Share anything. It files itself.
A tweet, a PDF, a forty-minute YouTube video, a photo of a page, a two-hour meeting. Collecta reads the actual thing, writes it up in the format and tone you asked for, and drops it in the bucket where it belongs. You never file anything. You never name anything.
Free to start — ten captures, no card. Runs in any browser today.
One line. Everything goes in it.
There is no "new note" button, no folder picker, no tag field, no template to choose. There is one box, and whatever you put in it gets understood.
- Links
- Articles
- Tweets & X posts
- YouTube
- PDFs
- Screenshots
- Photos
- Voice notes
- Meeting audio
- LLM chat exports
- Plain thoughts
Each type is handled properly rather than generically. A YouTube link is watched — frames and audio, with the transcript pulled separately so nothing rests on the title. A tweet's images and video are downloaded and shown to the model, so a screenshot-only post is still read. A PDF is read as a document. A photo of a label, a whiteboard or a page is read as what it is, not transcribed string by string.
On Android, Collecta sits in the system share sheet — you share from any app straight into it and carry on with your day. On iOS and desktop it installs to the home screen or dock as an app.
It comes back filed, in a bucket you wrote in a sentence.
Minutes later the thing you shared is on your board — titled, broken down, tagged, and sitting under the interest it belongs to.
Phthalate exposure and preterm birth risk
What the study actually measured, what it found, and the two caveats the thread skips over.
Mixture-of-Experts routing, in plain terms
The paper's core claim, the mechanism behind it, and where it disagrees with the two papers already in this bucket.
Pricing tiers, and whether Pro is underpriced
Two speakers, four topics, one decision, three action items — and a note that the pushback on timing never got answered.
The teardown of a $2M agency, minus the padding
Forty-three minutes reduced to the six moves that mattered, with timestamps for the two worth rewatching.
Bucket names are the only coloured text in the whole product. That is deliberate — colour here is information, not decoration, and clay means exactly one thing.
It writes things the way you asked to be told.
Most tools summarise in one house style. Collecta asks you once, at setup, and then every note it ever writes obeys.
Bullets, paragraphs, outline, or Q&A
Pick how a breakdown is laid out. It holds across a tweet, a research paper and a meeting alike.
Summary-first, key points, deep dive, narrative
How far in it goes. Deep dive on a paper you're studying; key points on things you're only tracking.
Neutral, casual, professional, concise, enthusiastic
Set once. There is no chirpy assistant voice underneath waiting to leak through — it writes in the third person about the source, never as you.
Read it in your language
The analysis comes back in the language you chose. Verbatim quotes stay in the language they were spoken or written in — they are evidence, and translating evidence loses it.
There's also a free-text instruction box, and it is honoured literally. "Always flag anything that contradicts something I've already saved." "Explain it like I have a biology degree but no maths." "Never use the word 'delve'."
Made out of sentences, not folder names.
You make a bucket by saying what you care about and why — "I'm interested in pregnancy because my wife is pregnant" — and Collecta writes the bucket from that. The reason is the part that matters: filing is matched against a bucket's description, not its name, which is why an article about design tooling doesn't get shoved into "ai" just because both touch technology.
When something arrives that fits nowhere, it does one of three honest things: files it where it truly belongs, proposes a new bucket if you've clearly started collecting a new interest, or leaves it uncategorised for you. It is deliberately conservative about proposing — a new bucket has to look like somewhere you'll file five more things, not a home for this one item.
Record the room. Get back what was actually said.
Hit record in the browser — no bot joining your call, no calendar integration, no invite to explain to anyone. Up to two hours in a single sitting.
Transcribed first, understood second
The audio is transcribed on its own pass, with speakers separated, in the language actually spoken. Only then does a second pass read that transcript and write the synthesis — and that pass never gets the audio.
This split exists because of a real failure: given audio and a summarising job at once, a model forms a hunch about what kind of thing it's hearing and bends the transcript to fit. A casual conversation in Sindhi came back as a "Sikh prayer ritual." Split in two, the transcript is fixed before anything is allowed to interpret it.
What you get back
An overview, then every topic in the order it came up — each speaker's point kept as their own, with counterarguments preserved rather than merged into "the team discussed X". Then decisions, action items with owners, and open questions nobody answered.
Then the part no meeting-notes tool writes: sentiment and dynamics. Each person's register, cited to the moments that show it. Where the pushback came from and whether it ever resolved. Where the energy in the room turned. Followed by a handful of sharp questions the conversation should leave you with.
The recording itself is deleted after 30 days. The transcript and the write-up are permanent — the expensive part to store is the audio, and the valuable part is the words.
Ask your own library a question and get a cited answer.
"What did I save about sleep and cortisol?" runs a keyword search and a meaning-based search at the same time, fuses the two result lists, then has a model score every candidate for whether it genuinely answers the question — not merely whether it mentions the words. Anything that only glances off the topic is dropped before the answer is written.
The answer arrives with the notes it drew on, so you can go straight to the source. And when nothing in your library actually answers you, it says so and offers the three closest things instead of manufacturing a confident paragraph out of near-misses.
The collection is the point, not the storage.
Insights
A briefing on what you've been filing — the thread you kept pulling on, the bucket that carried the week, a tension between two things you saved. Scannable in thirty seconds, citing your actual notes, no praise and no motivational filler. It runs when you ask for it, not on a schedule.
Research this
Take any note's claims out to the live web and come back with a grounded report attached to it. Or point it at a whole bucket — "what's the current thinking on this?", "what's changed since I started collecting?" — with your existing notes as context.
Talk to a note or a bucket
Argue with a paper. Ask a bucket what it thinks it knows. The conversation is kept with the note, so picking it up in three weeks resumes rather than restarts.
Your own annotations
Add your own thinking on top of any note, with images or files attached. Your words stay visibly yours — never blended into the machine's write-up.
Things it notices connect
Collecta pulls the specific people, papers, companies, places, events and claims out of each note and links notes that share them. Every one has to be backed by a real quote from the source, so connections are evidence, not vibes. Dated things resolve against when you captured them — "next Friday" in a note from June stays in June.
Share a bucket
Publish a note or a whole bucket at a link anyone can read without an account. Or open a bucket for contributions — anything a collaborator shares in lands in your bucket and stays in their own library. Revoke a link and it's dead.
It would rather tell you it couldn't read something.
A second brain that quietly invents things is worse than no second brain, because you'll trust it later and won't remember which part was real.
A URL is not content
If a page is paywalled, login-walled, bot-blocked or rendered entirely in JavaScript, Collecta will not write a summary from the headline and the address bar. You get an honest saved reference with a clean title and a one-line hint: paste the text in and re-run it. Most tools will happily hallucinate a confident three-paragraph summary of a page they never loaded.
Thin is reported as thin
A vague voice memo comes back as a short honest note, not inflated into an insight. Inaudible stretches of a recording are marked as inaudible rather than smoothed over. A rambling call with no agenda is written up as exactly that — "nothing was decided" is a correct result, and one most summarisers are incapable of producing.
Everything in here is yours, and leaves with you.
Any browser today. Native apps in build.
Priced against the two tools it replaces.
A read-later brain and a meeting recorder, in one place, for less than either pair costs. Cancel any time; you keep the plan until the period you paid for ends.
Annual pricing and the exact number behind every line above are on the pricing page.
Stop filing things. Start sharing them.
Ten captures free. Long enough to see whether it puts things where you would have put them yourself.