cosmo

KO | EN

gitlite — search

#javascript #react #ai #typescript #golang #android #machine-learning #deep-learning #macos #cli #kubernetes #vue #awesome-list #open-source

(★ 16)

AI-powered data analytics with GPU-accelerated graph visualization. Two AI agents debate your data autonomously.

cosmo 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Cosmo

[![Support me on Patreon](https://img.shields.io/badge/Patreon-Support%20my%20work-FF424D?style=flat&logo=patreon&logoColor=white)](https://www.patreon.com/AndersBjarby)

**AI-powered data analytics with GPU-accelerated graph visualization.**

Drop any CSV or JSON. Two AI agents (Claude + Gemini) autonomously analyze your data, debate findings, create charts, and navigate a live graph — all in real time.

![Stack](https://img.shields.io/badge/Frontend-Vite%20%2B%20Cosmograph%202.0-blue)
![Stack](https://img.shields.io/badge/Backend-FastAPI%20%2B%20Pandas-green)
![Stack](https://img.shields.io/badge/AI-Claude%20%2B%20Gemini%20via%20OpenRouter-purple)

## Quick Start

### 1. Clone and install

```bash
git clone <repo-url> && cd cosmo-filesystem

# Frontend
npm install

# Backend
python3 -m venv venv
source venv/bin/activate
pip install -r backend/requirements.txt
```

### 2. Configure

```bash
cp .env.example .env
# Edit .env and add your OpenRouter API key
```

### 3. Run

```bash
# Terminal 1 — Backend
source venv/bin/activate
export $(cat .env | xargs)
python backend/server.py

# Terminal 2 — Frontend
npm run dev
```

Open **http://localhost:5199**

### 4. Use

1. Drop a CSV or JSON file onto the page
2. Ask a question: *"give me insights"*
3. Enable the Companion checkbox for autonomous multi-agent debate

## Features

### Multi-Agent Debate
Two LLMs analyze your data simultaneously. The analyst (Claude Sonnet 4.6) responds first, then the companion (Gemini 3.1 Pro) challenges and extends. They alternate for 4 rounds, each updating the graph.

**Companion themes:**

| Theme | Behavior |
|-------|----------|
| Default | Dig deeper, challenge conclusions |
| Devil's Advocate | Poke holes, find counterexamples |
| Anomaly Hunter | Focus on outliers and oddities |
| Optimizer | Find actionable improvements |
| Connector | Build relationships between data points |
| Storyteller | Weave data into a narrative |

### Self-Building Analytics
The AI writes custom Pandas code on the fly via `run_pandas_code`. It can also:

- **Add computed columns** — `add_column`: create derived features (risk scores, bins, categories)
- **Create graph edges** — `create_links`: define relationships between rows with code
- **Save plugins** — reusable analyses stored as named snippets

### Inline Charts
Bar, line, scatter, pie, doughnut, radar, polar area, bubble. Created by the AI via `create_chart`, rendered with Chart.js. Click any chart to expand full-screen.

### GPU Graph Visualization
Powered by [Cosmograph 2.0](https://cosmograph.app) (WebGL + DuckDB-WASM). The AI controls the graph:

- **zoom_to** — fly camera to specific nodes
- **filter_view** — isolate a subset
- **select** — highlight matching nodes
- **reset** — restore default view

### Multi-Dataset Workspace
Upload multiple files. The AI discovers foreign key relationships between them (`find_connections`), joins datasets (`join_datasets`), and builds cross-dataset graphs.

### Full Transparency
Every tool call is visible above each AI message — which Pandas code ran, what it returned, which graph commands were issued.

## Project Structure

```
cosmo-filesystem/
  index.html          # Frontend (Cosmograph + Chat UI + Charts)
  vite.config.js      # Vite dev server config
  scan.js             # Optional: filesystem scanner for ~/Projekt
  backend/
    server.py          # FastAPI + OpenRouter tool-use loop
    analyzer.py        # Pandas engine, multi-dataset, plugins
    requirements.txt   # Python dependencies
  data/
    filesystem.json    # Generated by scan.js (optional)
```

## API Endpoints

| Endpoint | Method | Description |
|----------|--------|-------------|
| `/api/chat` | POST | Send a message, get AI response with charts + graph commands |
| `/api/upload` | POST | Upload CSV/JSON dataset |
| `/api/stats` | GET | Current dataset statistics |
| `/api/schema` | GET | Current dataset schema for debugging |

## AI Tools

The LLMs have access to these tools via OpenRouter function calling:

| Tool | Description |
|------|-------------|
| `query_dataframe` | Filter, group, count, sum, describe |
| `run_pandas_code` | Execute arbitrary Pandas/NumPy code |
| `create_chart` | Generate Chart.js visualizations |
| `control_graph` | Zoom, select, filter the Cosmograph |
| `add_column` | Add computed columns to the dataset |
| `create_links` | Create graph edges with Pandas code |
| `find_connections` | Detect foreign keys between datasets |
| `join_datasets` | Merge two datasets |
| `save_plugin` | Save reusable analysis code |
| `run_plugin` / `list_plugins` | Execute saved analyses |
| `list_datasets` | Show loaded datasets |
| `save_snapshot` | Bookmark an investigation state |

## Configuration

### Models

Edit `backend/server.py` to change models:

```python
MODEL_ANALYST = "anthropic/claude-sonnet-4.6"    # Primary analyst
MODEL_COMPANION = "google/gemini-3.1-pro-preview"  # Companion researcher
```

Any OpenRouter-supported model works. See [openrouter.ai/models](https://openrouter.ai/models).

### Ports

- Frontend: `5199` (configurable in `vite.config.js`)
- Backend: `8000` (configurable in `server.py`)

## Filesystem Scanner (Optional)

To visualize your local filesystem as a graph:

```bash
npm run scan                    # Scans ~/Projekt (depth 3)
node scan.js /path/to/dir 4    # Custom path + depth
```

Generates `data/filesystem.json` which can be loaded via the upload button.

## Tech Stack

- **Graph**: Cosmograph 2.0 (WebGL, DuckDB-WASM, force-directed layout)
- **Charts**: Chart.js 4
- **Markdown**: marked.js
- **Backend**: Python 3.11+, FastAPI, Pandas, NumPy
- **AI**: OpenRouter API (Claude Sonnet 4.6 + Gemini 3.1 Pro)
- **Bundler**: Vite 6

## License

ISC

cosmo

cosmo 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

cosmo

cosmo 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README