Shellway connects a state-of-the-art AI to your shell. Ask it to run a command, fix a script, diagnose a crash, or manage a remote server — it reads the context and acts.
JWT_SECRET to AUTH_SECRET but your production .env wasn't updated. Run:
Shellway wraps your shell session with a context-aware AI that sees what you see — your working directory, running processes, recent output.
Type a question or instruction the way you'd say it to a teammate. No special syntax — just describe what you want done.
Shellway sends the current directory, recent output, and environment metadata. The AI forms a plan grounded in your actual state.
Every proposed command is shown before it runs. Approve with one key, edit it, or say "go ahead" to execute the whole plan.
Shellway relays context to the AI and relays commands back — it never stores shell output, never touches your credentials, and never runs anything without your explicit approval.
Every shell command proposed by the AI requires your approval before execution. You can preview, edit, or reject any step.
Shell output sent as context is used for the current inference and discarded. We don't train on your command history.
Shellway reads only what you explicitly share per session. Env files and ~/.ssh are never scanned automatically.
Every AI-proposed command and its status (approved / edited / rejected) is logged locally and optionally to your own storage.
Shellway handles the kinds of tasks that slow you down: debugging, scripting, remote ops, and anything in between.
Paste or pipe an error. Shellway reads your stack trace, log tail, and recent git history, then gives you a focused explanation and a fix — not a generic Stack Overflow answer.
Describe what a script should do in plain language. Shellway writes it, shows it to you, and runs it only after you confirm. It fixes errors on the next turn if anything fails.
SSH into a machine and use Shellway there too. Check disk usage, restart services, tail logs, or coordinate changes across a cluster — with the same approval flow.
Highlight a block of bash, a config file, or a pipeline. Ask "what does this do?" and get an explanation anchored to your actual codebase, not a generic example.
Ask Shellway to "set up a staging environment" or "migrate the database and verify row counts." It breaks the task into steps, shows the plan, and executes them in order.
Git, Docker, kubectl, npm, pip, cargo — Shellway calls them exactly as you would. It uses your local config files, respects your aliases, and stays inside your environment.
Every plan includes the core AI shell integration. Upgrade for larger context windows, audit logs, and self-hosted relay.
All prices in USD. INR pricing billed via Razorpay. Annual plans save 20%.
Yes — every command the AI proposes requires explicit approval before execution. The AI cannot run anything without you pressing Enter (or typing "yes"). You can also restrict Shellway to read-only mode in your config, which prevents any write commands from even being proposed.
Only if you explicitly share them. By default, Shellway reads shell output from your current session — it does not scan .env files, ~/.ssh, or system credential stores. You can configure exactly what context is sent in your ~/.shellway/config.toml.
Shellway currently routes to Claude (Anthropic) on all plans. Enterprise customers can bring their own API key and route to any compatible model, including models deployed in their own VPC.
Shellway captures recent shell output, the current working directory, and metadata about running processes. The amount of history included depends on your plan's context limit (8k on Free, 32k on Pro). You can explicitly include a file or log snippet with the @include command.
The inference step requires a network call to the AI provider (or your own model endpoint on Enterprise). The CLI itself works offline for non-AI tasks. Air-gapped deployments are available with a self-hosted relay and an on-premises model — talk to us about Enterprise.