Prime Agent for VS Code

Run Prime Agent — the self-improving RLM coding agent — inside VS Code.

A chat sidebar to give the agent tasks, watch it work, and manage the files it changes — without leaving your editor.

Features

Chat that shows the work

Streamed replies with collapsible reasoning. Python cells, shell cells, and file edits each get their own card, showing the source it ran, how long it took, and output and errors separately — so you see what the agent did, not just its summary.

Real subagents, live

Prime Agent spawns child agents that work in parallel. The Subagents view shows them as a live tree — status, activity, tokens, and model — and lets you watch a child's work as it happens. Delegation, recursion, skills, goals and heartbeats all behave exactly as they do in the terminal.

Review and revert every change

The Changes view lists every file the agent created, modified, or deleted, with line counts, native VS Code diffs, and one-click revert. Changed lines are marked in the gutter of open editors. Your Git history is never touched.

Pick your model, keep your keys

Model and reasoning-level pickers, live context and cost. Bring any provider — OpenAI, Anthropic, or a custom OpenAI-compatible gateway you configure in a few clicks. Credentials stay on your machine.

Long-running, resumable work

Persistent goals, context compaction, and continual refinement. Sessions outlive the window and keep their state — reopen VS Code and pick up where you left off, or take over a session that is still running in the background.

Safe by default

An optional approval gate can require your confirmation before the agent runs a tool or edits a file. Off by default; one setting turns it on.

Move your setup between machines

Export the agent's learned memories, skills and settings to a single file, and import it on another machine — so a second computer starts already knowing what the first one learned.

Getting started

Open the Prime Agent view in the activity bar and follow Get Started. If Prime Agent isn't installed, the extension offers to install it for you — including the Python kernel it needs. Then enter your provider key, pick a model, and start typing.

Everything runs locally. The extension launches Prime Agent and talks to it directly; nothing is sent to any server but your chosen model provider.

Requirements

• VS Code 1.85+
• Node.js 20.6+ on your PATH
• On Windows, Git for Windows (for the agent's shell cells)

The extension checks all of this and offers a one-click fix for anything missing.

Settings

Support & contact

Questions, bug reports, or feedback: abald.fh@gmail.com

License & credits

Licensed under the MIT License. Built on Prime Agent by Prime Intellect. This extension is an independent client and is not affiliated with Prime Intellect.