Introduction

Fewshell is a mobile AI-assistant for MLOps, DevOps and Sysadmins. Connect to your infrastructure securely from any device and safely troubleshoot issues on the go, or just run things.

What is Fewshell?

Fewshell is a human-gated, collaborative AI terminal agent. It runs on your server and syncs state directly across your desktop and mobile clients — no cloud service required. Every command is staged for your review. Nothing executes until you say so.

Modern LLMs excel at terminal tasks and keep getting better — but letting an autonomous agent run shell commands against production infrastructure is too risky. Fewshell gives you the AI leverage while keeping a human in the loop.

Key Features

• Human-gated — AI assists, you approve. No command runs without explicit confirmation. There is no autonomous mode — it can't be accidentally enabled or misconfigured.
• Self-hosted — The agent runs on your server. Clients connect directly over SSH tunnel.
• Real-time sync — Snippets, prompts, and settings sync across desktop and mobile with no cloud service needed.
• Collaborative — Multiple engineers can join a session for real-time incident response.
• Multi-project / multi-session — Organize infrastructure into projects, each with multiple concurrent sessions and settings.
• Shell snippets — Quick-access command templates for common operations. Snippets also give the AI context about your infrastructure.
• Custom system prompts — Tailor the AI behavior per project to match your infrastructure and workflows.
• Secrets management — Secrets are stored in the client device keychain and automatically redacted from history. Never stored on disk or sent to LLM.
• Session history & export — Full audit trail of every session. Export for postmortems, compliance, or experiment tracking.
• BYOM — Bring your own model, or use any OpenAI/Claude-compatible endpoint. Self-host or connect to your preferred provider.

Use Cases

• Incident response — DevOps and sysadmins investigate logs, triage alerts, and perform root cause analysis from any device.
• ML labs — Monitor, restart, or adjust training runs and rollouts. Troubleshoot GPU nodes and pipelines via shell commands.
• Infrastructure health — Check system state, diagnose issues, and tune performance across your fleet.
• Serverless & cloud CLI — Works with serverless and cloud infrastructure through AWS, gcloud, Azure, or any CLI — for when you need hands-on access beyond what IaC covers.*

Note: For serverless, you need to run one micro instance that hosts the agent and has the CLI installed.

Next Steps

Ready to get started? Head over to the Quick Start guide.