FYOS07 · Python · Web · v0.3.0 · MIT
FLE Studio
A local chat and lab for llama.cpp and FLE: install, models, live data, all on your own machine.
- operating systems: macOS, Linux, Windows
- 3
- accounts, telemetry or cloud
- 0
- default – reachable on the network only on request
- 127.0.0.1
What it is
FLE Studio is one local app: a chat in your browser, live numbers while the answer forms, a model loader and launch profiles. It works with the FLE engine and with plain llama.cpp on macOS, Linux and Windows – no account, no telemetry, nothing runs in the cloud. Python standard library only, the installer brings its own Python.
What it does
- Chat. Markdown, code blocks, collapsible reasoning, attachments (text, PDF, images), tools and MCP servers; edit, branch, search and export.
- Live data. Generation speed, draft acceptance rate, GPU use and the FLE expert-cache hit rate while an answer is running.
- Model loader. Shows whether a model fits your machine before you download anything; every file is checked against the SHA-256 from Hugging Face.
- Launch profiles. A sensible start command for your machine, your own profiles with the full command, context size, GPU, threads and cache budget.
- Forge. Build your own quants: take expert layers from quants you already have, prune with REAP, check against the original (KL divergence, perplexity).
- Host and client. Studio on your laptop, computing on the GPU machine – by pairing code or over an SSH tunnel that Studio builds and watches itself.
Get started
$ curl -LsSf https://raw.githubusercontent.com/sirfyyn/fle-studio/main/install.sh | sh
$ fle-studio # öffnet http://127.0.0.1:8642 im Browser