Bring your plan, not just your keys

Update, 24 September 2026: agent execution — chat, actions, drafts — moved from Pro to Standard, so everything below about running on the plan you already have now applies to both plans. Pro adds the Memory API and real sending through Composio.

Until now, using the execution layer meant pasting an API key. That is a fine ask for engineers and a strange one for everybody else — a second billing relationship, a console to visit, a spending limit to guess at, all to use a model you are already paying a monthly subscription for.

So we changed the default. Pro now runs on the assistant plan you already have. An API key still works, and for some people it is the better option. It is no longer the price of entry.

How it works

If you already use one of the major assistant subscriptions and have its official command-line tool installed and signed in, ShogunAI can run agent work through it. You opt in explicitly, once, in settings. From then on drafting, reasoning, and tool use happen inside the plan you are already paying for — no separate API bill to set up. Plans include a finite allowance for this kind of use; if you ever reach it, ShogunAI says whose limit it is and offers the key path so you can keep working.

Two things we do not do, and will not:

• We do not read other applications' credential stores. No rummaging through another app's keychain entries or config files. ShogunAI launches the vendor's own tool, the way you would in a terminal, and that tool handles its own authentication.
• We do not reimplement anyone's consumer sign-in. Every request runs through the vendor's official client, under their terms, with their rate limits. That is the only version of this that is safe for your account.

If the tool is not installed or not signed in, ShogunAI says so and offers the key path instead — it never switches your billing silently.

Why the two lanes stay separate

There is a second kind of model work in ShogunAI that never touches your credentials: indexing, classification, the nightly review, and the morning brief. That work is batched, runs on our infrastructure, and is covered by your subscription. It is the reason Standard runs without a key.

Keeping the lanes separate is not bookkeeping. Batch work is high-volume by nature, and pointing it at a personal plan's allowance would burn through it in short order — taking your own assistant down with it. So the split is enforced: our capacity does the bulk work, your credentials do the work you asked for by name.

What this changes about the price

Pro is $99 a month billed annually. Before this change, the honest sticker price included whatever your API usage came to on top — which for heavy users was not a rounding error. Now, for most people, it is the sticker price, and the model spend is one you had already decided to make.

Standard is $49: capture, recall, search, the notch, connected tools for reading context, the nightly review, and the morning brief — none of which ask you for a key. Meeting transcription is the exception. It runs on a speech provider key you add yourself, on every plan, and the app says so before it records anything. Both plans start with a seven-day full trial.

The point underneath

This is the same argument as local-first, one layer up. Your memory should not be locked to a vendor, and neither should your inference. Bring the plan you already have, swap it later when a better one exists, and keep every day of context you have accumulated in the meantime.