ShogunAI

Execution layer

Answering is the easy half

Drafts, updates and actions inside the tools you already use, under three named levels of autonomy — and a stop before anything reaches another person.

Most memory products answer well and leave you to do the work: copy the draft, open the mail client, fix the recipient, hunt for the file, attach it, send it. The execution layer is the part that closes that gap, and the rules around it are visible before anything runs.

Press Option, and it writes at your caret

Inline composition reads the field around your cursor and the memory behind it, then writes the continuation straight into the app you are already typing in. It is a device-local write: nothing is sent, and you send it yourself.

Three levels, and the line does not move

Level one is reversible and local, and just runs. Level two is drafted and waits for you. Level three is anything that reaches another person — sending mail, posting a message, putting an event on someone’s calendar — and it always stops for approval. No prompt moves an action between levels.

Your plan or your key

Execution runs on the assistant subscription you already pay for, inside that plan’s limits, or on an API key you bring. You pick the provider and can switch without losing a day of memory. Keys live in the system keychain and nowhere else.

A record of what it did

Every action leaves what ran, on what evidence, and what left the device — marked when it passed through a third party. That log is what makes level one acceptable: automation you can audit afterwards is a different proposition from automation you have to trust in advance.

From context to a useful next step

Most memory products answer well and leave you to do the work: copy the draft, open the mail client, fix the recipient, hunt for the file, attach it, send it. The execution layer is the part that closes that gap, and the rules around it are visible before anything runs.

  1. Understand

    The request is read against the state of your work — who is involved, what was decided, what is still open — rather than the thread alone.

  2. Prepare

    The draft, update or tool action is assembled with the right file already attached and the open question already answered.

  3. Approve

    Anything consequential waits on one approval. Reversible work has already finished by the time you look.

What this changes

Spend less time rebuilding context

  • Send the follow-up with the right version attached
  • Walk into the 3pm already prepped
  • Close the loop you forgot was open
  • Keep the send button in your hands

Frequently asked questions

Clear answers before you start

Can it send something without asking?

No. Anything that reaches another person is level three and stops for approval. That is a property of how actions are routed, not a setting you have to get right.

Does driving it over the API skip the gate?

No. The same classifier and the same gates apply over MCP, CLI and REST. An agent calling in has no more authority than your own click.

Do I need an API key?

No. It can run on the assistant plan you already pay for. Bringing your own key is the alternative, not the requirement.

Which plan includes it?

Pro, along with the Memory API and the second-layer connections. Standard covers capture, recall and everyday execution.

Explore all Features

See how ShogunAI connects private memory, recall, and execution.

View overview
ShogunAI Command Layer

文脈はあちこちにある。AIを毎回ゼロから始めさせない。

一度つなげば、記憶は Mac の中にひとつ。そこから先は ShogunAI が仕事を終わらせます。

アーリーアクセスへの登録に費用は一切かかりません。

macOS ・ 7日間フルトライアル ・ 記憶は端末の中に。