DSAgent — AI Data Science Assistant for VS Code
AI-powered data science agent that integrates directly into VS Code. Analyze data, create visualizations, and build models using natural language.
Features
- Dual Chat Interface — Use the
@dsagent native chat participant or the custom chat panel
- Session Management — Create, resume, and manage analysis sessions
- Live Artifacts — View generated plots, CSVs, and other outputs in the sidebar
- Notebook Sync — Session notebooks sync with the DSAgent kernel
- Variables Explorer — Inspect DataFrames, variables, and imports in real time
- File Attachments — Upload CSV, Excel, JSON, Parquet, and other data files
- Human-in-the-Loop — Approve, reject, or modify agent actions before execution
Requirements
- A running DSAgent server (
dsagent serve)
- VS Code 1.85.0 or later
Getting Started
Start the server:
dsagent serve
Open VS Code — the extension auto-connects to http://localhost:8000
Open the chat: Cmd+Shift+P → DSAgent: Open Chat
Remote Server
To connect to a remote DSAgent server:
Cmd+Shift+P → DSAgent: Configure Server Connection
- Enter the server URL (e.g.
https://remote-host:8000)
- Enter the API key (if the server has
DSAGENT_API_KEY set)
Or configure in settings:
{
"dsagent.serverUrl": "https://remote-host:8000",
"dsagent.apiKey": "your-api-key"
}
Extension Settings
| Setting |
Default |
Description |
dsagent.serverUrl |
http://localhost:8000 |
DSAgent server URL |
dsagent.apiKey |
|
API key for remote server authentication |
dsagent.autoConnect |
true |
Auto-connect on startup |
dsagent.model |
gpt-5.1 |
Default LLM model |
dsagent.hitlMode |
none |
Human-in-the-Loop mode |
Available Models
gpt-5.1 (default)
gpt-5.2
gpt-5.2-codex
claude-sonnet-4-20250514
claude-opus-4-20250514
gemini-3-pro-preview
gemini-3-flash-preview
groq/openai/gpt-oss-120b
groq/qwen/qwen3-32b
groq/moonshotai/kimi-k2-instruct-0905
openrouter/qwen/qwen3-coder-next
openrouter/minimax/minimax-m2.1
HITL Modes
none — Agent runs autonomously
plan_only — Approve before executing a plan
full — Approve every plan and code execution
plan_and_answer — Approve plans and final answers
on_error — Approve only when errors occur
License
MIT
| |