An AI coding agent in your editor — chat, read and edit files with reviewable diffs, and run shell commands you approve. Backed by seven top coding models under one HiveMind account, with live cost tracking and a spend cap.

Features

Seven models, one account. Switch mid-session between HiveMind's own fast default and Claude Sonnet 5, GPT-5.3 Codex, Gemini 3.1 Pro, Grok Build, Qwen3 Coder Plus, or Kimi K2 Code — no separate API keys or subscriptions.

File edits you can actually review. Every change renders as a real diff in the chat, and opens in VS Code's native diff editor so you can see exactly what moved before you keep it.

Shell commands ask first. Any command the agent wants to run surfaces a prompt naming the exact command. Nothing runs until you allow it.

Steer without starting over. Type while the agent is working and your message reaches it at the next step — no need to kill a long task and lose everything it has already done.

Standard and Pro modes. Standard runs everything locally and free, offline included. Pro adds hosted code-aware embeddings, so semantic search finds code by meaning rather than keyword overlap — ask "how does login work" and it finds verifyToken() even though neither word appears. Switch it from the dropdown in the chat toolbar; Pro needs a signed-in account and bills against your balance.

Costs stay visible. A live readout shows tokens and spend per turn and per session. Set a budget and the session stops cleanly when it's reached, at a turn boundary, never mid-edit.

Long sessions keep working. Context compacts automatically as it fills, and sessions are saved as you go — close the window and pick the conversation back up later.

Every chat is a tab. Conversations open in the editor area, not the sidebar, so you can keep the Explorer where it is, run several chats at once, and split them beside the code they're about. The sidebar lists your conversations instead: click one to reopen it exactly where you left off.

Your conversations stay on your machine. Transcripts are written to ~/.config/hivemind/sessions on this computer and nowhere else — plain JSON files you can read, back up, or delete yourself. Nothing is uploaded. Chats you haven't opened in 14 days are deleted automatically; change that with hivemind.sessionRetentionDays, or set it to 0 to keep everything. A chat that's open in a tab is never deleted, however old it is.

Getting started

This extension is a front end for the hivemind command-line tool — install that first, or Start Session will report that the CLI wasn't found.

macOS / Linux

curl -fsSL https://raw.githubusercontent.com/BibhabenduMukherjee/HiveMind-releases/main/install.sh | bash

Windows (PowerShell)

irm https://raw.githubusercontent.com/BibhabenduMukherjee/HiveMind-releases/main/install.ps1 | iex

Then:

Run HiveMind: Sign In from the Command Palette to authenticate.
Open HiveMind in the activity bar and click + to start a chat. It opens in its own editor tab; click + again any time for another.

If the CLI lives somewhere else, point hivemind.executablePath at it in settings.

Commands

Command	What it does
`HiveMind: New Chat`	Open a new conversation in its own editor tab
`HiveMind: Refresh Conversations`	Re-read the saved conversation list
`HiveMind: Sign In`	Authenticate your HiveMind account
`HiveMind: Select Model`	Switch the active model
`HiveMind: Select Reasoning Effort`	Set thinking depth (models that support it)
`HiveMind: Set Session Budget`	Cap spend for this session
`HiveMind: Undo Last Turn`	Revert the last turn's file changes
`HiveMind: Force Compact History`	Summarize older turns to free context

Settings

Setting	Default	Purpose
`hivemind.executablePath`	`hivemind`	Path to the CLI binary
`hivemind.mode`	`standard`	`pro` enables hosted code-aware embeddings
`hivemind.defaultModel`	(CLI default)	Model to select on start
`hivemind.defaultReasoningEffort`	(none)	Reasoning effort on start
`hivemind.defaultBudgetUsd`	(none)	Session spend cap in USD
`hivemind.apiKey`	(none)	API key, if not using sign-in
`hivemind.extraArgs`	`[]`	Extra CLI arguments