• 

  Chirp imports your X profile to generate brand guidelines

• 

  You select what topics Composio should do Deep Research on

• 

  Gemini generates different viral tweet formats in your style

• 

  We can add our human voice

• 

  Every tweet we edit, updates the brand guideline

• 

  It posts directly to X

• 

  Once our voice is refined, we can set YOLO mode to autonomously research + post tweets for us.

Chirp

Inspiration

Generic AI posts are destroying personal brands on X. No sources, cookie-cutter takes, zero value. But manually researching timely stories takes hours. We built Chirp to solve this: an agent that finds real stories, writes in your voice, and learns from every post you make.

What it does

Chirp is a multi-agent system that automates X posting while maintaining authenticity:

1. Search Agent (You.com) - Finds trending, relevant articles with real citations
2. Writing Agent (Google DeepMind) - Generates 3 post variations with different expert angles
3. Publishing Agent (Composio) - Posts to X automatically

The continual learning piece: Every choice you make teaches the system. Pick technical angles? Chirp learns and prioritizes those next time. Edit posts shorter? It adjusts. Over time, it writes like YOU, not generic AI.

How we built it

Tech Stack:

• Next.js 14 + TypeScript frontend
• You.com Search API for real-time article discovery
• Google DeepMind for multi-angle reasoning
• Composio for publishing automation
• Simple preference tracking system (no complex ML needed)

Architecture: Sequential agent workflow with clear handoffs. User provides topic → Search agent finds articles → Writing agent reasons through perspectives → User picks favorite → Publishing agent posts.

Challenges we ran into

Multi-agent coordination: Getting three agents to work together without breaking UX. Solved with clear handoffs and progressive loading states.

Learning without data: Can't learn user voice without training data. Solved by tracking behavioral signals (which angles they pick, what they edit) instead of analyzing text.

Speed vs quality: Multi-step reasoning takes time. Solved with streaming responses and optimized API calls to keep perceived speed under 10 seconds.

Accomplishments that we're proud of

True agentic behavior: Each agent makes autonomous decisions. The search agent decides which You.com queries to run. The writing agent evaluates sources and generates distinct perspectives. No hardcoded rules.

Continual learning that works: Simple preference tracking achieves 70% of fine-tuning benefits with zero model training. Every post makes the next one better.

Citation integrity: 100% of sources are real URLs from You.com. Zero hallucinations.

What we learned

You.com's APIs are built for agents: The Search API returns structured data perfect for autonomous tool use. No parsing messy HTML or dealing with rate limits.

Learning from choices > learning from text: Behavioral signals (which angles users pick) teach faster than analyzing writing samples.

Agents need specialization: One agent that does everything performs worse than three specialized agents working together. Clear roles = better results.

Users trust agents when they see the work: Showing the research, giving choices, making learning visible—transparency builds trust in agentic systems.

Built With

• composio
• conda
• docker
• fastapi
• googledeepmind(gemini)
• lovable
• next.js
• render
• tailwindcss
• typescript
• uvicorn
• you.com