# Structure your context. Share it across your team and agents. > One workspace for the context, skills and terms your team's AI tools work from. Connect Claude, Codex or any AI tool that takes a custom MCP connector. Free during the beta. ## 1. Create your workspace Give your team and agents a shared home for knowledge, decisions and skills. Keep what you learn together, so everyone can find it and build on it. ## 2. Connect your AI tools Add the MCP to the AI tools your team already uses. Share the same context and instructions across tools, so useful knowledge carries from one session to the next. ## 3. Run /setup. Start small. Your agent checks your current setup, then guides. Turn what you already have into structured, actionable context, with sources, timestamps and instructions attached. ## 4. Invite your team Invite teammates with one click. An email link takes each of them through onboarding, and you choose who can see and update which parts of your context. ## 5. Connect GitHub repo Connect your team's private GitHub repository to back up your workspace every night. You own all your content. Midland just handles the structure and maintenance. You’ve explained the company to your AI more than once. Midland keeps that context where every teammate’s agent finds it, and keeps it current as the company changes. ## Built and maintained so you don't have to. 1. **Save time**: Spend your time and tokens on your own work, not on the plumbing underneath that connects your team’s AI. 2. **Catch contradictions**: Contradicting answers show up depending on who asks and which tool they use. Midland flags the conflict, its owner and the timestamps so every connected agent sees it before repeating the answer. 3. **One source**: Useful context stays stuck with the person who created it. One structured layer serves the same current answers, skills and terms to Claude, Codex and any MCP tool, so the rest of the team inherits the setup the power user built. 4. **Tracked decisions**: Yesterday’s decisions keep getting treated as today’s truth. When a new decision is saved, Midland surfaces the older claims it replaces. Update expectations, ownership, scheduled Loops and control questions keep agents off the superseded answers. 5. **Typed items**: The way the team organises knowledge stops working as soon as the work changes. Every item says what it is and how to write to it, so the layer stays coherent as new decisions and use cases appear. 6. **Save rules**: Everything useful produced in a session dies with the chat. Agents follow rules for where plans, reports, transcripts and customer emails belong, how to write them, and which existing items to update or link, so session output becomes reusable shared context. 7. **Assign owners**: No one is clearly responsible for keeping the shared layer honest. Every item has an owner, access rules and a clear update expectation. Agents and the overview show who is responsible and when something is past due. 8. **Self-maintaining**: Keeping the context layer usable eats the time and tokens meant for actual work. Built-in skills let any connected agent set up the layer. Loops and automated writers handle the ongoing checks so people spend tokens on their own work. 9. **Less noise**: A growing library of markdown files makes results worse instead of better. You choose what to save and can archive what is no longer useful. Midland organises the rest into named items with descriptions, so agents can find and read what is relevant to the task. 10. **Full visibility**: You cannot see what an agent actually read, wrote or relied on. Every item carries source, owner and timestamps. Problems surface inside the agent’s work, and the overview shows what is being used. 11. **Any tool**: Each AI tool starts from a different, incomplete picture of the team. Works in the AI tools you already use via MCP. The same setup is managed from one place and available in every connected session. 12. **Your content, backed up**: Managing a shared Git repo means handling access, merge conflicts and maintenance. Midland hosts your context in the cloud, manages access and runs checks for issues. A planned automatic backup will copy workspace-wide content to a GitHub repo you own, without merge conflicts. Your content remains yours. 13. **Versioning**: Past versions vanish and there is no reliable way to go back. Every version is kept, so you and your agents can look back at how a file has changed. If an update removes something you still need or takes the content in the wrong direction, open its history and restore an earlier version. ## Making AI work useful beyond the session. I designed Midland after months of trying out different memory systems. Each one worked at the start, then drifted and needed more and more maintenance. I never trusted any of them enough to roll out to clients or coworkers. If you've had the same experience, Midland is designed for you. I use Midland to take on work with more moving parts than I can keep in my head. It gives my agents a place to find earlier decisions, instructions, and details I might otherwise forget to bring into the next task. Midland is a shared layer that structures your context, shows who owns each part and when it is due an update, and serves it to any AI session. I built it for small teams without data engineers, who would rather spend their time and tokens on agents for their own business than on the plumbing under them. I want every model you use to know your business from its first session, as if it had been trained on it. Now I am opening up the beta. ## Get started ### Hosted Create a workspace and connect your first tool. Free during the beta. [Create workspace](https://console.midland.md/signup) ### Self-hosted Run Midland on your own infrastructure. The code will be open source. Leave your email for early access. [Join the waitlist](https://midland.md/) ### Assisted roll-out We help your team structure its context into one shared layer. [Contact](https://cal.com/fred-abrahamson-smh9i6/30min) ## More - [FAQ](https://midland.md/faq.md): What Midland is, who controls what goes in, how your data is kept, and how to get started. - [Privacy](https://midland.md/privacy.md): What personal data Midland handles, why, for how long, and your rights. - [Terms](https://midland.md/terms.md): The terms for using the hosted Midland service. - [DPA](https://midland.md/dpa.md): How Midland processes personal data in workspace content on a customer's behalf. - [Subprocessors](https://midland.md/sub-processors.md): The suppliers Midland uses to run the hosted service.