The AI briefing layer
Brief AI properly. Once.
This is the page for how it actually works. If you want the short version, the overview is on the home page. Below: the pipeline, the brief, and exactly what does and does not leave your Mac.
Where it fits
The step between scattered context and the AI tool.
A chat window is built for long tasks and heavy for tiny ones. Shikomi is built for the thirty small transforms you do every day — without opening anything.
A clipboard manager remembers what you copied. Shikomi prepares it into a brief the next tool can act on. The brief is the product; the archive is a side effect.
How it works
Six steps, copy to handoff.
- 01
Capture, on your terms
Shikomi keeps your last twenty copies in Recents and lets you paste any of them with ⌘1 through ⌘6 — no digging. Copied an image? macOS's built-in Vision text recognition turns it into text you can use. Passwords, 2FA codes, and API tokens are detected and dropped before a clip is ever stored.
- 02
Select the fragments
Pick several clips at once — the thread, the figures, the stray note. Shikomi reconciles them into one coherent context instead of gluing them end to end: overlap is merged, the order is made sensible, and the noise is left out.
- 03
Shape the outcome
Before anything generates, you say what you actually want — the goal, what “good” looks like, and the limits. Shikomi frames the problem from your fragments so the brief is about the result you need, not the raw mess you started with.
- 04
Generate the brief
Shikomi writes a clean, five-section Markdown brief. The transform runs on the cloud key you bring and goes straight to the provider you chose — nothing relays through us.
- 05
Review in Manaita
The brief opens in Manaita, your review surface, before it goes anywhere. Read it, edit it, pin it, or stack it with related clips. Nothing leaves until you say so.
- 06
Hand off to any tool
Copy the Markdown and paste it into ChatGPT, Claude, Codex, Claude Code, Cursor, or Gemini. Same packet, same shape, whichever tool you reach for next.
The brief
Five sections. Always the same five.
A brief is plain Markdown with a fixed shape, so any model reads it the same way. Each section does one job. Skip one and the AI fills the gap with a guess.
- Goal
- The single outcome you want, in one sentence — the result, not the topic. “A reply that asks for one number without sounding defensive,” not “email about the deck.”
- Context
- Only the fragments that matter: the thread, the figures, the note. The AI stops guessing because the relevant detail is already on the table.
- Success criteria
- How you will know the output is right — length, tone, what to lead with, what to avoid. The part people skip, then re-prompt three times to recover.
- Constraints
- The hard limits: match the prior tone, no new commitments, stay inside the budget. The lines the AI must not cross.
- Output
- The exact shape you want back — a draft reply, a bullet list, a table. So you get something usable, not something you have to reshape.
## Goal A one-page decision memo the team can approve in tomorrow's standup. ## Context - Five Slack threads on the migration - The cost estimate (eng, 14:02) - Two objections from design review ## Success Criteria - One page, scannable headings - States the recommendation first - Names the open risk, not just upside ## Constraints - No new scope - Decision framed as reversible ## Output A Markdown memo: recommendation, rationale, risks, next step.
One worked example — five fragments reconciled into a brief any tool can act on. The home page shows a different one.
Manaita
Keep what you figured out, not just what you copied.
Most clipboard tools remember what you copied. Manaita remembers what you worked out. When you piece together a proposal from five threads or a fix from scattered notes, open Manaita and keep the method — pin it, stack the related clips, name it.
The next time a similar problem lands, you already have the shape. Briefs become a reusable playbook that lives in Shikomi, not inside any single AI tool. That is the part that compounds: one solved problem becomes a dozen you do not have to solve from scratch.
Trust & boundaries
Cloud-only, on your key — and nothing more than that.
Your key, your provider, direct
Every transform runs on the cloud key you bring — OpenAI, Anthropic, OpenRouter, or a local model you point to. The request goes straight to that provider. No servers in the middle, no relay, no telemetry endpoint. You pay your provider; we never touch your credits.
It sees only what you copy
No Accessibility permissions, no Screen Recording, no window scraping, no ambient monitoring. Shikomi reads the clipboard when you act on it — nothing else.
Secrets are filtered first
Passwords, 2FA codes, and API tokens are detected and dropped before a clip is stored or a transform runs. They never reach Manaita and never reach a provider.
The backend is disclosed
Before anything is sent, Shikomi tells you where it is going and on whose key. No quiet routing, no surprise destinations.
Requirements
Built for Apple Silicon.
- Apple Silicon Mac (M1 or newer)
- macOS 26 or later
- A cloud API key you bring — OpenAI, Anthropic, OpenRouter, or a local model
- Intel Macs and earlier macOS are not supported
Questions
The details people ask about.
Which Macs can run Shikomi?+
Apple Silicon Macs (M1 and newer) running macOS 26 or later. Intel Macs and earlier macOS versions are not supported.
Does my clipboard get sent to a server?+
No. Transforms use your own API key and the request goes directly to your chosen provider. We run no servers in between — no relay, no telemetry endpoint.
Do I need my own API key?+
Yes. Shikomi is bring-your-own-key: point it at OpenAI, Anthropic, OpenRouter, or a local model you run. You pay your provider directly — we never touch your credits.
How is this different from ChatGPT or Claude Desktop?+
Chat tools are great for long tasks and heavy for tiny ones. Shikomi is built for the thirty small transforms you do every day — inside the app you are already in, with no chat window to manage.
Is Shikomi a clipboard manager like Paste?+
No. Paste remembers what you copied; Shikomi prepares it into a brief the next tool can act on. Manaita lets you review and reuse, but the brief is the product, not the archive.
Will Shikomi watch my screen?+
No. Shikomi only sees what you copy. No Accessibility permissions, no window scraping, no ambient monitoring. Passwords, 2FA codes, and tokens are filtered out before any transform runs.
When does it launch?+
We are in pre-launch. Join the waitlist and you will be among the first to receive an early download link.
Better AI answers start before you ask.
Shikomi is in private beta. Join the waitlist for an early download link.