A modular switchboard for social-media agent swarms

What it does Agentfy is a Python orchestrator that takes a user prompt, reasons over it with an LLM, and dispatches work to modular sub-agents that crawl, analyze, or interact with social platforms like TikTok, Instagram, and X. The system uses a JSON agent registry to discover capabilities and dynamically chain tasks—finding buyers, generating content, sending DMs—without hard-coding every workflow. It exposes a CLI and a Streamlit interface, but at heart it is a routing layer that leans on external platform APIs to do the actual work.

The interesting bit Rather than treating itself as just another automation script, Agentfy frames its architecture as a prototype “open agent protocol” inspired by Google’s A2A and Anthropic’s MCP. The pitch is an HTTP-like interoperability layer where a “TikTok Agent” can call a “CRM Agent,” though the current codebase is closer to a well-structured registry of platform glue code.

Key highlights

• Registry-based modularity: new agents are added by dropping code into agents/ and updating agent_registry.json, with no core logic changes.
• LLM-driven intent routing: the system parses natural language and selects the right agent functions dynamically.
• Covers the full social ops stack: buyer discovery, cross-platform posting, automated messaging, creator outreach, and video performance tracking.
• Sponsored by TikHub, which supplies the underlying data APIs for TikTok, Instagram, YouTube, and others.
• Early groundwork for exposing agents as callable microservices via CLI, Streamlit, or FastAPI.

Caveats

• You must supply your own API keys for every platform you want to touch; the README warns this can be a hurdle and offers a support email for help.
• Several integrations are aspirational: the README states it “will integrating deeply” with platforms, suggesting the current surface area may be narrower than the feature table implies.
• Documentation has rough edges—broken sentences, missing words, and placeholder instructions—that signal a very early-stage project.

Verdict Worth a look if you are building multi-agent social media tooling and want a reference architecture for LLM-based task routing. Skip it if you need a polished, batteries-included automation suite that works out of the box.