Comparison · 5 min read

Petals after Rewind and Limitless

Your ambient memory tool is gone. What happened, what it got right, and what comes next.

2026-06-11 · Last reviewed 2026-06-11 — products change; we re-check these claims.

What Rewind and Limitless got right

Passive ambient capture was real magic. Not a metaphor — actual magic. Rewind sat on your Mac and recorded everything you read, wrote, and looked at, compressing it locally so you could search your own life like a filesystem. No friction, no discipline required. If your work lived in a browser and a text editor, you had a searchable archive of your entire working day without doing anything.

Limitless built on that and went further: a small aluminum pendant you clipped to your shirt captured the conversations that never made it to the screen — the hallway aside, the dinner table idea, the meeting where no one took notes. It was the first serious attempt to capture the ambient layer of a life, not just the digital one.

That combination — screen memory plus in-person audio — pointed at something worth wanting: an AI that knew what you actually did, not just what you typed into it.

What happened

In December 2025, Meta acquired Limitless. The Limitless team joined Meta's Reality Labs group to work on AI-enabled glasses. The standalone product did not survive the acquisition.

The timeline matters if your data is in the system:

  • Pendant sales ended December 5, 2025.
  • The Rewind Mac app stopped capturing on December 19, 2025.
  • EU, UK, Brazil, and several other countries lost access immediately — hard cutoff, no grace period.
  • Existing Pendant owners have cloud access for roughly one more year (through approximately end of 2026), after which the device has no backend to talk to.
  • Limitless did ship a data-export tool before shutdown, so transcripts can be retrieved — but the window is finite.

New users cannot sign up for anything meaningful. The product is wound down.

The lesson, stated plainly

Rewind launched with a strong local-first promise: your data stays on your machine. That was the privacy story and the product differentiator. Two years later, the architecture had shifted to a "Confidential Cloud" model. Another year later, Meta owned the cloud.

The local-first promise broke retroactively — not because the company was dishonest about the shift, but because the data was never structurally the user's. When it lived on servers someone else operated, the trust relationship was always contingent on who owned those servers next.

This is an architecture problem, not a character problem. The lesson is not that the founders were untrustworthy; the lesson is that data ownership is an architecture choice, and that choice has to be made at the foundation, not the marketing layer.

Petals is a cloud product — that is not hidden. Content passes through servers to power the memory AI. What Petals does differently is make the data structure fully visible and user-editable: the graph is yours to browse, correct, and export. The memory is not a hidden profile somewhere; it is a legible structure you can inspect claim by claim. That is a different kind of ownership than local-first, but it is ownership with teeth rather than ownership on trust.

Where Rewind and Limitless were genuinely better

This section exists because honesty requires it.

Petals does not do passive ambient capture. It never has. You have to actively feed it — paste a note, upload a document, link a file, run a conversation. Nothing happens in the background without your participation.

That means Petals cannot replace what Rewind did for you. If your workflow depended on continuous screen OCR surfacing what you read six months ago, or on the Pendant capturing what was said in a meeting room before anyone opened their laptop — those things are not in Petals, and there is no equivalent today.

Rewind and Limitless were also stronger on meeting intelligence: native summarization across Zoom, Google Meet, and similar services, automated and polished. Petals can work with meeting transcripts you bring in, but it does not sit in the call with you.

The closest path forward for ambient capture

The only serious open-source heir to Rewind is Screenpipe. It runs locally, captures your screen and audio continuously, and stores everything on your own machine. It is technical — you configure it, you run it, you maintain it — but it is real and it is yours in the structural sense that Rewind's original model was.

Petals has an ingestion API with two endpoints for pushing content in: one for documents, one for transcripts. The shape is: capture in Screenpipe, run a small script that pulls recent content from Screenpipe's local database and posts it to Petals, watch it surface in your memory graph.

That pipeline is not a product. It is a bring-your-own-script pattern, and it requires comfort with a terminal. Scripts authenticate with an API key you create at Settings → API keys. The Screenpipe guide walks through a realistic implementation; read it and decide if the setup cost is worth it for you.

If your workflow was Rewind and the ambient capture layer was the whole point, the honest answer is: nothing on the market replaces it turn-key today.

Where Petals is different

For the parts of your life that do touch a keyboard, Petals offers something Rewind and Limitless never had: a memory you can see and correct, connected to an assistant that can act.

The graph is the concrete difference. Every claim Petals holds about you — a person you mentioned, a project you are working on, a commitment you made — is a node with provenance you can trace. You can open the memory explorer, find what the assistant knows about a person, verify where it came from, and correct anything wrong. That is different from a search-your-life archive; it is a structure you can audit.

The assistant layer is also different. Petals can set automations, run code, reach into connected apps, and send you a morning brief that draws on what it knows. Rewind could recall; Petals can act on the recall.

The connectors, the automations, the multi-model access (Claude, GPT, Gemini) — these are live today. The memory graph and its editing UI are live today. The Screenpipe pipeline is buildable today with the tools in the guides.

If what you wanted from Rewind was the recall, and the app you built your workflow around is gone, Petals is a different kind of answer — more intentional, less passive, but deeper where it does work.

Start freePricing