IsoCode

Local agentic AI coding assistant for VS Code. ReAct-style agent that reads, edits, and runs tools in your workspace—runs on Ollama, LM Studio, or any OpenAI-compatible API. No cloud required.

Screenshot & Demo

IsoCode in VS Code — Chat, Agent, and Agent+

IsoCode chat interface in VS Code sidebar

Demo video (≈1 min): Watch here

What it does

IsoCode is a local-first coding agent inside VS Code. You get:

Chat — Direct Q&A with your local model (streaming).
Agent — Multi-step ReAct loop: plan → act (read/write files, search, shell, git, lint, test) → observe. Proposes file edits as diffs; you approve or reject.
Agent+ — Same agent with full autonomy: auto-applies edits, more steps, no approval prompts.

All inference runs on your machine (Ollama, LM Studio) or your own API key (OpenAI). The server and extension are open source; no telemetry, no locked-in cloud.

Why IsoCode vs other AI IDEs

Many IDEs and coding assistants rely on cloud APIs and API keys: your code and requests go to third-party services, usage is metered, and you depend on their pricing and policies.

	IsoCode	Other AI IDEs (cloud / API-key based)
Where it runs	VS Code (or any editor with the extension)	Often tied to a specific IDE or SaaS
AI	Your own models (Ollama, LM Studio) or your API key	Vendor’s cloud models; optional bring-your-own in some
Cost	Free (you pay for hardware or your own API if you choose)	Typically subscription or per-use API costs
Privacy	Code stays on your machine or your API; no required cloud	Code and prompts may be sent to vendor services
Lock-in	Open source; switch editors or models anytime	Tied to vendor product and pricing

Use IsoCode when you want a local-first agent (plan → edit → approve) inside your editor, with no required subscription and no required cloud. Use other AI IDEs when you prefer their integrated experience and are fine with their API keys and data policies.

Features

Area	Details
Modes	Chat (streaming), Agent (diff approval), Agent+ (autonomous)
Tools	`read_file`, `write_file`, `replace_in_file`, `apply_diff`, `codebase_search`, `search_files`, `list_files`, `glob_files`, `run_shell`, `run_lint`, `run_tests`, `git_status` / `git_diff` / `git_commit` / `git_log` / `git_branch`, `memory_read` / `memory_write`, browser (screenshot, click, type, extract), MCP tools
Context	Add files to context, @-mention, auto-gather from codebase; project rules from `.isocode/rules.md`, `AGENTS.md`, `.cursorrules`
Stack	Node.js server (Express) + VS Code extension (TypeScript); supports Ollama, LM Studio, OpenAI-compatible APIs

Requirements

Node.js 18+
VS Code 1.85+
LLM: Ollama (e.g. ollama pull qwen2.5-coder:7b), LM Studio, or any OpenAI-compatible endpoint

Install

1. Clone and run the server

git clone https://github.com/rushi32/IsoCode.git
cd IsoCode
npm install
cp .env.example .env
# Edit .env if needed (LLM_API_BASE, LLM_MODEL_ID)
npm start

Server runs at http://localhost:3000 by default.

2. Run the extension (development)

In a separate terminal:

cd IsoCode/extension
npm install
npm run compile

Open the extension folder in VS Code → press F5 (Extension Development Host) → open the IsoCode sidebar from the activity bar.

3. Configure (optional)

Server: .env — LLM_API_BASE (e.g. http://localhost:11434/v1 for Ollama), LLM_MODEL_ID, PORT.
Extension: VS Code Settings → search “IsoCode” — set Server URL if not http://localhost:3000.

Usage

Open the IsoCode view (sidebar).
Choose Chat, Agent, or Agent+.
Select a model (dropdown; populated from your server).
Type your request. In Agent/Agent+ the assistant will plan, call tools, and (in Agent mode) propose diffs for you to approve.

Project-specific instructions: add .isocode/rules.md in your workspace. Optional MCP servers: configure in the extension settings or .isocode/mcp-servers.json.

Project structure

IsoCode/
├── assets/              # Screenshot & demo video 
├── server/              # Agent server (Node)
│   ├── index.js         # HTTP + SSE
│   ├── agent.js         # ReAct loop
│   ├── tools.js         # Tool implementations
│   ├── llm.js           # Ollama / LM Studio / OpenAI
│   ├── config.js        # Env + user-config
│   └── ...
├── extension/           # VS Code extension
│   ├── src/             # TypeScript
│   ├── media/           # Webview UI, styles
│   └── package.json
├── .env.example
├── CONTRIBUTING.md
├── OPEN_SOURCE_GUIDE.md
└── LICENSE

Docs

License

MIT. See LICENSE.