DEDUPED Hot AI repos, turned into use cases Get the Monday 5

How to run Crush on a local Ollama model with Crush

Add Ollama as a provider in crushrc and register a model so Crush works fully offline.

Short answer: Install Crush and run Ollama locally; then in a crushrc add the provider: provider add ollama --type ollama --base-url "http://localhost:11434/v1"; register the model: model add ollama/llama3.3 --name "Llama 3.3" --context-window 128000. All steps.

Who it is forA developer who wants a terminal coding agent on private code without cloud calls.

The problemCloud models send your code off the machine and cost money per token.

The outcomeCrush configured with an Ollama provider and model, using your local server.

What you need first

  • A terminal on macOS, Linux, Windows, Android or BSD
  • An LLM provider: Charm's Hyper subscription, an API key for Anthropic, OpenAI, Gemini or OpenRouter, or a local server such as Ollama

Steps

  1. Install Crush and run Ollama locally
  2. In a crushrc add the provider: provider add ollama --type ollama --base-url "http://localhost:11434/v1"
  3. Register the model: model add ollama/llama3.3 --name "Llama 3.3" --context-window 128000
  4. Optionally auto-approve safe tools with permissions allow view edit
  5. Start Crush and pick the Ollama model in the model picker

Prompt for your agent

Configure charmbracelet/crush to use my local Ollama server with a project crushrc that adds the ollama provider and registers <model>, then run a small refactor on <file> with it.

Caveat. crushrc and Crush.json are trusted code that run in a shell; never launch Crush in a directory whose config you have not reviewed.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for charmbracelet/crush and "Run Crush on a local Ollama model"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer