What website is this?
Goldfish is a local AI writing assistant for Mac: it reads text from your current app via macOS Accessibility APIs and builds searchable work memory on your device. Press ⌥ Option in any text field to reply, continue, or summarize from on-screen context while gradually aligning to your tone and recurring workflows. Unlike general chat tools that expect you to paste background each time, it emphasizes on-device memory plus invoke-anywhere access; memory stays in local SQLite by default, and only the snippets needed for an active request are sent to Azure OpenAI.
Key Features
- Press ⌥ Option in any text field to generate replies or continuations from current screen context
- Store cross-app work memory in a local SQLite database with search and browse support
- Learn tone, audience, and scenario differences across apps for output closer to your voice
- Detect repetitive workflows and surface reminders for follow-ups that may slip
- Connect Mac context to Claude Desktop through local MCP
- Start full-context conversations from the Dynamic Island/menu bar “aquarium” interface
Use Cases
- Product or ops staff switching between Slack and email who do not want to copy project background into every thread reply.
- Freelance writers drafting in Notion and researching in the browser who use Option to continue with prior context intact.
- Founders handling follow-ups across WhatsApp and email who want consistent tone and fewer forgotten to-dos.
- Claude Desktop users who want on-screen Mac context brought into their existing writing flow via MCP.
Who is it for?
- Knowledge workers who switch among many Mac apps and frequently reply or continue text
- Users who want work memory kept locally rather than fully resident in the cloud
- People already on Claude Desktop who want device context in the same writing environment
- May not fit: users who cannot or will not grant macOS Accessibility permission
- May not fit: people mainly on Windows/Linux or workflows that rarely involve text fields
- May not fit: organizations that need team-wide shared memory or multi-user orchestration
How It Compares to Similar Tools?
Tools like ChatGPT usually rely on manually supplied context; Grammarly focuses more on grammar and style and generally does not build persistent cross-app local memory. Goldfish bundles screen context, local memory, and a shortcut invoked anywhere—suited to users tired of re-explaining background. If you prioritize open self-hosting or fully offline models, this hybrid “local memory + cloud inference” path may not fit; if you only polish occasional single paragraphs, a lighter grammar tool is often simpler.
What Our Customers Say
Maya (Knowledge Worker)
When handling project threads across Slack and email, they first ask whether copy-paste can be reduced and replies still sound like them; after granting Accessibility access, they also check whether memory scope and privacy exclusions are granular enough.
Daniel (Freelance Writer)
Switching between Notion and the browser while drafting, they want Option invocations to continue with prior text; if tone learning is still uneven, they often edit a few lines before sending.
Rina (Claude Desktop User)
Already using Claude for thinking and writing, they watch whether MCP truly brings Mac context over; they also cross-check privacy policy against claims of sending only needed snippets per request with zero retention before deciding scope.
FAQs
Q: Does Goldfish require installation? Which systems are supported?
A: A client download is required (the site lists macOS and Windows); core capabilities center on Mac local memory and Accessibility screen reading—check current release notes for actual support.
Q: Is my data uploaded to the cloud?
A: Work memory defaults to encrypted SQLite on your device; only the context needed for an active writing or Q&A request is sent to Azure OpenAI under a zero-retention configuration.
Q: How is it materially different from ChatGPT or Grammarly?
A: The difference is persistent local cross-app memory plus shortcut-based generation from current screen context in any text field, rather than mainly manual paste or single-paragraph grammar fixes.
Q: What does it cost today?
A: The site FAQ states it is free during beta; future pricing depends on official updates—do not assume it stays free long term.
