Persisted Agent Chats: The Feature That Sounds Small and Changes Everything

What may look like a table-stakes feature from the outside is actually quite a major unlock for the future, and immediate boost in performance and quality today.

Much More Than Saved Chats

On the surface, workspace chats with Dreambase agents are now persisted to the database. That sounds like basic chat history. It is not.

Every persisted chat carries the full context management, retrieval, compression, agent framework, and tool access that already powers our dashboard chat experience, which we also upgraded as part of this release. Your conversations with Dreambase agents are not just a log you can scroll back through. They are a living, queryable memory layer that the agent actively uses to reason better on every future turn.

This matters because analysis is not a single question and answer. It is a process. You ask something, the agent answers, you follow up, you refine, you come back the next day with a related question. Before this release, that context disappeared the moment you closed the tab. Now it does not.

Why This Changes the Quality of Everything Downstream

Here is the part that actually matters for your work.

When you are analyzing your Supabase Postgres data alongside your connected APIs and MCP sources, historical chat context means the agent already knows what you have asked before, what you were trying to figure out, and what it already told you. You get better inline analytics because the agent is reasoning with more signal, not starting from zero every time.

That improved reasoning carries directly into dashboard planning. Plan Mode conversations are chats. When those chats persist with full context, the agent asks sharper questions and proposes better plans because it understands the history of what you care about and how you have described your business before.

And better planning means better dashboard generation. Fewer rounds of back-and-forth edits. Fewer dashboards that need to be rebuilt because the first attempt missed the mark. The entire chain, from conversation to plan to generated dashboard, gets more accurate because the foundation underneath all of it now has memory.

This is the kind of upgrade that does not show up as a single headline feature in a changelog. It shows up in every dashboard you build being a little better than the last one, and every conversation being a little smarter than it would have been a month ago.

What This Sets Up Next (MCP)

Persisted, context-rich workspace chats are also the proving ground for something bigger we are building: MCP and REST API access for external consumption.

We are rolling out a Dreambase MCP server that will let you set up, configure, use, and integrate Dreambase agents, dashboards, and datasets directly inside the AI apps and workflows you already use every day. Your own agents, schedules, routines, and automation loops will be able to call Dreambase the same way Dreambase's own agents call PostHog or Linear today.

The persistence and context architecture we shipped in this release is the same architecture that makes that possible. Before we could expose Dreambase agents to the outside world reliably, we needed the internal chat experience to be durable, context-aware, and self-improving. It is now. The MCP server is next.

Open Dreambase →