VESPR documentation
VESPR is a smart development environment (SDE): a workspace where your team and your AI agents write code in the same files at the same time, with the whole team in charge.
On this page
What an SDE is, in plain words
A code editor is built for one person at a keyboard. When an AI assistant is added to it, the assistant usually works for that one person, in a private chat, and its changes show up as a pile of edits to review afterward.
An SDE keeps the familiar workspace (a real editor, files, a terminal, git) and changes who is in it. People and agents are in the same room, looking at the same files, and the team decides what the agents may do. In VESPR that comes down to five things:
- Everyone has a cursor. People and agents appear live in the same file, with names, the way collaborators do in a shared document.
- Agents are teammates. Every chat with an agent is visible to the whole workspace, and an agent asks the room before it changes or runs anything. The first answer on an Approve or Deny card decides, and everyone sees who gave it.
- Every line has an author. Hover a line to see who wrote it (a person or an agent), which model, and the request behind it.
- You can run it and rewind it. Ask the agent to start the project and it posts a preview link. Scrub back through a finished session: edits, cursors, chat and approvals.
- The team stays in charge. Five permission modes, set for the whole workspace, decide what an agent may do alone and what it must ask about.
Put shortly: an SDE is what a development environment becomes when agents are teammates and not tools.
What is under it
- Real VS Code in the browser (Code - OSS) with the VESPR extension, so your extensions, themes and settings work. Built
- A realtime server that keeps one shared document per workspace and merges everyone's edits without locks. Built
- Agents. A worker that joins a workspace as an ordinary participant and calls NVIDIA Nemotron models through Nebius Token Factory. Without a key it runs a scripted demo mode and says so in its messages. Built
- An app shell for the whole team. Agents on the left, the editor in the centre, the conversation beside it, and a History of what was done. It is a prototype in plain HTML and JavaScript, and a few panels are not connected to a backend yet. Partial
Map of these docs
- Getting startedRun VESPR on your own computer. A throwaway demo stack needs no keys and no Docker; the full local stack uses Docker.
- Tour of the appThe top bar, agents, approvals and modes, authorship on hover, History, Code Map, replay, Run for Python, skills, and opening a GitHub repository.
- ConceptsRooms and the shared document, people and agents, modes, approvals, runners, skills, claims and provenance.
- Client APIThe small library the app uses to join a room. Use it to build your own front end on the same shared document.
- FAQWhat is real today, what is planned, and the limits you should know about before you trust it with anything.
How to read the status tags
Every statement in these docs is meant to be true today. Where it is not, it is marked.
- Built exists and runs today.
- Partial works in part. The text next to the tag says which part.
- Planned is not built. It is on the plan, and nothing here should be read as available.
Some command IDs, environment variables and folders still carry the project's earlier working name, murmio (for example murmio.codemap and MURMIO_HUMAN_RUNS). They will be renamed in a later release. The product is VESPR.