Building an AI Social Media Agent

Generic AI writing tools produce copy in nobody’s voice, so every draft needs heavy editing before it can go out. A social media agent closes that gap by learning your voice first: it scrapes your best-performing historical posts, analyzes the style, stores that profile permanently, and drafts and publishes in it — so the output stays on-brand without a rewrite each time.

The stack

Four tools, one per phase of the workflow.

The workflow

1. Ingest. The user supplies a target handle. ScrapeGraph navigates to the profile and extracts the text of the most popular posts — around the top ten viral posts is usually enough of a baseline to replicate a style reliably.

2. Analyze and store. Nebius AI examines the posts across tone, formatting, vocabulary, and cadence, producing a voice profile. Memori persists that profile so the agent carries the same voice into every future session.

3. Generate and post. When the operator gives a new topic, the agent pulls the stored profile from Memori, Nebius AI drafts a post matching that voice, and — once approved — Composio publishes it live via API.

Implementation

Keep API keys in a .env file; never hardcode them.

# Nebius AI Configuration
NEBIUS_API_KEY=your_nebius_api_key

# ScrapeGraph Configuration
SGAI_API_KEY=your_scrapegraph_api_key

# Composio Twitter Integration
COMPOSIO_API_KEY=your_composio_api_key
TWITTER_AUTH_CONFIG_ID=your_twitter_auth_config_id
USER_ID=your_unique_user_identifier

Install dependencies via requirements.txt or a modern manager like uv:

• streamlit — user interface
• composio — API execution
• langchain-scrapegraph — data ingestion
• memorisdk — persistent memory
• langchain-nebius — LLM orchestration

Split the logic into three modules:

1. app.py — the Streamlit dashboard the operator drives.
2. twitter_agents.py — core logic: ScrapeGraph ingestion, Nebius analysis, Memori storage.
3. create_tweet.py — the execution script that connects Composio to the account and posts.

Why it matters

Delegating the drafting and posting turns the operator’s job from tactical to strategic — you set topics and approve, the agent handles the rest. And because the voice profile lives in Memori, the same profile and Composio integration can be pointed at other text-heavy channels — LinkedIn, Instagram — once the core pipeline is stable, without re-learning the voice from scratch.