SageArchive

How it works

Last updated 8 September 2026

SageArchive brings selected recordings, pages and files into a personal Mac archive, alongside notes you write yourself. The useful loop is simple: capture material, find a passage, check the original, then use it in your own work. This page explains the capabilities and their limits. The user guide has the full operating instructions.

Getting the notes off your devices

Sources have different routes. Plaud and HiNotes can use their cloud accounts. reMarkable can use cloud or direct SSH over USB/Wi-Fi; HiDock P1 also has a direct USB route. Inq pen pages first travel through the iPhone app over Bluetooth and into the inq cloud account.

Listing, downloading and converting are separate steps. The app records progress so supported downloads can resume, and keeps source-specific failures in Queue. Retry the failed operation after fixing its connection or model problem; other sources can continue. An unchanged sync should not create another copy within that source.

File imports deduplicate identical bytes and retain multiple folder locations. Device transports have separate identities: switching between cloud and cable may leave separate notes. Follow the setup guide for your source.

Writing it down

With Local selected and the required speech model installed, transcription runs on your Mac without uploading the recording for recognition or paying a per-minute SageArchive processing charge. Optional cloud transcription uses your selected provider, key and source permission instead.

Processing speed depends on the model, hardware and audio. Choose a model and try a short recording before importing a large backlog. Audio is saved before transcription, so a failed conversion can be retried from the note.

Long recordings are cut into pieces before processing, and the cuts are made at a quiet moment near each boundary where possible, to reduce cuts through speech. The timings are then shifted back onto the real clock, which is why clicking a result three hours into a meeting lands where you expect.

About who said what

SageArchive's current on-device transcriber does not separate speakers. Its optional Deepgram transcription can return speaker labels; the Groq and OpenAI transcription paths do not. Cloud transcription requires your API key and permission for that source. An empty speaker list means the transcript has no labels, not that the recording contains only one voice.

Reading handwriting

Handwritten pages can be read locally by macOS's own text recognition, which is good at telling us where on the page each line sits. With permission, an optional cloud model can provide another reading of what the line says. The two readings are lined up against each other so that matched text can inherit available page coordinates.

Not every recognized line has page coordinates. Highlights use the available coordinates and can be imperfect. Unplaced lines have a small dot meaning "found and read, but we could not place it precisely." We would rather show you a gap than a confident box in the wrong place.

Finding things

Search combines the available routes; meaning search requires its local model:

Structured search treats supported identifiers as values rather than unrelated pieces of punctuation and numbers. It can match an IP address against a network range. Date handling can use a note's capture date when a year is missing and retain more than one interpretation when a date is ambiguous. It is useful for recovering technical details, not a guarantee that every number or identifier format has special handling.

The available result lists are combined by their ranks. You can narrow the library with sources, folders, tags and filters, and check a result in the original note. Recognition mistakes and incomplete source text can affect what search finds.

Answering questions

When you ask a question rather than search for a phrase, the app finds the most relevant passages within the selected scope, then hands your question and those passages, with their titles and source metadata, to a language model, and asks it to answer using them.

Clickable citations use the retrieved passages' stored source positions. The app resolves recognized passage IDs to the original note, transcript segment or page; unknown IDs do not become valid source links. This helps you inspect the evidence, but does not guarantee that the generated answer is correct or that a cited passage supports its conclusion. Read the source before relying on an answer.

The default answering engine uses Apple's on-device model when that model is available. Apple Intelligence must be enabled and the model ready; macOS 26 alone does not guarantee it. Answers are grounded in retrieved passages, and you can open their sources to check them. Optional cloud answering requires your API key and the relevant source permissions. It sends the retrieved text needed for the question.

Using a model you run yourself

Instead of the built-in choices, you can point SageArchive at an OpenAI-compatible server you run: Ollama, LM Studio, MLX-LM, llama.cpp, or a hosted gateway that speaks the same API. Presets fill in the usual addresses. Check connection reads the model list the server offers; once you have chosen a model from it, checking again also asks that model a short grounded question, so you learn whether it cites the passage it was given before you rely on it. Small models have small context windows, so SageArchive fits the retrieved passages to the length you configure and tells you when it had to leave some out. The first passage is always sent, even where it alone exceeds a very small budget, so a server may still refuse or truncate that request.

