AB
Ash Beacon
20 hours ago
deepseek harness desktop ate 14GB then froze my M2
Deepseek Harness Desktop
Electron desktop client for DeepSeek Harness workflows
tried the Electron client for DeepSeek Harness over the weekend. pulled Qwen2.5-Coder 14B through it instead of raw ollama CLI.
first run: fine for ~20 min. then the window locked, Activity Monitor showed the renderer at 14.2GB, fans full blast. no crash dialog, just a dead UI. had to force quit.
curious if anyone else is using this on Mac or if it's mostly a Linux/Windows thing. the harness workflows look useful; the desktop shell feels half-baked.
2 comments
Join the discussion
Log in to comment.
same class of bug on my Linux box last week — Chromium renderer leaks if the harness keeps streaming tokens into a big chat buffer. I ended up running the CLI and pointing the UI at localhost only for config.
Electron + local LLM is cursed until they cap the message history. did you notice if RAM dropped after closing the chat pane, or was it stuck until kill?
mac here too. i bounced after two freezes and just stuck with ollama serve + a thin openwebui tab. less pretty, but it doesnt hold my machine hostage when a 14B loop goes sideways.
the harness recipes themselves are fine — its the desktop shell. weird that a "$0 electron wrapper" is the part that burns more RAM than the model.