Skip to content
| Marketplace
Sign in
Visual Studio Code>AI>Magnexis Agent StudioNew to Visual Studio Code? Get it now.
Magnexis Agent Studio

Magnexis Agent Studio

magnificent-language

| (0) | Free
A local-first coding agent workbench for VS Code. Magnexis is a connector, not a model: it routes chat, edits, and tools to any OpenAI-compatible provider you choose, using your own API key.
Installation
Launch VS Code Quick Open (Ctrl+P), paste the following command, and press enter.
Copied to clipboard
More Info

Magnexis Agent Studio logo

Magnexis Agent Studio

A local-first coding-agent workbench for VS Code, desktop, web, and CLI.

Version 0.4.0 TypeScript Approval model Local first Multi provider Contributing Security

Magnexis Agent Studio is an original developer tool for working with coding agents across multiple surfaces. It combines:

  • a VS Code-native chat and agent companion
  • a desktop workspace for provider management, run inspection, and settings
  • a focused web workbench for launching coding tasks
  • a local CLI for provider checks, doctor commands, and scripted runs

Magnexis is a connector, not an LLM provider. It does not host its own model, inference endpoint, or hidden routing layer. It connects your workspace to the provider and model you choose under your own credentials.

Version 0.4.0 includes:

  • the VS Code Agent Lab experience
  • the Electron desktop workbench
  • the web dashboard workspace
  • shared provider and model routing
  • verified model-context metadata
  • per-model local guardrails
  • Supabase-based local auth scaffolding
  • update notification plumbing
  • release bundling for previews, docs, media, and VSIX packaging

Table of Contents

  • Why Magnexis
  • Surface Previews
  • What You Can Do
  • Architecture at a Glance
  • Supported Providers
  • Quick Start
  • Run Each Surface
  • Authentication Setup
  • Configuration
  • Safety Model
  • Release Workflow
  • Monorepo Layout
  • Troubleshooting
  • Contributing and Security
  • Status

Why Magnexis

Magnexis is designed for people who want a serious coding-agent product shape without handing control over to a black box. The product direction is:

  • local-first by default
  • approval-first for file edits and command execution
  • multi-provider instead of model-locked
  • repo-aware across editor, desktop, web, and CLI
  • original Magnexis branding and implementation rather than copied product code or assets

Where Magnexis talks to OpenAI directly, it follows the current official OpenAI Node SDK pattern:

  • Node package: openai
  • preferred API path: Responses API
  • standard env var: OPENAI_API_KEY
  • CLI override support: --api responses or --api chat-completions

For non-OpenAI providers, Magnexis preserves its OpenAI-compatible routing path so local and third-party endpoints continue to work.

Surface Previews

These are real project screenshots stored in the repository so the GitHub README shows the current UI rather than mockups.

Web Workspace

Magnexis web dashboard workspace overview

Magnexis web dashboard composer and route controls

The web dashboard is the focused agent workspace. It centers the composer, model/provider routing, context controls, safety state, and route inspector instead of trying to be a generic analytics dashboard.

VS Code Agent Lab

Magnexis VS Code split-panel overview

The VS Code surface is designed for compact sidebar use and split-panel use. It includes thread switching, provider selection, current-file and workspace context, slash commands, diff-aware flows, and auth-aware status controls.

What You Can Do

Across the product

  • connect external LLM providers with your own API keys
  • select verified models and inspect known context-window metadata
  • chat with repository context
  • plan coding tasks before applying changes
  • review diffs before accepting edits
  • keep terminal and file edits behind approval gates
  • track workflow templates and route preferences

In VS Code

  • open the Agent Lab sidebar
  • run chat beside the editor or in a separate window
  • attach current file, selection, or workspace context
  • switch providers and models from the in-product picker
  • restore sessions and use protected commands through the shared auth layer

In Desktop

  • manage providers in a higher-bandwidth workspace
  • review model stats and route settings
  • inspect agent activity, settings, and tool capabilities
  • use the integrated llm-stats.com viewport inside the native shell

In Web

  • start focused coding tasks from a clean agent-workspace surface
  • inspect safety, docs, and route summaries from the right-side inspector
  • use branded dropdowns with provider marks and model routing context

In CLI

  • run provider and workspace diagnostics
  • inspect current routing
  • launch scripted runs for review or planning flows