A server reached at localhost is treated as its own category, with a permission separate from cloud answering, and the note records where the text was sent. It is not described as staying on your Mac: the request leaves SageArchive for your server, and what your server does next is outside what this app can see. A server on your network or the internet is treated as cloud, appears under What Has Left This Mac by its address, and shows no cost estimate, because there is no published price to read. Servers that also answer the transcription endpoint can write out recordings as well; Ollama, LM Studio and MLX-LM do not.

Answering in Claude, from your archive

SageArchive can run a read-only server on this Mac that speaks the Model Context Protocol, so Claude Desktop, Claude Code and other MCP apps can search and read your notes and quote them back. Each passage carries a link: a recording opens at the second those words were said, and a handwritten passage opens its note, with the page numbers returned beside the text rather than encoded in the link. It is off until you turn it on, it listens only on this machine, and it requires a token kept in your Keychain. Connected apps can search, read and list; they cannot write, edit, delete or sync.

Only sources you have allowed cloud answering for are shared, because the text an AI app reads goes to that app's provider under their terms, and the setting you already made for cloud processing is the one that decides it. The app lists which sources are being withheld. Replacing the token disconnects every app set up with the old one. Switching sharing off stops the server taking new requests; one already past authentication can finish returning the passages it had found.

Local storage and optional network services

Your archive is stored on your Mac, without a SageArchive account or an archive hosted by us. Vendor account sync and model downloads use their respective services. Local processing does not upload your notes to us.

Optional cloud services can transcribe audio, improve handwriting conversion, answer questions, generate requested summaries or editing proposals, and narrate a note. Cloud permissions start off and are controlled by source and purpose. Depending on the operation, a provider receives audio, page images, retrieved passages or the text of the selected note. The privacy page explains those paths.

Your archive lives in your macOS account's application-support folder. The audio and page images are kept as ordinary files outside the database, so backing up does not mean copying forty gigabytes every time one note changes.

Writing with multiple references

Create an authored note with rich-text editing and autosave. Use headings, lists and links to build a draft or review, and organize it with library folders and tags. A note can link to several other notes. Linked from shows ordinary incoming note links; Referenced by lists the notes quoting a timed source.

For a manuscript review, link to the imported manuscript and your notebook, then add quotations from different interview transcripts. Timed quotations preserve their saved text, source revision and segment range. Renaming the source keeps the relationship; re-conversion does not replace the saved quote. Unavailable sources are identified when you try to follow them.

Ordinary links open whole notes. Timed quotations return to stored audio segments, which may include more words than an exact character selection. Exact anchors into untimed manuscript paragraphs or page regions are not supported. Follow the worked review example.

You can also request a summary or propose an addition or rewrite to your own note, inspect it and apply or discard it. These operations use the selected answering engine and the relevant permissions. Review generated wording before treating it as part of your work.

Capturing and reusing material

Quick Add accepts a typed thought, files and supported podcast URLs. Capture microphone audio, system audio or both with macOS permission. Recording is saved before transcription; this is not live interim dictation. Short recordings can be filed as Clips, with optional copying of completed Clip text. A configurable global shortcut can open capture from another app.

Turn text into saved narration using a local voice or an allowed cloud service. Export selected notes as Markdown, plain text or JSON; SRT/VTT require stored transcript timings and skip untimed notes. Print is for saved summaries. Exports are useful for another writing tool, but are not a complete backup of your database, audio and pages.

Back up the whole archive while SageArchive is closed, or configure its Time Machine inclusion and verify the backup in Time Machine. Keep a matching app/schema and archive when restoring, and preserve newer notes first. Read the recovery instructions.

Choosing whether it fits

SageArchive is a personal Mac archive for people who want to retrieve and reuse material from several sources. It is free to try for 14 days, then $49 once. It is not a shared team workspace or a phone app, and it does not provide a graph view or automatic local speaker separation. Device-account access and vendor subscriptions have their own requirements.

Start with a short recording or a few pages, check the conversion against the original, and try your own search-and-writing workflow. Supported routes do not imply acceptance of every device model, firmware or account failure case. The guide and support answers explain setup and current limits.

Requirements

Download free trial Buy — $49