Skip to content

Brainstorm AI

Most people are terrible at talking to AI.

Code Video Cerebral Valley

Core trick

Asks 2-3 clarifying questions, then compiles the answers into one structured prompt and routes it to Gemini with stored user preferences.

Tags

productivityagents

Description

Most people are terrible at talking to AI. They type vague, incomplete prompts, often in mixed languages like Hinglish and get mediocre results. Then they go back and forth correcting, re-explaining, frustrated. But most don't knows how to write that perfect prompt. Brainstorm AI fixes this You talk like a normal human, sloppy, vague, in whatever language you think in. The system figures out what you actually mean, asks 2-3 smart clarifying questions (because humans are better at answering questions than writing instructions), then builds a perfectly structured single-shot prompt and fires it at Gemini. One try. That perfect answer. It remembers you across sessions your skills, your preferences, your projects, your language. It selects the right reasoning strategies from a skills database. It connects tools automatically. And it gets smarter every time you use it.
pragnyanramtha/brainstorm★ 11 forksPythonCreated Feb 14, 2026 · last push Feb 14, 2026

README excerpt

# Brainstorm AI

A local-first, intelligent middleware between you and AI models.

## Project Status

- **Backend**: ✅ Complete (Core Engine, API, DB)
- **Frontend**: 🚧 In Progress (Structure built, dependencies pending)

## Prerequisites

- Python 3.11+
- Node.js 18+ (for frontend)
- API Keys: Google Gemini (Required), Anthropic Claude (Optional)

## Quick Start (Backend)

1. **Setup Environment**:
   ```bash
   python scripts/setup.py
   ```

2. **Configure API Keys**:
   Copy `.env.example` to `.env` and add your keys:
   ```bash
   cp .env.example .env
   # Edit .env and add GEMINI_API_KEY
   ```

3. **Run Backend Server**:
   ```bash
   .venv/Scripts/python -m backend.main
   ```
   Server will start at `http://localhost:3847`.
   API Documentation: `http://localhost:3847/docs`

## Architecture

- **Core Engine**:
  - `Intake`: Analyzes intent using Gemini Flash.
  - `Clarifier`: Asks smart questions when ambiguous.
  - `Optimizer`: Builds perfect single-shot prompts.
  - `Executor`: Routes to best model (Gemini/Claude) and handles tools.
  - `Memory`: Stores user profile and context in SQLite.
  - `Skills`: Auto-selects best prompt engineering techniques.
  - `MCP`: Connects to local tools via Model Context Protocol.

- **Tech Stack**:
  - FastAPI, SQLAlchmey, SQLite
  - React, Vite, Tailwind CSS (Frontend)

## Development

To run both backend and frontend (once frontend is fixed):
```bash
python scripts/dev.py
```

Team