Llamaforge v0.16: a control panel for upstream llama.cpp with local agents whose quotes are code-verified

Sleepybear2611 · reddit · 2026-10-14

An indie dev released Llamaforge v0.16 — not a llama.cpp fork but a browser control panel that drives real upstream llama.cpp, so new models are usable the same day via the panel.

Key features:

Pre-load VRAM preview: using actual model sizes measured on your machine, it shows where each model will land on each GPU, to scale, and warns if it won't fit — no more OOM crashes. The main model goes on the fastest GPU that fits; worker models stay off it.
Run several models at once.
Embers, local agents: watch topics you pick, read your sources overnight, build their own wiki, and write a morning brief. The core rule: every claim must quote its source word for word, and plain code checks the quote against a saved snapshot — if it's not there, the claim is dropped. The model proposes, the code decides. No browser, no tools, nothing leaves the machine.
Smaller updates: real loading progress; a thinking switch (auto/on/off plus token budget) that reads the chat template and tells you if off won't actually stick; vision can be disabled to free the projector's VRAM without deleting files; a voice tab (the demo voiceover is qwen3-tts running locally).

Stack: Python stdlib backend, vanilla JS frontend, no build step. One-line installers for Windows/Linux/macOS; author runs Windows + NVIDIA 5080 and 5060 Ti.

Original post →

Llamaforge v0.16: a control panel for upstream llama.cpp with local agents whose quotes are code-verified

More from coding & agent