Architecture at a Glance

Magnexis is organized as a monorepo:

apps/
  desktop/           Electron desktop workbench
  vscode-extension/  Extension app wrapper and package metadata
  web-dashboard/     Focused browser workspace

packages/
  agent-core/        Shared agent orchestration primitives
  auth/              Supabase-backed auth provider and callback flow
  config/            Shared config model and defaults
  indexer/           Workspace indexing helpers
  llm-router/        Provider presets, verified models, routing metadata
  shared/            Shared utilities
  tools/             Tool contracts and registration model
  types/             Shared TypeScript contracts
  ui/                Shared UI tokens and metadata

src/
  VS Code runtime host, webview glue, CLI commands, settings, tools

media/
  Shared branding, provider marks, CSS, and webview script

The current codebase keeps the shared product language in the packages while each surface consumes the same routing, provider, config, and auth ideas.

Supported Providers

Magnexis currently exposes 19 provider presets:

  • OpenAI
  • Anthropic
  • Google Gemini
  • Z.ai
  • Mistral
  • Kimi / Moonshot
  • Groq
  • OpenRouter
  • Together AI
  • DeepSeek
  • xAI
  • Perplexity
  • Cerebras
  • Fireworks AI
  • SambaNova
  • NVIDIA NIM
  • Ollama
  • LM Studio
  • Custom OpenAI-compatible endpoint

Important distinction:

  • Magnexis provider presets are routing presets
  • verified model metadata is pinned where official docs are available
  • local and gateway routes stay marked as endpoint-reported rather than invented

For provider policy and verified-model notes, see docs/PROVIDERS.md.

Quick Start

Requirements:

  • Node.js 20+
  • npm
  • VS Code 1.92+
  • Python with Playwright only when regenerating screenshots

Install and compile:

npm install
npm run compile

Run Each Surface

Primary start commands

npm run start:extension
npm run start:desktop
npm run start:web
npm run start:cli

Preview-oriented commands

npm run start:extension:preview
npm run start:desktop:preview
npm run start:web:preview

CLI examples

npm run cli -- status
npm run cli -- doctor --workspace .
npm run cli -- providers openai
npm run cli -- run "Review this repository and summarize the main risks." --provider openai --api responses

After npm link, you can call:

magnexis status
magnexis run "Summarize this repo" --provider openai --api responses

VS Code extension

  1. Open this repository in VS Code.
  2. Press F5.
  3. Choose Run Magnexis Agent Lab.
  4. In the Extension Development Host, run Magnexis: Open Sidebar.
  5. Use Magnexis: Pin Chat Near Editor or the pin button in the Agent Lab header to keep the chat beside your code.
  6. Open the provider picker, select a provider and model, then save the API key.

Build and install the VSIX:

npm run package
code --install-extension magnexis-agent-studio-0.4.0.vsix --force

Publish note:

  • run vsce publish from the repository root, because the root package.json is the real extension manifest
  • apps/vscode-extension/package.json is only a workspace wrapper and now proxies publishing back to root through npm run publish

Desktop application

npm run desktop

The desktop runtime stores provider keys through Electron safeStorage. The Model Stats area embeds llm-stats.com in a restricted native view.

Authentication Setup

Magnexis includes a shared Supabase-backed auth layer for VS Code and desktop with:

  • email/password sign-in
  • sign-up
  • GitHub OAuth when configured
  • Google OAuth when configured
  • localhost callback capture
  • secure session restore
  • protected feature gating

Environment

Copy .env.example to .env and fill in:

SUPABASE_URL=
SUPABASE_ANON_KEY=
AUTH_CALLBACK_URL=http://localhost:54321/auth/callback
AUTH_CALLBACK_PORT=54321
APP_DEEP_LINK_SCHEME=magnexis
MAGNEXIS_GITHUB_RELEASES_URL=
MAGNEXIS_VSCODE_MARKETPLACE_URL=
MAGNEXIS_DESKTOP_MANIFEST_URL=
MAGNEXIS_UPDATE_MANIFEST_URL=

Redirect URLs

In Supabase Auth, configure:

  • http://localhost:54321/auth/callback
  • magnexis://auth/callback

Validation

npm run compile
npm run auth:check

VS Code auth flow

  • Magnexis: Sign In
  • Magnexis: Sign Up
  • Magnexis: Sign Out
  • Magnexis: Show Account
  • Magnexis: Refresh Session

Sessions are stored in VS Code SecretStorage.

Desktop auth flow

Open Settings > Account and sync and use the account actions there. Sessions are stored with Electron safeStorage encryption.

Configuration

Project configuration lives in .magnexis/config.json; user-level defaults live in the user's .magnexis/config.json. VS Code settings use the magnexis.* namespace.

Setting Default Purpose
magnexis.provider zai Active external provider
magnexis.model glm-5.1 Model identifier sent to that provider
magnexis.approvalMode agent chat, agent, or fullAccess
magnexis.reasoningEffort medium Requested planning effort
magnexis.maxToolRounds 12 Maximum model/tool turns
magnexis.commandTimeoutMs 120000 Terminal command timeout
magnexis.autoContext true Attach recent editor context
magnexis.maxWorkspaceFiles 80 Workspace-map path cap
magnexis.maxFileBytes 24000 Per-file context byte cap

Per-model limits are local caps. They are not claims about account-level provider quotas.

Safety Model

Manual approval is the default posture.

Read-only inspection can run directly, but the following remain guarded:

  • workspace edits
  • shell commands
  • package installation
  • browser-control style tools
  • custom tool manifests
  • sensitive-file changes

Additional security expectations:

  • provider keys never belong in Git, config files, or exported logs
  • Supabase uses PKCE and a browser-based OAuth flow
  • raw passwords, tokens, codes, and service-role keys must never be logged
  • full-access mode should only be enabled in trusted workspaces

See docs/SECURITY.md and SECURITY.md.

Release Workflow

Package the extension and assemble the release bundle:

npm run package
npm run release:bundle

Native desktop packaging is wired for future release builds through electron-builder:

npm run package:desktop:dir
npm run package:desktop:win
npm run package:desktop:mac
npm run package:desktop:linux

Outputs:

  • VSIX: magnexis-agent-studio-0.4.0.vsix
  • local release folder: dist/releases/magnexis-agent-studio-0.4.0/
  • zipped release bundle: dist/releases/magnexis-agent-studio-0.4.0.zip

The release bundle includes:

  • docs
  • previews
  • screenshots
  • media assets
  • VSIX
  • CHANGELOG.md
  • RELEASE_NOTES.md
  • release-manifest.json

What is not fully automated yet:

  • desktop native installers
  • VS Code Marketplace publication
  • GitHub Releases publication
  • signed multi-platform desktop distribution

Those still require external publishing credentials and a real distribution pipeline.

Monorepo Layout

apps/
  desktop/
  vscode-extension/
  web-dashboard/

packages/
  agent-core/
  auth/
  config/
  indexer/
  llm-router/
  shared/
  tools/
  types/
  ui/

src/
media/
docs/
workspaces/runtime/

Reference docs:

  • docs/ARCHITECTURE.md
  • docs/DEVELOPMENT.md
  • docs/PROVIDERS.md
  • docs/REPO_STRUCTURE.md

Troubleshooting

Callback never returns

Make sure AUTH_CALLBACK_PORT is free and the redirect URL exactly matches AUTH_CALLBACK_URL.

OAuth provider mismatch

Confirm the same provider is enabled in Supabase and in the flow you launched.

Missing env variables

Run:

npm run auth:check

Expired session

Use Magnexis: Refresh Session first. If refresh fails, sign in again.

Preview port already in use

The preview server automatically falls forward to another local port if 4173, 4174, or 4175 are occupied.

Browser cannot open

Check that the local machine allows VS Code or Electron to open the default system browser.

Contributing and Security

  • contribution flow: CONTRIBUTING.md
  • root security policy: SECURITY.md
  • implementation security details: docs/SECURITY.md

Status

This is an active 0.4.0 development build.

The strongest current path is:

  • OpenAI-compatible provider routing
  • VS Code Agent Lab workflows
  • desktop provider management and model stats
  • web workspace previews and release bundling

Still intentionally incomplete:

  • cloud sync
  • billing
  • marketplace publication
  • signed desktop installers
  • complete remote release automation

License

Proprietary. See LICENSE.

  • Contact us
  • Jobs
  • Privacy
  • Manage cookies
  • Terms of use
  • Trademarks
© 2026 Microsoft