whittle

swizzcheeze/whittle

★ 0 stars Python 🤖 AI/LLM Updated 1d ago

Local-first, LLM-driven workspace for paring down text datasets. CLI + MCP server, embed via Ollama, project with UMAP, curate by conversation.

View on GitHub →

Quick Install

Copy the config for your editor. Some servers may need additional setup — check the README.

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "whittle": {
      "command": "uvx",
      "args": [
        "whittle"
      ]
    }
  }
}

Or install with pip: pip install whittle

README Excerpt

<div align="center"> *Embed locally · Project to 2D · Curate by conversation · Save the kept subset* </div> --- > **Why this exists.** The bottleneck in modern ML is no longer model architecture — it's data quality. But every "data-centric AI" tool ships your text to a SaaS dashboard. This project keeps the entire loop on your laptop: embeddings via Ollama, dimensionality reduction via UMAP, an in-browser scatter view via Spotlight, and an MCP server so any LLM client (Claude Desktop, Claude Cod

Tools (10)

close_viewerfind_near_duplicatesfind_outliersget_rowlaunch_viewerloadmarkproject_2dsearchstatus

whittle

Quick Install

README Excerpt

Tools (10)

Related AI/LLM Servers