Build a custom CRM for tracking contacts, farms, or any collection of records without writing SQL migrations by hand.
Prototype a database backed app where the schema evolves through natural language requests instead of code changes.
Learn how to constrain an AI model to a safe, typed set of actions instead of letting it generate raw SQL.
| leandrobon/mutable-crm | 0xkinno/neuralvault | 0xmayurrr/ai-contractauditor | |
|---|---|---|---|
| Stars | 1 | 1 | 1 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | moderate | hard | easy |
| Complexity | 3/5 | 4/5 | 2/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires Docker and an Anthropic API key.
mutable-crm is a CRM, a customer database tool, whose structure changes based on what you type to it rather than requiring you to write code or SQL yourself. When you type something like "contacts need a company field," an AI model proposes a specific change to the database's structure. You review that proposal, click to apply it, and the app's tables rebuild themselves to match. Ask for something bigger, like "a CRM to track my farm," and the model can propose a whole new set of tables at once. The interesting design choice is that the AI model is never asked to write SQL directly. Instead it picks from four specific actions: create tables, add a column, rename a column, or change a column's type. The application itself turns that choice into the actual SQL. This means the model cannot produce broken SQL syntax, and it cannot attempt to delete or drop data because that option simply does not exist in its toolkit. The model also never sees the data stored in your CRM, only the names and types of the columns, so your actual records stay private from the model. Every change made is stored along with its exact reverse, so it can be undone later. Undo only works on the most recent change still in effect, since undoing an older change could conflict with something built on top of it afterward. Undoing an added column deletes whatever was written into it, so that kind of undo requires a second confirmation click. The tool is built with Next.js, TypeScript, Postgres running in Docker, Drizzle, and the Anthropic SDK for the AI model. It needs Docker and Node 20 or newer to run locally, plus your own Anthropic API key. It currently has no support for user accounts, multiple users, permissions, or production deployment, and there is no way to delete tables or columns. It is meant for single-user, local use only, and is released under the MIT license.
A local CRM where you change the database structure by typing requests to an AI, which proposes safe schema changes instead of writing SQL.
Mainly TypeScript. The stack also includes TypeScript, Next.js, PostgreSQL.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.