ollama_code

A coding assistant with ollama as LLM backend. This is very experimental software.

The premise is to have a full replacement for Claude Code, backed by Ollama and an LLM that runs on the local machine.

Requirements

For this to work, you need a computer with a dedicated graphics card.

Confirmed to work:

• Apple Computers with M-series chipsets

Other requirements:

• Ollama
• Rust 1.86 or later

Usage

To get started, you need to pull llama3.1:8b first:

$ ollama pull llama3.1:8b

Once the model is downloaded, run it. You'll get a prompt:

cargo run
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.21s
     Running `target/debug/ollama_code`
Let's get started. Press [ESC] to exit.
? Prompt : Tell me about this codebase
📁 Listing directory: .
   Found 15 items
📄 Reading file: Cargo.toml
   Read 1360 characters
📄 Reading file: README.md
   Read 61 characters

Based on the files I found, Cargo.toml and README.md, it appears that this codebase is for a Rust project called 'ollama_code' which uses an ollama backend. The project has several dependencies including clap, reqwest, serde, and tracing among others. It seems to be designed for building a coding assistant with an ollama LLM (Large Language Model) as the backend.

Limitations

The biggest limitation is the graphics card and its VRAM. The default model here right now is llama3.1:8b to ensure it'll work on machines with 16GB of RAM. Additionally, we can only use Open Source models that support tool calling.

Product limitations:

• No write access to your codebase (yet)
• It doesn't have as much memory as Claude Code - you cannot build infinitely long conversations.