Skip to content
| Marketplace
Sign in
Visual Studio Code>Programming Languages>NeonCodex AI: Coding Agent with Claude, GPT-5.5, Gemini and GrokNew to Visual Studio Code? Get it now.
NeonCodex AI: Coding Agent with Claude, GPT-5.5, Gemini and Grok

NeonCodex AI: Coding Agent with Claude, GPT-5.5, Gemini and Grok

Nero Softwares & IT Services Pvt. Ltd.

|
73 installs
| (0) | Free
AI coding agent with 30+ models including Claude Opus 5.5, Sonnet 5.5, GPT-5.5, Gemini and Grok. Agent mode reads your project, edits files, runs commands with your approval and fixes the problems VS Code reports, with checkpoints you can restore. Custom modes, hooks, parallel agents in git worktree
Installation
Launch VS Code Quick Open (Ctrl+P), paste the following command, and press enter.
Copied to clipboard
More Info

NeonCodex AI for VS Code

Claude Opus 5.5, Sonnet 5.5, Fable 5.1, GPT-5.5, Gemini 3.8 Flash, Grok 4.7, DeepSeek V4 Pro and more, inside VS Code.

Chat about your code in the sidebar with your open files as context, fix and refactor with a diff before anything changes, generate tests, run security scans, review your git changes, and pick up any conversation you started on neoncodex.io, in the browser extension or in the CLI.


What's new in 0.12.0

  • Continue in Cloud. NeonCodex: Continue in Cloud (or the cloud button in the chat view) hands the work in your repository to a NeonCodex Cloud Agent: your unpushed commits, uncommitted changes and new files go with it, plus this conversation if you want. You can close VS Code; the agent opens a pull request when it is done, and VS Code tells you if it is still open. Works with GitHub, GitLab, Bitbucket and Azure DevOps repositories you connected at neoncodex.io. Files that look like credentials stop the hand-off until you confirm.
  • Cloud Agent runs. NeonCodex: Cloud Agent Runs lists your runs: see what one changed, open its pull request, stop it, or bring its branch back as a branch or a new worktree.

What's new in 0.11.0

  • Custom modes. Save your own modes next to Ask, Plan and Agent: instructions, a tool level (read only, edit files, or full) and a model. Put them in .neoncodex/modes/*.md to share with your team, or ~/.neoncodex/modes/ for yourself. Pick them under More in the mode bar. Command: NeonCodex: New Custom Mode.
  • Hooks. Run your own commands before and after the agent's tools, at the start, on each message and when it finishes, to format files, block risky commands or run the tests. A project's .neoncodex/hooks.json runs only after you trust it, and you are asked again when it changes. Command: NeonCodex: Edit Agent Hooks.
  • Parallel agents in git worktrees. NeonCodex: Run Agents in Parallel starts 2 to 6 agents on the same task (one model or several), each in its own git worktree, so your folder isn't touched. Review each result, then apply one, keep it as a branch, or discard it. NeonCodex: Show Agent Worktrees lists them later.
  • Browser control. In Agent mode the agent can open pages in a real browser (Chrome, Edge or Chromium), read them, click and type, for example to check the app you are building on localhost. It uses a separate temporary profile, asks before each new site, and never types passwords or card numbers. Turn it off with neoncodex.agentBrowser.

What's new in 0.10.0

  • Claude Sonnet 5.5. Pick it from the model list or the neoncodex.model setting. It always thinks, so there is no Thinking off option: the level list runs from Minimal (lowest) to Max, and more thinking costs more tokens.
  • Auto routing now sends more kinds of work to Sonnet 5.5 (writing, legal, resumes, support, explanations and questions about images) and reads your prompt more carefully. Choose a model yourself and it is used as picked.

What's new in 0.9.0

  • Approvals for everything the AI does. Before the agent runs a command, changes a file, opens a web page or uses an MCP tool, a card in the chat asks you: allow once, allow for this conversation, always allow, or deny and tell the AI what to do instead. File changes show the diff first. Keys 1-4 answer, Enter confirms.
  • Web pages. The agent can read documentation and issues on the web, one site at a time with your approval. Sites you always allow are saved on your NeonCodex account, so the web app, the browser extension and the CLI stop asking too.
  • Cloud chats ask here too. When a chat on NeonCodex wants to open a page, change data in a connected app or run code in the sandbox, you can answer in VS Code.
  • NeonCodex: Manage what the agent may do without asking lists what you always allowed and lets you take it back.

