djinn-rs

a Rust playground for running language models and some other machine learning models.

ollama-cli

a CLI/TUI for interacting with an Ollama server.

djinn-cli

the main entrypoint for running either a server (with djinn-server) or a one shot inference.

djinn-core

core ML functionality and model implementations.

djinn-server

an HTTP server with an API and streaming front-end for running models

Frontend

djinn-server ships a small UI served from ./djinn-server/assets/.

How it works

1. index.html is served as a static file by ServeDir.
2. app.js intercepts form submission and POSTs the prompt as JSON to /complete/stream.
3. The server streams tokens back as Server-Sent Events.
4. Tokens are appended to the response area as they arrive.

development

just check       # typos, clippy, tests
just coverage    # HTML coverage report → target/llvm-cov/html/index.html
just schema      # regenerate JSON Schema files after config struct changes

examples

run the server on a Macbook M-series:

cargo run --release --features djinn-core/mac -- server-config --name test

breakdown: - cargo run --release to build in release mode for best performance - --features djinn-core/mac to enable CoreML acceleration - -- everything before this are cargo args and everything after are djinn args - server-config command to run the server from a config file - --name test to run the config named test, in ./configs/server/test.toml