Hey Product Hunt community! 👋 I'm Tejas, the founder of Winsen Labs and the creator of Platos! I've spent the last year trying to take AI agents to production. Tried every framework. Tried hosted platforms. Tried writing the runtime ourselves. Failed, more times than I'd like to admit. The library tier ships you parts. You spend months stitching them yourself: memory, observability, scope, auth, rate limits, prompt caching, the streaming loop. Months you don't have. The hosted tier ships you the runtime. You give up your data, your model choice, and your cost line. Vendor lock by another name. The breakthrough was finding trigger.dev. Background tasks that survive restarts, deploys, and actual production load. Once we had that under the agent loop, the rest of the stack started to click. An agent runtime sitting on top of a durable task engine. That is the formidable force nobody had shipped together. So we built Platos! git clone, docker compose up, and you go from prompt to live agent in minutes. Full monitoring, prompt caching, rate limit caps, persistent memory across sessions, MCP gateway, agent clusters, custom skills, all of it. Wrap any function in any backend with one decorator and it becomes a typed tool the agent can call. BYOK any LLM. Built on Vercel AI SDK, the most widely used agent layer out there. Built on trigger.dev so your agent can kick off long-running operations that survive crashes and deploys. Apache 2.0. Self-host the whole thing on your own infrastructure. If you have been trying to ship agents to production and felt like every framework is one piece short, this is the one. https://play.platos.dev runs the live demo. The chat bubble in the corner of https://platos.dev is our React widget pointed at the same runtime you would self-host. Would love your feedback! Tejas

Platos | The runtime for Managed Agents

Your agents. Your servers. Your keys. Your model.

Your agents. Your servers. Your keys. Your model.