AI-powered personal branding agent for Twitter/X. Define your brand "spheres" via free-text descriptions, and the agent generates search queries, discovers relevant web content and news, drafts original tweets grounded in real-time research, and learns from your feedback over time. Built for the Continual Learning Hackathon by Creators Corner (Feb 2026).

• Python 3.11 — Core backend logic
• FastAPI — API framework
• SQLite — Lightweight persistent storage
• HTML / CSS / JS — Frontend dashboard
• Conda — Environment management (hackenv)

• You.com — Search + Live News APIs for real-time content discovery (~10 results per query)
• Composio — Twitter OAuth + tweet posting
• Google Gemini — LLM for query generation + tweet drafting
• Render — Hosting + deployment

1. Authenticate — User logs in with Twitter/X via Composio OAuth
2. Define Sphere — User writes free-text description of what they want to post about (e.g., "AI developer sharing practical takes on new tools")
3. Query Generation — Gemini generates 5 focused search queries from the sphere description
4. Query Review — User can accept queries as-is or tweak wording before search
5. Content Discovery — Each query runs through You.com Search + Live News APIs (~10 results per query)
6. Source Selection — Results displayed grouped by query (title, snippet, URL). User checks which to use as tweet material
7. Tweet Drafting — Gemini drafts original tweets based on selected sources + sphere context + past feedback
8. Draft Review — User approves, edits, or rejects each draft
9. Posting — Approved tweets posted to Twitter/X via Composio
10. Learning Loop — All feedback (approved/edited/rejected drafts, query tweaks, source selections) stored in SQLite and fed into future Gemini prompts

• Local (one command): python run.py — backend on port 3000 (or set PORT).
• Docker (backend + frontend): Runs backend on 3000 and Next.js dev server on 3001.

  docker-compose up

  Then open http://localhost:3000 for the app; the API is at http://localhost:3001 (and /api is proxied from the app).
• Docker (backend only): Build and run the API (and optional pre-built static):

  docker build -t brander-agent . && docker run -p 3000:3000 brander-agent

• Render: Use the repo’s render.yaml (Blueprint). Build runs pip install, builds the React app into static/, then start runs uvicorn main:app --host 0.0.0.0 --port $PORT. Set env vars (e.g. YOUCOM_API_KEY, COMPOSIO_API_KEY, GOOGLE_API_KEY) in the Render dashboard.

git clone https://github.com/farzanmrz/brander-agent-clhack.git
cd brander-agent-clhack
conda env create -f environment.yml
conda activate hackenv

Create a .env file in the project root:

YOUCOM_API_KEY=your_youcom_api_key
COMPOSIO_API_KEY=your_composio_api_key
GOOGLE_API_KEY=your_google_gemini_api_key
DATABASE_URL=sqlite:///./brander.db

Phase 1: Setup (Complete ✓)

• Conda environment created
• Memory bank initialized

Phase 2: Core Implementation (In Progress)

• Create .env.example template
• Initialize FastAPI app structure
• Define SQLite schema (spheres, queries, sources, drafts, feedback)
• Implement Twitter OAuth via Composio
• Implement sphere creation endpoint
• Implement Gemini query generation (5 queries from description)
• Implement query review UI
• Integrate You.com APIs (Search + Live News)
• Implement source selection UI (checklist grouped by query)
• Implement Gemini tweet drafting (sources + feedback → tweets)
• Implement draft review UI (approve/edit/reject)
• Implement tweet posting via Composio
• Implement feedback loop storage
• Deploy to Render

Stretch Goals

• Reply-to-tweets feature (search for tweets to reply to)

4 developers building this in 5.5 hours (11 AM - 4:30 PM PT, demo at 5 PM).

Dev Focus:

• End-to-end working demo over UI polish
• MVP UI: Simple checklists and text inputs
• No over-engineering — plain Python functions, no agent frameworks

This project is licensed under the MIT License.