> Hello, friend.

An AI dev environment
that runs on your machine.

Monaco editor, a real terminal, file explorer, built-in browser and a tool-using AI agent — together in one desktop window. Bring your own OpenAI-compatible model. Your keys never leave your computer.

Download for desktop See what's inside

v0.1.0 · MIT licensed · Windows, macOS & Linux

agent.ts tokens.css mcp.ts

mastermind

Explorer

src

main

agent.ts

agentTools.ts

mcp.ts

lsp.ts

terminal.ts

renderer

App.tsx

package.json

README.md

1import { runTools } from './agentTools'

2import { getActiveProvider } from './providers'

4// stream a turn, then execute the tool calls

5export async function step(ctx: Context) {

6 const provider = getActiveProvider()

7 const turn = await provider.chat({

8 messages: ctx.history,

9 tools: ctx.tools,

10 maxSteps: 32

11 })

12 return runTools(turn.calls, ctx.root)

13}

Agent

You

Find the retry bug in the provider stream and fix it.

✓search "retry" · 6 hits

✓read_file providers.ts

✓edit_file providers.ts +8 −3

✓run_command npm run typecheck

Mastermind

The backoff never reset after a 429. Fixed and typechecked clean.

/plan│

~/mastermind $ npm run typecheck

✓ tsc --noEmit · 0 errors 1.8s

One window

Everything the agent needs, already in the room.

No browser tabs, no context copy-paste. The editor, shell, filesystem and web are all first-class surfaces — for you and for the agent.

Monaco editor + LSP

The editor from VS Code, with real language servers for completions, diagnostics and go-to-definition.

typescript · javascript · python · rust · go · c/c++

A real terminal

Full PTY-backed shells with xterm rendering. The agent can propose a command and you approve it, or let it run.

run_command · propose_command

Built-in browser

Open your dev server next to the code. The agent reads the DOM, screenshots the page and describes what it sees back to itself.

browser_open · browser_read · browser_screenshot

MCP servers

Extend the agent with Model Context Protocol servers. Add them from settings or let the agent wire one up itself.

add_mcp_server · list_mcp_servers

Skills & memory

Teach it once. Reusable skills and persistent project memory keep conventions and decisions across sessions.

~/.mastermind/skills · read_memory · write_memory

Git & GitHub built in

Review a diff, commit from the chat, or walk the repo backwards in time to find where a change came from.

github_commit · github_status · git_time_travel

SSH remote hosts + SFTP Voice input Whisper transcription Vision QA screenshots read back Command palette ⌘K Project search across the tree Sessions parallel agent threads 20+ themes Tokyo Night, Nord, Dracula…

The agent

It doesn't just suggest. It works.

The agent reads and writes real files, greps the tree, runs commands, opens URLs and commits. Slash commands aim it at a specific job.

/plan /diff /code-review /security-review /verify /run /init /memory /context /compact /worktree /mcp /skill-creator /status

agent · session 3

> /security-review src/main/mcp.ts

✓ read_file src/main/mcp.ts

✓ search spawn( · 4 hits

✓ read_memory project conventions

✓ web_search MCP stdio hardening

▲ 1 finding · server args are not escaped

✓ edit_file mcp.ts +12 −4

✓ propose_command npm run typecheck

Patched and verified. Want the diff?

Your model, your keys

Bring your own provider.

Point Mastermind at any OpenAI-compatible endpoint — a hosted frontier model or something running on localhost. Set a base URL, a model and a key. That's it.

Endpoint: Any OpenAI-compatible base URL, hosted or local
Keys: Stored locally, encrypted by your OS keychain — never bundled with the project
Vision: Auto-detected on capable models for screenshot review
Switching: Multiple providers side by side, swapped per session

Settings · Providers

base_url

https://api.your-provider.com/v1

model

your-model-name

api_key

•••••••••••••••••••• local only

✓ connection ok · vision supported

An AI dev environmentthat runs on your machine.

Everything the agent needs, already in the room.

Monaco editor + LSP

A real terminal

Built-in browser

MCP servers

Skills & memory

Git & GitHub built in

It doesn't just suggest. It works.

Bring your own provider.

Built for all three desktops.

Windows

macOS

Linux

Hello, friend.

An AI dev environment
that runs on your machine.