Forum Discussion

harshankbansalalgoworks's avatar
harshankbansalalgoworks
Copper Contributor
Sep 27, 2026

Clarification on Copilot Conversation Length Limits and Copilot Studio Agents‎

TLDR; We need clarification on the undocumented Copilot conversation-length limit: how it is determined, whether the M365 Copilot add-on changes it, and whether the same limit applies to Copilot Stu...
copilot chat
microsoft copilot
Paras Dodhia's avatar
Paras Dodhia
Iron Contributor
Nov 10, 2026

Hi Harshank,

I work on Copilot Studio architecture for clients in regulated industries, and this question comes up a lot when we're scoping agents.

Short answer: there's no published number. I've looked for one too. A Microsoft moderator confirmed on your Q&A thread that Microsoft doesn't document a fixed conversation limit for M365 Copilot Chat, and that it depends on the service and the context. The "30 turns" you'll see everywhere is from the old Bing Chat days, so I wouldn't plan around it.

On the add-on license, I wouldn't count on it lifting the limit. The license changes what Copilot can ground on. It doesn't change how long one chat can run.

The Copilot Studio part is the one that I think matters the most for your design, and it depends on what kind of agent you build.

A declarative agent runs inside M365 Copilot's own chat and orchestrator. I'd assume it hits the same wall as regular Copilot Chat. Microsoft hasn't spelled that out, but that's how it's built. Being billed by consumption doesn't change it.

A custom agent built in Copilot Studio runs on its own runtime, and those limits are documented. On the new harness, the agent keeps the last 100 messages and drops older ones as the conversation grows. The chat doesn't lock. The catch is that it can forget things from early in a long conversation. See the "Agent runtime limits" page on Learn.

For a CRM agent over MCP, I'd lean toward a custom Copilot Studio agent if your users will have long working sessions. You get documented limits you can actually design around, plus proper ALM and governance, which matters once your client's risk team gets involved. A few other things I'd do either way:

  • Write anything important back to the CRM, so a fresh chat doesn't lose anything
  • Keep the MCP responses lean, with filters, paging and only the fields you need. Big payloads use up context fast
  • Pilot it with some realistically long conversations and note where they break. You'll want that evidence if you need to open a ticket or ask Microsoft for a limit increase

Happy to dig into the declarative vs. custom decision if it helps.