Twitter Agent

An AI-powered content automation system that researches trending AI topics, writes authentic developer-focused posts, and schedules them optimally on X (Twitter). Built with Mastra and TypeScript.

What it does

The system runs a three-stage agentic pipeline:

1. Researcher — Searches the web and X for trending AI news, model releases, infra patterns, and developer tools from the last 7 days. Produces a structured research brief with 5–8 content angles.

2. Writer — Transforms the research brief into 4–6 authentic X posts (individual tweets or threads). Writes in a practical, builder-focused voice: no hype, no emojis, no em dashes.

3. Scheduler — Assigns optimal posting times based on developer audience activity windows (EST/PST), with a minimum 90-minute gap between posts.

Stack

• Mastra — Agent orchestration, memory, storage, and observability
• xAI Grok — Powers the researcher, writer, and scheduler agents
• X API (XDK) — Real-time X search for trending posts
• Hono — HTTP server with Mastra integration
• LibSQL + DuckDB — Local storage and observability data
• dotenvx — Encrypted environment variable management

Setup

bun install
cp .env.example .env
# Fill in your API keys in .env

Required environment variables:

Variable	Description
XAI_API_KEY	xAI API key for Grok models
X_API_BEARER_TOKEN	X API bearer token
X_API_CLIENT_ID	X OAuth client ID
X_API_CLIENT_SECRET	X OAuth client secret
DATABASE_URL	Neon database connection string

Usage

# Start Mastra Studio (dev UI at localhost:4111)
npm run dev

# Run the full research → write → schedule pipeline
npm run test:agents

# Run individual agents
npm run test:researcher
npm run test:writer
npm run test:scheduler

The server runs on port 3010. Mastra Studio is available at localhost:4111 for visualizing agent traces and runs.

Agents

Researcher

Searches for AI content signals across the web and X. Focuses on frontier model releases, agentic frameworks, inference infrastructure, applied AI use cases, and emerging developer tools.

Writer

Converts research into X posts with a specific voice: practical, technical, first-person, contrarian when warranted. Threads are 3–6 tweets, prose-based (no bullet points), max 1 hashtag, no emojis.

Scheduler

Assigns posting times targeting developer-heavy windows: 8–10 AM, 12–2 PM, 5–7 PM, and 9–11 PM EST. Puts the most broadly appealing content in prime slots, niche/technical posts in off-peak times.