Hi all,
I’d like to share a new package we’ve been working on: Bilge.jl — an AI coding assistant that runs directly inside the Julia REPL.
Code Base:
https://github.com/taf-society/Bilge.jl
Docs (dev):
https://taf-society.github.io/Bilge.jl/dev/
Bilge connects to OpenAI, Ollama, or any OpenAI-compatible API, and exposes a set of structured tools that allow the model to interact with your working directory (read/edit files, search, run commands, etc.). The idea is to provide a REPL-native AI workflow rather than an editor plugin.
What it does
Inside a Bilge session, the model can:
- Inspect and read project files
- Propose and apply edits
- Search through the codebase
- Execute shell commands
- Help with refactoring, documentation, debugging, and tests
It’s designed to be interactive and iterative — closer to a pair-programming experience in the terminal.
Installation (development version)
Since the package is under active development:
using Pkg
Pkg.add(url="https://github.com/taf-society/Bilge.jl")
Then:
using Bilge
# Default (OpenAI-compatible, uses OPENAI_API_KEY)
bilge()
# With Ollama
bilge(ollama=true, model="gpt-oss:latest")
# Custom endpoint
bilge(api_key="...", base_url="https://...", model="...")
Motivation
There are great AI integrations in editors, but many Julia users spend significant time in the REPL and terminal. Bilge explores what an AI assistant looks like when the REPL is the primary interface.
We’re especially interested in feedback from:
- People working on larger Julia codebases
- Package authors
- Anyone experimenting with LLM-assisted workflows in Julia
The project is early and evolving, so comments, design suggestions, and contributions are very welcome.
Happy to answer questions ![]()