FAQ
Straight answers about how Sage works, what it costs, what happens to your code, and what to do when something breaks.
What is Sage AI?
Sage is a local-first AI coding assistant. It is a command line tool you install with pip, a web app, iOS and Android apps, and an SMS bridge that lets you fire tasks at your own computers from a text message. It runs open-weight models on your hardware by default, so your code stays where it is.
How do I install it?
Run pip install sage-ai-cli, then sage login to connect the CLI to your account. Launch sage from the root of the project you want to work on. The full walkthrough, including every slash command, is on our Documentation page.
What does "local-first" actually mean?
It means the default path involves no network at all. When you run a local model, your prompts, your source code, and your files are processed entirely on your own machine. We do not receive them, we could not read them if we wanted to, and there is no background telemetry sending your prompt contents anywhere.
What happens if I use a cloud model?
A cloud request has to leave your machine to be answered — there is no way around that. Before it does, Sage strips identifying fields such as your user ID, email, and session identifiers, and scrubs email addresses and phone numbers out of the prompt body. The upstream provider gets an anonymized request that is not linked to you. Choosing a cloud model is always an explicit action on your part.
Which models can I use?
More than 200 open-weight models, including the Llama, Gemma, Qwen, Mistral, Phi, DeepSeek-Coder, and Devstral families. You can pull them as local GGUF weights with sage pull, or route to a cloud-hosted copy. Use /models inside a session to see what is available to you, and /model to switch without losing your conversation.
Do I need a GPU?
No, but it helps. Smaller open-weight models run acceptably on a modern CPU with enough RAM; larger ones want a GPU with enough VRAM to hold the weights. If your machine cannot comfortably run a model locally, a cloud model on a paid plan gives you the same interface without the hardware requirement.
What do the plans cost?
•
Free — $0/month. Browser-only AI with 7 in-browser models.
•
Starter — $19/month. 300,000 output tokens per billing period.
•
Pro — $49/month. 2,000,000 output tokens per billing period.
•
Premium — $99/month. 10,000,000 output tokens per billing period.
Why are you billing by output tokens?
Because output is what actually costs money to generate. Input tokens are not billed against your allowance, which means pasting a large file, a long stack trace, or an entire specification does not eat into what you paid for. You are charged for what the model writes, not for what you show it.
What if I go over my allowance?
Usage beyond your plan's allowance is billed at the published per-1,000,000-output-token overage rate for your plan. Your consumption is visible live on the Usage page and from /status inside a session, so it should never come as a surprise.
How do I pay?
PayPal, Apple Pay, or Google Pay. Those providers handle the payment credentials — Sage never stores your card or bank details.
Is the Free plan really free?
Yes. $0, no card required, and it does not expire. It is browser-only, with 7 in-browser models and no output token allowance for cloud models, so it is best treated as a way to evaluate Sage before you decide whether a paid tier is worth it.
Can I cancel whenever I want?
Yes, from the Billing page. Cancelling stops the next renewal and leaves your current period intact until it ends, after which the account reverts to Free. Refund terms are on our Refund Policy page.
Do you train on my code?
No. We do not use your code or your conversations to train models, we do not sell your data, and we do not run advertising or behavioural tracking. For local models the question is moot — we never see the data in the first place.
Can I delete everything?
Yes, from the Account page, without contacting anyone. You can export a JSON archive of everything we hold, delete your conversation history, wipe your SMS bridge devices and allowlist, or delete the account outright. Account deletion is immediate and permanent — no shadow copies, no 30-day window during which your data still exists.
Which editors does it work with?
VS Code, Cursor, JetBrains IDEs, Neovim, Sublime Text, and Zed, plus GitHub CLI and Git Hooks for version control workflows, and Godot 4 and the Unity Editor for game development.
How does the SMS bridge work?
You register a computer on the Devices page and allowlist the phone numbers or email addresses permitted to reach it. From iMessage, Google Messages, or email you name the device and describe the task; Sage runs it on that machine and replies over the same channel. Only allowlisted senders are accepted, and you can revoke a device or a sender instantly.
Is it safe to let Sage run commands?
Sage can read files, write files, and execute shell commands you approve, which is exactly what makes it useful and exactly what deserves care. Review what it proposes, work inside version control, and do not give it privileges it does not need. The /undo command reverses its most recent file changes if an edit goes wrong.
My session keeps losing track of things
Long conversations eventually crowd out the earliest context. Run /compact to summarize the session and discard the raw history, /files to check what is actually loaded, and /read to pull a specific file back in. If Sage is reasoning about a stale version of a file, /read is usually the fix.
Something is broken. What do I do?
Run sage update to make sure you are on the current release, then /help to confirm the command you are using exists in your build. If it still misbehaves, email support@sageworksai.com with the output of sage --version, the command you ran, and what happened instead of what you expected.
My question is not here
Email support@sageworksai.com. Questions that come up repeatedly get added to this page.