M-Drive, the self-hosted AI agent platform.
Run a whole team of AI agents on your own server. Each agent has its own instructions, memory and tools, works with the model you choose, and replies on the web, WhatsApp and email.
Everything a team of agents needs
One platform to build, run and govern AI agents, from a single assistant to a multi-agent system.
-
Multi-agent teams
Each agent has its own role, instructions, memory and tools. Agents can pass work to each other, like a coordinator handing billing questions to a billing specialist.
-
Your choice of model
Gemini, Claude, OpenAI models, or open models running locally. Set a main and a fallback model per agent, and switch later without rebuilding the agent.
-
Knowledge and memory
A knowledge base built from your documents, with the source shown for each answer, and a memory that keeps decisions and preferences across conversations.
-
Routines and rules
Agents run tasks on a timetable in your time zone, such as a 7:00 am summary or a check every 15 minutes, and rules turn common requests into fixed actions.
-
Tools and integrations
Built-in tools for web search, documents and images, custom tools written for each agent, and MCP servers to connect other systems or expose your agents to them.
-
Runs on your server
Installs on Linux or macOS. Conversations, documents, memory and API keys stay on your machine, and upgrades keep your agents and their history.
Where your team already works
People talk to M-Drive agents in the browser, in the Android app, on WhatsApp, by email and by voice. The same agent keeps the same memory on every channel.
- WhatsApp groups and chats, including voice notes
- Email: reads and replies, in Gmail or other email accounts
- Web chat, the Android app and voice conversations
Example: the same operations assistant on WhatsApp (left) and email (right)
You stay in control
See what every agent did and why, decide who can use and change each agent, and keep costs within limits.
- Roles and access for your team
- Traces of each run: which tools an agent used and what it answered
- Quality checks to review answers and spot weak replies
- Usage tracking and cost limits
- Backup and restore: move an agent with its memory and history
From install to first agent
We set it up with you, or your team runs it on its own.
Install
One installer on your Linux or macOS server. You sign in from any browser.
Add models
Connect Gemini, Claude, OpenAI or a local model with your own keys.
Create agents
Start from a template or from scratch, then give each agent its instructions, tools and knowledge.
Connect channels
Link WhatsApp groups and email, set routines, and invite your team.
M-Drive questions
Anything else, just ask.
Is M-Drive cloud software or self-hosted?
Self-hosted. It runs on your own Linux or macOS server, so your conversations, documents, memory and API keys stay with you.
Which AI models can M-Drive use?
Gemini, Claude, OpenAI models and open models run locally. Each agent can have a main and a fallback model, and you can change them at any time.
Can it run without sending data to an AI provider?
Yes, with a local open model, nothing leaves your server. With a hosted model such as Gemini or Claude, the text of each request goes to that provider.
Can agents work together?
Yes. Agents can send work to each other, so a coordinator can hand a question to a specialist and combine the answers into one reply.
Can agents reply on WhatsApp and email?
Yes. Agents can post to and reply in WhatsApp groups and chats, and read and reply to email, in Gmail or other email accounts.
How can we try it?
Email us for a demo. We show it working on a task from your own business, then help you set it up on your server.
See M-Drive on your own task
Tell us what your team spends time on. We will show you an agent doing it.
Ask for a demoOr write to mathan@km-ts.com