Creator OSDocs

Social Agents, the Creator OS agent

Social Agents is Creator OS’s own agent harness: a ready-to-run agent with the CLI built into its codebase. Instead of wiring the toolkit into your own AI coder, you run Social Agents and get an agent that already knows it. It onboards you with a few setup questions, writes a brand file it re-reads before every run, and manages your accounts end to end.

Social Agents ships as a fork-and-run starter repo: github.com/kevinbadi/social-agents (open source, MIT). Fork or clone it, npm install(bootstraps the CLI and agent surfaces), set your Anthropic API key (the agent's brain), and start. Your Creator OS workspace key is collected during the onboarding interview, masked as you type:

$ git clone https://github.com/kevinbadi/social-agents && cd social-agentsCopy
$ npm installCopy
$ export ANTHROPIC_API_KEY=...Copy
$ npm start creatoros social-agentsCopy

Onboarding: Social Agents builds your brand file

On first run Social Agents interviews you and saves the answers to a brand file (brand.md) that anchors everything it writes and replies to:

Your brand

What you’re marketing and what you’re selling: the products, offers, and positioning Social Agents writes toward in every post and reply.

Your links

URLs for your products and services, so posts, replies, and DMs always send people to the right place.

Your profiles

The username for each connected profile, so the agent addresses the right account on every platform.

Automations

Whether you want automations set up. Say yes and Social Agents preconfigures them for you during onboarding.

What Social Agents can do

This is the exact Creator OS API surface Social Agents ships with. Every other endpoint is gated off from the agent, so it can’t touch anything outside this list.

Posting

Every post type, to one account or many at once.

  • •Short-form video: TikTok, Reels, Shorts
  • •Long-form video
  • •Carousels
  • •Single posts: image or blog-style text
  • •Threads & tweets
  • •Multi-posting: one call publishes to several platforms
  • •Scheduling: server-side, nothing stays running on your machine

Analytics

The same numbers the dashboard shows, readable by the agent.

  • •Platform analytics: follower growth, views, and per-platform performance
  • •Post analytics: per-post metrics across every connected platform

Messaging & comments

Webhook-driven: Social Agents reacts as events arrive, no polling.

  • •Message (DM) replies: X, Instagram, Facebook
  • •Comment replies on every platform except TikTok: X, Instagram, Facebook, Threads, YouTube, LinkedIn
  • •Comment-to-DM workflows: Facebook & Instagram

Launch the Social Agents dashboard

The Social Agents repo ships with a local web dashboard for watching what your agent is actually doing, and verifying it’s working. From the social-agents folder:

$ npm run dashboardCopy

It opens at http://localhost:4180 (override with SOCIAL_AGENTS_DASHBOARD_PORT). After npm link you can also launch it from anywhere with kai dashboard. Zero external services: it reads the repo’s files, the agent’s structured activity log, and the Creator OS API with your already-configured credentials. Missing credentials never crash it; you get a friendly connect state instead.

Overview

Health strip, reply/DM/post counters, a GitHub-style year heatmap of agent activity, and a live feed

Agent

Full transparency into the agent’s understanding: persona, objective, KPIs, comment/DM rules, and the literal system prompt it runs on

Automations

Every workflow drawn as trigger → action → outcome node chains, each with an operating/armed/failing health badge and live executions

Brand

Your brand file rendered, editable in place

Training

Every workflow playbook with last-used-by-the-agent info, editable in place

Logs

The full filterable activity feed with raw JSON and real error payloads

Chat

The same Kai as the terminal, streaming in the browser

Every panel is fed by plain local JSON endpoints (/api/health, /api/activity, /api/automations, /api/chat…), and a panel is just one ES-module file in dashboard/public/panels/ with no build step, so you can add your own views against the same data.

PreviousAI agents