No more context-switching between SSH terminals, monitoring dashboards, and access panels. ScaleServe consolidates every fleet operation into a single interface.
⚡
Remote Command Execution
Execute any shell command on any Tailscale peer. stdout and stderr stream live into the UI as the process runs — no waiting for completion. Includes built-in presets for GPU inventory, Ollama health checks, PyTorch CUDA verification, and vLLM status. Auto-detects SSH users across common Linux distros and AWS/OCI instances.
ssh subprocesslive stdoutabortpresetsuser autodetect
📡
Script Broadcast
Stream any local script into a remote machine's stdin over SSH. No file upload, zero disk footprint. Supports Python, Bash, Node.js, PowerShell, Ruby, and Perl. Interactive Python mode lets you send runtime input after the script starts running.
stdin pipepython3 -interactive
🔀
Parallel Multi-Device Runner
Run scripts across N machines simultaneously. Per-device output panels update live. Stop any individual job or kill all with one button. Mix different scripts for different machines in the same batch run.
Future.waitper-job stopN×M parallel
🗂️
Distributed Compute Engine
Split an input file across N nodes proportionally by RAM + GPU capacity using a weighted scoring algorithm. Each node processes its chunk in parallel. Merge outputs with concat or a custom bash command. Supports CSV, JSONL, binary, and any file format.
weighted splitSCP chunksmerge
🧠
Distributed LLM Inference
Distribute prompt batches across GPU nodes running Ollama. Three modes: Batch splits prompts proportionally by GPU throughput. Consensus sends all prompts to every node for multi-perspective answers. Ensemble round-robins for load balancing. Results stream in real time with per-node prompts/sec metrics and ETA. Automatically checks VRAM capacity before assigning work and unloads the model after the job to free GPU memory.
Ollama APIbatch · consensus · ensembleVRAM checkreal-time PPSJSONL export
📊
Fleet Resource Monitoring
Poll CPU, RAM, and GPU metrics from fleet machines every 5 seconds via SSH. Platform-aware: uses nvidia-smi, rocm-smi, or Apple Silicon APIs.
5s pollGPU detectNVIDIA/AMD/M-series
🔑
SSH Key Management
Generate RSA keypairs, auto-install public keys on remote machines via a one-time bootstrap key, and sync the team keypair through the backend. SSH Doctor runs diagnostics on any node to identify and explain connectivity issues.
RSA keygenteam syncSSH Doctor
👥
Team Access Control
Workspace-scoped membership with owner, admin, and member roles. Granular permission flags control exactly what each member can do. New members join via OTP invite — no shared credentials ever. Full audit trail of every remote execution.
OTP invitepermission flagsaudit logs
🖥️
Device Fleet Browser
Searchable, real-time inventory of all Tailscale-connected peers. Filter by name, IP, OS, or online status. Devices with SSH profiles link directly into Remote Runner with the machine pre-selected.
tailscale statussearch & filterOS icons