes notebook data and talks directly to GitHub. The small authentication service only exchanges and renews credentials, never notebook content.