Thanks to everyone who joined us live for Enterprise AI that Delivers with David Hsu! 
We had some fantastic questions come through during the session, so we pulled together a recap of the Q&A in case you missed it—or want to keep the conversation going.
BTW: Kent (who led the Agents launch) and @matei (already building with them) are chatting live TODAY at 10AM PT for Builder Talks. No pitch, just a real convo about using Agents in the wild. Join us here →
Check out the highlights below and feel free to drop any follow-ups or ideas in the thread—we’d love to hear how you’re thinking about building with Agents.
Product & Platform Vision
Q: Is Retool changing its business model and product offering?
A: The Retool offerings you're familiar with are not going away! We see Agents as a new opportunity to build on all of the Retool features you're familiar with and make them work in the age of AI. You can call agents from your existing apps, use workflows as tools, hook them up to RetoolDB to read and write data and more, which makes them even more powerful because they sit on top of our platform.
Q: So now Retool covers UI, identity, backend, and agents?
A: More updates to come here! We're focused on continuing to make Retool the platform that helps bring good software to the world and the how of that is always changing!
Q: Why not simply embed the Agents module as part of the Workflow modules? Was there a reason behind separating the two?
A: While tools like n8n have an agent block that can run as part of their workflows, we view agents in Retool as a whole new primitive. By making them more than just a workflow block, this means you can trigger your agents via apps, email, workflows (yes there still is an agent block in workflows) as well as the dedicated agent chat interface. We believe this makes them even more powerful than an agent that can only run in the context of a single workflow.
Technical Setup & Capabilities
Q: How does the gettranscript tool know which meeting to get the transcript notes from?
A: Since it's a custom tool, it all depends on how you set it up. In the demo we showed, it was looking for the most recent meeting.
Q: How do I trigger with a webhook from the configuration page?
A: That isn't set up from the configuration page. You would create a Retool workflow to be triggered via webhook. Then, your agent can be called from a block within that workflow.
Q: Can agents be tested with a mock data set OR a real company source but not push any actions or updates until released to production?
A: The best solution to this currently is to create a mock version of your tools, where the tools themselves return mock data. This allows you to test a given model's tool calling ability and see a run of your agent end to end, but with no risk to your production data. If you would prefer to work in production, taking advantage of the "tool approval" functionality means that you can limit the tools an agent can call and manually inspect them before your agent takes any action.
Q: Will I be able to review bot action history? This UI looks good, hopefully there is a powerful search behind it too to dig into specific complaints.
A: We're working on making both the log and monitoring views even more powerful. Users will have full control and transparency over both what their agents are working on, and what they've done in the past.
Pricing & Deployment
Q: Will Agents be available on self-hosted?
A: Yes, Agents will be available for self-hosted customers as a preview release in June and then as part of the stable release in August.
Q: Is the pricing model based on the number of hours even for enterprise self-deployed plan? What if we use the agent with an internal LLM proxy?
A: The pricing model for Retool Agents is always based on the number of hours your agent is running. If you are using a self-managed LLM, the price is a flat $5 per hour of agent run time.
Q: Is Retool AI going to be unusably expensive vs. something like Lambda?
A: Pricing for Retool Agents depends on which model your agents are using. We only charge you for the time your agents are running (thinking, calling tools, creating output) and we feel that this creates a pretty compelling case for using agents in Retool to solve problems, especially compared to other ways you could accomplish the same task.
LLMs, Hallucinations & Safety
Q: How do we reduce hallucinations and ensure trust in agent decisions?
A: This is where a combination of evals and monitoring come in handy. The likelihood of hallucinations depends on a combination of the model and the prompt you're using, but using logging to catch them after the fact (and correct the prompt) or evals to enforce known "good" input/output combinations work well.
The key tradeoff with agents is to scope the decisions that the LLM is making by writing a clear prompt. You want your agent to be choosing the tools, but then you want the complexity of the work to live in the tools themselves. This allows you to make these tools deterministic and reduce the risk from hallucinations because you're more easily able to inspect and verify the function of the tools your agent is choosing.
Q: Can AI tools like ChatGPT access Retool info if we point them to docs?
A: Yes—paste relevant documentation into ChatGPT or Perplexity. You could build an MCP server on top of Retool’s API, but we’re not exploring that internally.
Q: If we want guided help on ‘how to’ develop using what we just saw—if working in Replit as well as in Retool—is the information available to be ‘seen’ by ChatGPT or Perplexity or Gemini if we tell it where to go?
A: You've got a couple options here. The most straightforward is to paste the relevant page(s) of documentation into a tool like ChatGPT or Perplexity.
Are you looking for a tool to give an LLM control of managing Retool? If so, it would be possible to build an MCP server on top of the Retool API, but that's not something we're exploring internally.
Integrations & Ecosystem
Q: When will Retool integrate with Microsoft Office?
A: We’ve seen exciting developments from Microsoft around MCP and other integrations we’re keeping an eye on.
Q: Can Retool integrate with Microsoft Co-pilot?
A: There's a few different ways to interpret this question. Can you reach out to keanan@retool.com so we can chat more specifically about what you're looking to do with Microsoft's Co-pilot?
Q: Can I replace Salesforce or MS agents with Retool agents?
A: Since you have full control over how you configure your agents and what tools you build in Retool, this is definitely something that's possible depending on what your other agents are currently tackling. In fact, it's one of the benefits on a platform that's both data and model agnostic. Salesforce is a very popular Retool integration and we already have at least a couple agents running internally that use Salesforce data. This also means that, if you need to bring in other data or make additional modifications, you're not locked into a single ecosystem.
Developer Experience
Q: Why use Retool over n8n + Langchain + tools + DBs?
A: A significant value of having a platform like Retool to begin with is not needing to cobble together a variety of different pieces that don't always fit. This has never been more of an issue than with agents, as in order to be useful, they need deep and specific tools inside your company's stack. That is what builders have been creating with Retool for years, and now they're by default available to an agent. Bringing all the pieces together in a seamless and trusted surface means that orgs can focus on their true value instead of drudgery.
Q: I am hoping there will be something like cursor / vscode. I am doing a lot of cutting and pasting between Retool web interface and chat.
A: Retool's AI-assisted development experience is coming soon! Take a look at this video for more information or join the waitlist.
Tagging a few folks who asked great questions—feel free to jump back in if you’ve got follow-ups or want to share how you’re thinking about using Agents:
@taylorsearle @richrono @ageorgia @Shawn_Optipath @tom-blazek @matt.em