AgentSwarms

AgentSwarms is an interactive school for Agentic AI that pairs every concept with a runnable, forkable example in a live browser-based playground. It covers the full stack from prompts and RAG to multi-agent swarms, guardrails, observability, and text-to-SQL agents — all in one guided environment with zero setup required to start. The platform is free forever for learners, with a BYOK (Bring Your Own Key) mode for power users who want unlimited requests via their own provider accounts. AgentSwarms positions itself as the layer that gets you ready to use production tools like Langflow, Dify, or LangGraph by teaching the underlying concepts first.

• Guided Curriculum — Six structured lesson modules covering prompts, RAG, tools, guardrails, multi-agent swarms, and observability, each with interactive checkpoints.
• Live Playground — Run real agents directly in the browser; switch models mid-conversation, inspect token counts, and view full execution traces.
• Multi-Provider Gateway — Supports OpenAI, Anthropic, Gemini, Bedrock, Azure, OCI, Qwen, and Grok; start with the free AgentSwarms AI gateway or plug in your own API keys.
• Swarm Canvas — Drag-and-drop canvas for composing multi-agent workflows with typed handoffs between Router, Worker, and Reviewer nodes.
• RAG & Knowledge Bases — Upload PDFs or URLs to create knowledge bases; agents retrieve and cite relevant chunks at query time.
• HITL Approvals Inbox — Gate risky agent actions (refunds, deletes, sends) behind human-in-the-loop approval flows with full audit trails.
• Traces & Observability — Every run logs full prompts, responses, tool calls, token counts, latency, and cost for debugging and eval-driven development.
• Prompt Library — Save, tag, version, and reuse system prompts across agents and swarms as first-class assets.
• Community Feed — Publish agents and swarms publicly; remix others' work to learn faster.
• Budget Caps — Set monthly, per-agent, and daily spend limits with auto-disable on limit to prevent runaway loops.
• MCP Server Support — Connect Model Context Protocol servers to expose standardized tools usable across any compatible agent client.
• Text-to-SQL Agents — Run SQL agents on included sample datasets with AST validation and table allow-listing for safe, grounded data queries.