What's inside
Everything you need to run a team of AI coders in one place.
Notes, a graph, a calendar, isolated workspaces, live terminals, and your choice of AI - all on your own machine, all working from the same files.
A manager that splits the work
Start a message with `task:` and the manager takes it, writes a delegation plan, and runs each member on its own slice - then grades what comes back and synthesises it. Anything else you type goes to the semi-manager, a read-only session that answers questions about the team and its runs. You never choose which one to talk to; the prefix decides.
Local models are first class
Point ReplicaMD at Ollama and a local model becomes an ordinary member: same panes, same memory, same manager. It needs no sign-in, so an entire team can be staffed from your own hardware and keep working with the network off.
Plugins, from the app or the terminal
`replicamd plugin install <slug>` works in any pane, and what it installs appears in Settings immediately because both go through the same store. The catalogue includes image and video tools that shell out to ImageMagick and ffmpeg for real work, not just prompts.
Notes that stay yours
Knowledge is a proper writing space built into the app: live preview, links, search. Every note is an ordinary file on your disk, so nothing is trapped in someone else's format.
See how it all connects
As you link notes, the graph turns them into a picture you can navigate - search, focus on one corner, and spot the notes that have drifted off on their own.
A calendar for the work
The calendar reads the same notes as the editor and the graph. Plan locally today; it's built so secure sync can slot in later without changing how your files work.
Agents that stick around
Instead of a chat that disappears, an agent is a lasting teammate. It can pause, resume, and hand work back to you without hiding what changed in your repo.
Everyone gets their own copy
Under the hood these are Git worktrees - a normal, well-understood Git feature, not a black box. Agents can try different fixes in parallel and leave clean diffs for you to review.
Real terminals, not transcripts
Every session is an actual terminal. Agents can run servers, tests, and tools, and a background daemon keeps them alive even when the window is closed.
Bring your own AI
Pick the tool that fits each task, budget, or privacy need. One agent can run Claude while another runs Codex, all in the same project, with no lock-in.
It all lives on your machine
ReplicaMD starts from your own files and folders. Anything remote would be added later behind clear, explicit boundaries - never silently.
Make it durable