Launching today
Most AI tools make you explain the context before they can help. Goldfish already has it. It privately remembers what you’ve been working on across your Mac, then helps you write better from any app. Press Option in a text field to draft replies, summarize threads, rewrite sentences, or recall important details from your recent work without copying, pasting, or re-explaining the whole backstory.
Wow, this hits right where it hurts. 🐠
The "goldfish memory" of current AI tools is easily the most frustrating part of my daily workflow. I waste so much time copy-pasting Slack threads or brief docs into Claude just to get a relevant reply. Having that context accessible natively with just the ⌥ Option key sounds like a massive productivity unlock.
Love the local database approach for privacy too, that's usually the biggest blocker for these kinds of tools.
Congrats on the launch Joel!
Quick question: how does Goldfish handle context switching? If I move from a client email to a technical dev task, does it easily separate the two "memories" when I press Option?
@keirodev Hey Kévin, great question!
Goldfish separates context by what you are actually doing when you press Option. It reads the focused field, the surrounding window, and relevant recent activity, then pulls in the memory that matches that situation.
So if you move from a client email to a technical dev task, it treats those as different contexts and avoid blending them. The goal is exactly to stop you from having to re-explain which thread, doc, or task you are in every time.
And yes, the local database part is a big piece of making that safe and usable.
Andsend
Congrats on the launch! I've tried the beta and really like the UI and ambition. I'm curious: what's your own/the team's favorite use cases in everyday work?
@per_clingweld Thank you Per for being an early supporter and for the comment! Mostly messaging for me, but I also use it to find old tweets. Weirdly, when I can’t remember a song, I sometimes use transcription in Spotify to see if it can match it. Works 50% of the time haha
the 'replies like you' part is what i'm most curious about. how do you handle voice consistency across contexts where someone writes very differently (boss vs partner vs friend)? is the model learning per recipient or one general voice trained on all past replies?
@thenameisarian love this question. it’s per-context, not one blob of “your voice” averaged over everything.
Goldfish looks at the surface you’re writing in, the current thread, who you’re talking to, and your past examples that are relevant to that kind of interaction. So a Slack reply to a teammate, a LinkedIn DM, and an email to a customer can all come out differently.
We’re trying pretty hard to avoid the uncanny “same tone everywhere” thing. It should feel like you, in that specific situation.
Congrats on the launch! I have a small question, with Goldfish capturing context across so many apps, how does it filter out irrelevant noise when you're rapidly switching between unrelated projects throughout the day?
@crystalmei Great question. Goldfish uses the current app, focused field, surrounding UI, and recent activity as the strongest signals, so it does not just dump one giant memory into the prompt.
When you press Option, it ranks the context by relevance to what you are doing right now, then pulls in only the pieces that match. If you have been jumping between unrelated projects, the current window and field usually anchor it pretty well, and the noisy stuff gets ignored. It is also designed to be explicit about uncertainty rather than blending contexts together.
This is great! How big can the context get?
I also saw you using graphs, I guess it is to increase accuracy. So how is the memory layer structured?
@vugar_javadov the context can get pretty large, but we keep it scoped rather than just dumping everything in.
The rough structure is:
1. Immediate screen/app context from the current field
2. Recent activity from the last few minutes/hours
3. Longer-term memory summaries and searchable snapshots
4. A small identity/voice layer for how you usually write
The graph is mostly there to connect people, projects, apps, threads, docs, etc. so the memory retrieval is more accurate than plain keyword search. When you press Option, Goldfish pulls the relevant slice for that exact surface and intent, then ignores the rest.
Hey Ben, Goldfish looks like a genuinely clever take on ambient AI. How does it handle context switches between work projects or clients that have very different tones?
@mbertone911 Thanks Marco! It uses the focused field and nearby context as the anchor first, then pulls in only the relevant recent memory for that surface. So a client email, a dev task, and a Slack reply each get treated as separate writing situations, with different tone and context.
Have you tried it? What do you think? 🐠
The challenge with AI memory is often trust and relevance. How do you prevent important details from getting buried as more conversations accumulate?
@harini_mukesh This is basically the core design problem for Goldfish.
We don’t treat memory as one giant chat history. It’s split into layers: the immediate screen context, recent activity, longer-term summaries, and entity/project-level memory. When you press Option, Goldfish first anchors on what you’re doing right now, then pulls in only the memory that looks relevant to that surface.
Important details should also become easier to retrieve over time, not harder. So instead of relying on raw conversation logs forever, Goldfish distills repeated people, projects, preferences, and commitments into more structured memory.
And there are controls for excluding things you don’t want remembered, because trust only works if the user can shape the memory.
@kar_re This makes a lot of sense. The layered memory approach feels much more practical than treating everything as one long chat history. User controls for memory seem like a must-have for trust.