ops

The web interface

OPS chat, Local GPUs and Cluster: what each page shows, and the tools behind them.

The sign-in link opens the DeepLM Ops platform UI.

The pages

OPS chat is the assistant. Each answer lists the read-only tools it used. The right rail (a menu on a narrower screen) shows the model, your chats, and how full the chat's context is.
Local GPUs shows this machine: its GPU, memory and disks, GPU use over the last 15 minutes to 24 hours, and what is using the GPU.
Cluster is listed once a cluster is connected. It shows free GPUs, nodes, partitions and your jobs, and says so on every panel when it is the recorded demo cluster.

Insights, Baseline, Cloud GPUs and the recipes are listed as coming soon.

These pages run the same read-only tools as the chat. They refresh every 10 seconds (this machine) or 60 seconds (the cluster) while they are on screen, and DeepLM Ops shares each reading with every open page for a few seconds.

What the page keeps

The page loads nothing from the internet and keeps nothing in the browser but its session cookie. Chats live in DeepLM Ops' memory, at most 10 per sign-in. Signing out or restarting DeepLM Ops ends them.

The previous, simpler page is still available at /classic/. Use it in one tab at a time.

The tools

The assistant answers from read-only tools. There is no tool that submits, cancels, drains or modifies anything.

Tools	What they read	Needs
`machine_status`, `gpu_usage`, `gpu_processes`, `predict_fit`, `docs_search`	This machine's GPUs, and whether a model fits in GPU memory	Docker. Add `--gpus all` for an NVIDIA GPU. On a Mac, the host probe.
`cluster_status`, `my_jobs`, `job_info`, `job_history`, `limits_and_policy`, `node_gpus`	The Slurm queue, your jobs, limits and node GPUs	A connected cluster, or demo mode

Terminal chat

You can also chat in the terminal. Wait until the status bar says Ready, or the logs show "model ready", before you start.

bash

docker exec -it dlm-ops dlm-ops chat
docker exec dlm-ops dlm-ops ask "how busy is my GPU?"
docker exec dlm-ops dlm-ops doctor

chat is interactive: /new starts over and /quit exits. doctor prints diagnostics without secrets. The terminal chat shares the model the web server already runs, and never loads a second copy.