What's new in 0.8.0

  • Agent mode. Describe a change or a bug and the agent does the work in your project: it reads and searches the code, edits files, runs your tests or build (you approve each command, or allow it for the workspace), and repeats until it is done. Edits go through the editor, so open files update in place and Ctrl+Z works.
  • It uses what VS Code knows. The agent reads the Problems panel (compiler, type checker and linter errors), and uses go to definition, find references and the document outline to understand how the code fits together. After editing it checks the Problems panel again.
  • Fix with NeonCodex Agent. Hover any error or warning, open the quick fixes (Ctrl+.) and choose Fix with NeonCodex Agent. Or run NeonCodex: Fix problems with the agent from the editor's right-click menu.
  • Ask and Plan modes. Ask answers questions by reading the project, without changing anything. Plan explores the code and writes a plan; Build this plan carries it out.
  • Checkpoints. Every run lists the files it changed. Open a side-by-side diff of any file, revert one file, or restore everything.
  • Project rules. AGENTS.md, CLAUDE.md, .cursorrules, .github/copilot-instructions.md and .cursor/rules are followed automatically.
  • Codebase search. The agent keeps a local index of your project and can find code by what it does ("where is the session token checked"), not only by exact text. The index stays on your machine.
  • MCP servers. The agent can use tools from MCP servers: add them in .mcp.json, .cursor/mcp.json or .vscode/mcp.json in your workspace (you approve each one the first time), or in the neoncodex.mcpServers setting. Tools that change things ask before they run.

Agent mode needs a folder open in VS Code. Chat mode works as before.


Features

Agent mode. Pick Agent above the message box, then describe what you want. Quick actions: Fix problems, Find issues, Explain the project, Write tests, Plan a feature, Review my changes. Set neoncodex.agentAutoRun to let the agent run commands without asking.

Chat. Ask anything. The open file, your selection and other open tabs are sent as context; click the file name under the message box to turn that off for a message. Questions about "the project" also scan relevant workspace files.

Editor actions. Select code and right-click, or use the buttons in the chat: Explain, Fix, Write tests, Security scan, Refactor, Document, and Review my changes (git diff).

Apply and insert. Every code block has Copy and Insert. Fix, Refactor and Document answers have an Apply button that shows a diff first. When a fix spans several open files, Apply to N files updates them together; it only writes to files you already have open.

Inline completion. Press Ctrl+Alt+Shift+I to complete code at the cursor. Automatic ghost-text suggestions can be turned on with neoncodex.inlineCompletions.

Deep Review. On paid plans, large or multi-part questions may be handled by several sub-agents working in parallel, then combined into one answer. It runs on the model you picked; with Auto routing it uses Claude Opus 5.5, falling back to Claude Haiku 4.5 when your balance is low.


Keyboard shortcuts

Action Windows / Linux macOS
Open chat in editor Ctrl+Shift+A Cmd+Shift+A
Ask about selection Ctrl+Shift+Q Cmd+Shift+Q
Fix code Ctrl+Alt+Shift+F Cmd+Alt+Shift+F
Complete at cursor Ctrl+Alt+Shift+I Cmd+Alt+Shift+I

Getting started

  1. Create an API key at neoncodex.io/dashboard/developer.
  2. Run NeonCodex: Set API Key from the Command Palette and paste the key. It is kept in VS Code's secret storage.
  3. Click the NeonCodex icon in the Activity Bar.

Settings

Setting Default Description
neoncodex.model auto Default model. The model picked in the chat takes precedence.
neoncodex.includeFileContext true Send the open file and tabs with chat messages.
neoncodex.showDiff true Show a diff and ask before applying changes.
neoncodex.inlineCompletions false Ghost-text suggestions while typing (uses tokens).
neoncodex.apiUrl https://neoncodex.io/api Change only for a self-hosted server.
neoncodex.apiKey empty Optional. Prefer the Set API Key command.
neoncodex.agentAutoRun false Don't ask: the agent runs commands, changes files, opens web pages and uses MCP tools without asking. A project's hooks and MCP servers still ask.
neoncodex.agentBrowser true Let the agent use a browser in Agent mode (each site asks first).
neoncodex.agentBrowserHeadless false Use that browser without a window.

Privacy

  • Code is sent to NeonCodex only when you send a message or run an action.
  • No background scanning or telemetry.
  • Requests are encrypted in transit. See the privacy policy.

Free accounts can use the free models. Claude, GPT, Gemini, Grok and the other premium models, and Deep Review, need a paid plan: see plans.


MIT License. Built by Nero Softwares & IT Services Pvt. Ltd.

  • Contact us
  • Jobs
  • Privacy
  • Manage cookies
  • Terms of use
  • Trademarks
  • Your Privacy Choices
  • Consumer Health Privacy
© 2026 Microsoft