Sources

AgentSwarms — Our Biggest Release Yet: 100+ Executable Framework Notebooks, Voice Agents, Website Embeds & More TL;DR — what's new: -103 executable, in-browser notebooks across 9 agent frameworks — no setup, run every cell live. -Voice agents you build with speech-to-text + text-to-speech and talk to right in the browser. -Website embeds — drop a standalone agent or a multi-agent task bot onto your own site with one iframe. -30 interactive field-engineering blogs + 25 interactive presentation decks -9 hands-on failure-mode labs + a 38-cell failure-modes notebook covering 12 real production failures -17 free, no-login interactive AI tools -Detailed swarm execution traces — tokens, cost, latency, every handoff -20 runnable industry templates AgentSwarms is a learn-by-building platform for Agentic AI: build, run, test, and ship AI agents and multi-agent swarms — and actually learn the frameworks by running real code, not reading about it. This is our largest update to date. Here's everything that's new. 1. Framework-specific notebooks with executable cells — learn every major agent framework by running it We shipped 103 notebooks with live, editable cells that execute right in your browser — zero installs, zero API keys to wrangle. There are dedicated, in-depth tracks for OpenAI Agents SDK, Google ADK, LlamaIndex, Mastra, LangChain, LangGraph.js, and AWS Strands SDK — plus Vercel AI SDK and VoltAgent. Each track walks the whole arc: fundamentals → typed tools → multi-agent handoffs → memory → guardrails → evaluation, with clear, beginner-friendly explanations next to code you can change and re-run. Start here: https://agentswarms.fyi/notebooks 2. Build voice agents with TTS + STT — give any agent a voice and talk to it You can now turn any agent into a voice agent using a speech-to-text → LLM → text-to-speech pipeline, pick from 11 studio voices, and talk to it live in the Voice Playground — tap the mic, speak, and hear it answer back. Fork a ready-made sample to get started in seconds. Because voice runs on the same engine as chat, your tools, RAG, memory, and guardrails all carry straight over. Try it: https://agentswarms.fyi/voice-playground 3. Build an AI backend and embed agents + multi-agent task bots on your website Ship what you build. Embeds turns any agent or swarm into a widget you paste into your own site with a single iframe — no backend to stand up. Choose an Agent embed (a streaming chat panel that answers with the agent's full brain) or a Swarm embed (a run-and-return task bot where a whole multi-agent workflow executes server-side and returns the result). Every embed gets domain-locked API keys, a per-embed daily request cap, and serves anonymous visitors with no login. Details: https://agentswarms.fyi/docs/embeds 4. New interactive field-engineering blogs and presentations Our library now includes 30 field-engineering blog posts, each with custom interactive visuals you can click through — covering AI agents in financial services, multi-agent simulation, deploying MCP servers to AWS/Azure/GCP, voice agents, security, retrieval, cost control, and framework deep-dives. Alongside them are 25 interactive presentation decks on everything from LLM internals to multi-agent orchestration and LLMOps. Read them: https://agentswarms.fyi/blog 5. A lot more failure-mode labs and notebooks Real agents fail in ways demos never show, so we built the practice ground for it: 9 hands-on failure-mode labs in the swarm canvas that inject a real fault — a coordination deadlock, a cascading error, a runaway loop, a prompt injection — and then grade your fix. They're backed by a 38-cell "Failure Modes Lab" notebook that reproduces 12 production failure modes (hallucinations, sycophancy, RAG poisoning, context rot, cost blow-ups, goal drift, and more) and ships the exact fix for each. Start at https://agentswarms.fyi/learn 6. Free interactive AI tools We added 17 free, no-login browser tools for everyday agent work: a RAG chunking visualizer, prompt-injection tester, semantic chunker, swarm-architecture visualizer, PII redactor, multi-agent token-cost calculator, LLM-as-a-judge builder, tool-schema generator, GraphRAG triplet extractor, synthetic RAG-eval dataset generator, and more. Open and use them instantly: https://agentswarms.fyi/tools 7. Detailed swarm execution traces Every multi-agent run is now fully traced and replayable. Step through each node's inputs and outputs, every tool call and handoff, and read per-run token usage, cost, and latency on a timeline — the raw material for debugging, evals, and cost dashboards. No more guessing why a swarm did what it did. 8. More runnable industry templates We expanded the template library to 20 one-click, runnable industry templates — support triage / knowledge-base agent, research RAG, sales outreach, code reviewer, CRM data cleanser, long-context analysis, and more. Each spins up a working agent or swarm you can run immediately and adapt. Browse them: https://agentswarms.fyi/templates

Search

Featured matches

Other tools

Go to section

Search

Sources

Featured matches

Other tools

Help

People also viewed

Feedback and Incident Report

AI Options

Create AI Tools

Mini Tool

Vibe code an AI Tool