Leaving Microsoft Copilot is an inventory problem before it is a tooling problem. Agent definitions, grounding and prompt history live inside Microsoft's tenant, so you rebuild rather than export. This plan gives the order, then the open-source system the work lands in: Kortix.
Microsoft Copilot is Microsoft's hosted AI assistant built into Word, Excel, PowerPoint, Outlook and Teams, grounded in an organization's own content through Microsoft Graph and Work IQ. It runs as a Microsoft 365 service, and the assistant layer stays Microsoft's. (Microsoft Copilot)
Almost nothing you built inside Copilot leaves with you. Agent definitions are provisioned inside your Microsoft 365 environment, and deleting one removes tenant resources such as the app registration and service principals, so there is no portable file to carry out. (Copilot Studio)
Grounding is a tenant connection, so it does not travel as a file you own. Copilot reaches files, mail and chats only through Microsoft Graph and Work IQ, scoped to each person's permissions, and it stores the prompt, the response and the response's citations as that person's Copilot activity history alongside your other Microsoft 365 content. (Copilot data, privacy and security)
The intent behind each artifact is what you re-create: an agent becomes a file in a repo, a prompt becomes a skill, and the grounding becomes a connector your own system holds. The result is open source. Read it, fork it, audit it.
A migration stalls when it starts with the tooling. Start with the inventory instead: the workflows people actually run, the agents behind them, the connectors each one needs, and the team that owns each. Keep the list short enough that one team can move without waiting on another.
If the team is still weighing the decision, the Kortix vs Microsoft Copilot comparison covers the differences, and the guide to open-source Copilot alternatives lists the rest of the field.
kortix self-host init --domain kortix.example.com --admin-email you@example.comkortix self-host startkortix self-host doctorThe published self-host commands from Kortix's docs; init writes docker-compose.yml.
The open-source AI Operating System is Kortix: it holds your agents, their skills, your company memory, and every connector in one git repo you own. One isolated sandbox per session. Each session has its own isolated machine and branch.
Session work reaches main through a change request. Merge is default-deny for agents. Any model provider with your own keys, or the ChatGPT plan you already pay for.
Connectors: 3,000+ apps in a click, plus MCP, OpenAPI, Postman, GraphQL and raw HTTP. Run it on Kortix Cloud, in your VPC, or on your own on-prem network. Self-host is free. (Kortix docs)
Kortix keeps the configuration in a repo you own; Copilot keeps it in Microsoft's tenant.
There is no single cutover. The work runs one workflow at a time, so the length follows your inventory: more workflows and more connectors take longer, while standing up the platform is a few commands.
Yes. Moving the assistant layer off Copilot changes where the agents and memory live while leaving your Microsoft 365 plan and the apps your team opens every day in place. Kortix reaches those tools, and the rest of your stack, through connectors.
They stay in your Microsoft 365 environment and do not export as portable files. You rebuild each one as a file in your own repo and point it at the tools it reached before. (Copilot agent management)
Start on Kortix Cloud or self-host, then bring your agents over one team at a time.