InsightAI
InsiPilot for VS Code

Your Code.
Your Models.
Your Control.

A self-hosted AI coding assistant for VS Code. It runs on your own Ollama, so your code never leaves your machine.

Install from VS Code Marketplace Open in VS Code

Installing by hand? Download .vsix then choose Install from VSIX in the Extensions view menu.

  • Free
  • v1.0
  • VS Code 1.100+
cart.ts Illustration
12export function totalWithDiscount(items, pct) {
13  const sum = items.reduce((a, i) => a + i.price, 0);
14- return sum - pct;
14+ return sum * (1 - pct / 100);
15}
Accept Reject
Any bugs in src/?
  1. Find totalWithDiscount
  2. Read cart.ts, cart.test.ts
  3. Test 1 test failing
  4. Change cart.ts:14
  5. Test again all passing

cart.ts:14 subtracts the percentage instead of applying it. Fix is ready for review.

Why InsiPilot

An AI coding agent that runs entirely on your machine, even with a 7B model.

Local & private

InsiPilot runs on your own Ollama server. Your code never leaves that machine, and it keeps working offline. No API key, no cloud.

Chats listed per project in the side bar, with a green dot on the active chat

Built for small local models

Each request is broken into small steps: find the code, read it, run the tests, make the change and test again. The model handles one step at a time, so even a 7B model stays on track.

Step cards in the chat as InsiPilot works through a request

You approve every change

Every edit shows up as a diff with Accept / Reject. Terminal commands wait for Run, Always allow or Reject. Undo restores every file an answer changed.

A proposed change shown as a diff with Accept and Reject buttons
How it works

Up and running in three steps.

  1. Install Ollama

    Get Ollama 0.9.0 or newer on your computer or on your own GPU server.

    ollama.com/download

  2. Install InsiPilot, then Install model

    Add the extension from the Marketplace and click Install model. The default model downloads right inside VS Code.

  3. Ask

    Open a chat next to your code and ask about it, hunt for bugs or request a change. You review every edit before it lands.

What you can ask

Ask about your code. It checks its work.

You askInsiPilot
What does totalWithDiscount do? Finds the code, reads it and answers from it.
Any bugs in src/? Reads the code, runs the tests as evidence, lists bugs with file:line.
Fix, add or rename something Reads the code, runs the tests, makes the change, runs the tests again and fixes what broke.
Make the tests pass Runs them, fixes the failures and repeats, up to three rounds.
Inside the extension

Made for the way you already work.

Knows your project

Reads and searches the project, sees VS Code's errors and runs your tests to check its work.

Chats next to your code

Each chat opens as a tab beside your code. The side bar lists every chat for the project; rename or delete them, and the latest one reopens on its own.

Context, your way

The open file or selection is sent along automatically. Type @ to attach another project file, or use + and then Upload file to add any text file, like a log or a config file.

Project instructions

Reads AGENTS.md or .insipilot/INSTRUCTIONS.md at the start of every chat.

Slash commands

/handoff saves progress to .insipilot/HANDOFF.md, /standup reads it back with the git state, /clear starts fresh.

Use a bigger machine

Run the model on a GPU server over an SSH tunnel with the insipilot.endpoint setting.

Chats are saved in ~/.insipilot, outside your project, so they are never committed. Attachments are text only: images, PDFs and Office files are not supported.

Pick a model for your machine

From 0.5B to 32B. 7B by default.

Choose any qwen2.5-coder size. InsiPilot shows the download size, warns you when a model is too big for your RAM, and installs it from inside the extension.

  • 0.5B
  • 1.5B
  • 3B
  • 7B
  • 14B
  • 32B
Default
qwen2.5-coder:7b
Download
~4.7 GB
Free RAM needed
~7 GB

16 GB of RAM runs the default model comfortably.

Model picker showing download sizes and a RAM warning
Command approval box with Run, Always allow and Reject
Safety

Guardrails are on by default.

  • Only touches files in the open workspace, and never writes to .git or node_modules.
  • Files that often hold secrets (.env, private keys, .ssh, .npmrc and similar) are skipped in searches, and reading them needs your approval.
  • Commands stop after 2 minutes.
  • Dangerous commands (sudo, formatting a disk, deleting your home folder and the like) are always refused.
FAQ

Questions, answered.

Do I need a GPU?

No. A GPU helps but is not required: 16 GB of RAM runs the default 7B model comfortably. Speed depends on your hardware.

Does it cost anything?

No. InsiPilot is free on the VS Code Marketplace.

Is my code sent anywhere?

No. Requests only go to the Ollama endpoint you configure, which by default is Ollama on your own machine.

Can I run the model on my own server?

Yes. Open an SSH tunnel to the server and point the insipilot.endpoint setting at it.

Which editors are supported?

VS Code 1.100 or newer, with Ollama 0.9.0 or newer. InsiPilot is published on the VS Code Marketplace only.

What is it best at?

Questions about your code, finding bugs, renames, writing tests and small changes.

Can the model decide the steps itself?

Yes. The default guided workflow asks the model one step at a time. Set insipilot.workflow to agent to let the model decide every step.

Your code stays yours. Start now.

Free on the VS Code Marketplace. Needs VS Code 1.100+ and Ollama 0.9.0+.

Install from VS Code Marketplace Open in VS Code
code --install-extension doublekien.insipilot

Or download the .vsix (v1.0.0).

© 2026 Duong Kien. Available on the VS Code Marketplace.