Skip to content

MajestixAi/Majestix-Coder

Repository files navigation

Majestix Coder

Agentic AI coding assistant for VSCode. Reads files, writes code, runs commands, and searches your codebase autonomously — with human approval gates for every destructive action.

Powered by the Majestix AI multi-model platform with automatic routing across GPT, Claude, Gemini, Nemotron, Qwen, and more.


Getting Started

  1. Install Majestix Coder from the VSCode Marketplace.
  2. Sign up at majestix.ai and create an API key from the dashboard.
  3. Open the Majestix Coder sidebar (activity bar icon) and click Set API Key.
  4. Start chatting — describe a task and watch the agent work.

New accounts include 500 free credits to get started.

Features

Agentic Loop

Describe a task and the agent plans, reads files, edits code, runs commands, and iterates until the task is done. Every write, edit, and shell command shows an approval card so you stay in control.

Four Agent Modes

  • Code — full coding agent: read, write, edit, run commands.
  • Architect — read files and write planning docs only. Great for design discussions.
  • Ask — read-only Q&A. Never modifies your code.
  • Ask-on-write — ask questions and make code edits in the same session.

Multi-Model Routing

Automatic routing across the best models for each task — no manual model switching required. Or pin a specific model from the selector if you prefer.

Session Persistence

Conversations save to .majestix/sessions/ in your workspace and restore across restarts. Browse past sessions in the date-grouped history drawer. Long conversations use rolling LLM-generated summaries so you never hit context limits.

Context Awareness

Every request automatically includes:

  • Active file, selection, and diagnostics
  • Open editor tabs and recently edited files
  • Workspace structure and git status
  • Project rules from .majestix-rules / .clinerules
  • Workspace Problems panel
  • Files you drag-and-drop or @mention

Approval Controls

  • none — always ask (default, safest)
  • readOnly — auto-approve safe commands like ls, cat, git status
  • all — auto-approve everything (use with caution)

Optional command whitelist restricts shell execution to specific tokens. Per-session "Always allow" toggle lets you trust a single run without changing global settings.

Cost Awareness

Live credit balance in the status bar, per-run cost estimate during agent loops, and a configurable cost-warning threshold that pauses expensive runs before they blow your budget.

Secure by Default

  • API keys stored in the OS keychain (never in settings.json or disk)
  • All file operations sandboxed to your workspace
  • No external telemetry — diagnostics go only to the VSCode Output channel
  • Destructive operations require explicit approval

Commands

Command Keybinding
Ask AI Cmd/Ctrl+Shift+I
Explain Code Cmd/Ctrl+Shift+E
Fix Error Cmd/Ctrl+Shift+F
Review Code Cmd/Ctrl+Shift+R
Send Terminal Output to Chat Cmd/Ctrl+Shift+T
Refactor
Generate Code
Generate Commit Message
Generate PR Description
New Chat
Session History
Run Todo Items as Agent Tasks
Create Rules File

All commands are available from the Command Palette under Majestix. Right-click a selection in the editor for quick access to Explain / Refactor / Fix / Review.

Settings

Setting Default Description
majestix.apiUrl https://api.majestix.ai Backend API URL
majestix.defaultModel "" (auto-route) Pin a specific model
majestix.defaultMode code Default agent mode for new chats
majestix.autoApprove none Auto-approve level
majestix.maxIterations 50 Max loop iterations per task
majestix.commandTimeout 60 Shell command timeout (seconds)
majestix.costWarningThreshold 100 Pause above this many credits (0 = disabled)
majestix.commandWhitelist [] Optional command token whitelist
majestix.maxContextFiles 5 Max context files per request
majestix.sessionHistory.enabled true Persist conversation history
majestix.telemetry.enabled true Local diagnostics in Output channel
majestix.terminalMirror false Mirror agent commands to a real terminal panel

Requirements

  • VSCode 1.85.0 or newer
  • Internet connection
  • A Majestix API key — create one here

Privacy

Majestix Coder sends your prompts, file contents you attach or @mention, and the environment context described above to the Majestix API for model inference. We do not train on your data. See the full privacy policy at majestix.ai/privacy.

Support

License

MIT © Majestix AI LLC