Run Claude Managed Agents with Vercel’s Chat SDK: What’s New

Vercel has added Claude Managed Agents support to its Chat SDK. The integration lets developers ship AI agents, like a Slack research bot, without managing their own database or session state. The agent loop, including the model, tools, conversation history, and sandboxed web research, runs server-side inside Claude Managed Agents. Vercel provides a one-click deploy template for a Slack research agent as a starting point.

What happened

Feature	Detail
Integration	Claude Managed Agents now works with Vercel Chat SDK
Session storage	Conversation state stored inside Managed Agents, no developer-managed database needed
Streaming	Token-by-token output over a single streamed response
Platform reach	30+ platforms supported by swapping a few lines in the handler
Launch template	One-click Slack research agent deployable on Vercel

Vercel’s Chat SDK now connects directly to Claude Managed Agents, Anthropic’s server-side agent runtime that handles the model, tools, session state, and sandboxed web research in one managed layer. Previously, developers building stateful AI agents needed to wire up their own database to store conversation history and manage the agent loop themselves.

With this integration, none of that falls on the developer. The Managed Agents session holds the full transcript, so features like sidebar history, replays, and conversation continuity come for free, reading directly from the session rather than a developer-owned store.

What does Claude Managed Agents actually do?

Claude Managed Agents is Anthropic’s hosted agent runtime. It runs the loop that would otherwise sit in your own backend: calling the model, invoking tools, handling retries, and maintaining session context between turns. Web research happens inside a sandbox, so the agent can browse and summarize sources without any extra infrastructure on your side.

The Chat SDK wraps that in a developer-friendly interface. Responses arrive token by token over a single HTTP stream, and live tool calls (what the agent is doing mid-turn) are surfaced as an activity feed you can show directly in the chat UI. Users see a trace of the agent’s work as it happens, not just a final answer.

Why it matters for builders

The biggest practical win here is portability. According to Vercel, changing a few lines in the request handler moves the same agent from Slack to Microsoft Teams, Google Chat, Discord, WhatsApp, or any of 30+ supported platforms. That means a research bot built for one internal tool can expand across your communication stack without a rewrite.

For teams building AI integrations on top of existing chat platforms, removing the session-management layer cuts a significant chunk of backend work. You still own the front end and the deployment, but the stateful parts live in the managed layer.

This also fits a broader pattern worth watching: AI providers are moving more of the agent infrastructure server-side, reducing the surface area that product teams need to maintain. That is good for shipping speed, but it also means more dependency on the provider’s runtime behavior and pricing.

Our take

This is a genuinely useful reduction in boilerplate for teams already on Vercel and already using Claude. The no-database pitch is real: session state is one of the fiddliest parts of building agents that need to remember things across turns. Offloading that to the managed layer is a fair trade for most projects.

The portability story is compelling on paper, but it depends on how much platform-specific logic your bot accumulates over time. A pure research bot with simple inputs and outputs travels well. A bot that leans on Slack-specific features, like thread structures or emoji reactions, will need more than a few line changes to move platforms.

If you are evaluating this for production, start with the one-click Slack template, stress-test the streaming behavior under load, and check Anthropic’s pricing for Managed Agents sessions before you commit. The convenience is real; so is the vendor lock-in. For teams exploring workflow automation across multiple chat platforms, this setup is worth a serious look.

What to do about it

1. Deploy the one-click Slack research agent template on Vercel to see the full stack working end to end.
2. Review Anthropic’s Managed Agents session pricing before scaling beyond a prototype.
3. Test the live activity feed in your UI so users can see what the agent is doing mid-turn, it builds trust and reduces perceived latency.
4. Map out which platform-specific features you rely on before committing to the portability approach across 30+ targets.

The fastest way to evaluate whether this fits your workflow is to clone the template, deploy it, and send it a real research question your team already asks manually.

Source: Vercel Blog