Guide

What your first session with SourcePath looks like.

This walks through the product in the order you will actually use it: what to prepare, what each of the four stages produces, when to reach for each of the six analyses, and what "coverage" and "citations" actually mean. By the end you should be able to tell whether this workflow fits the material you have.

SourcePath beta builds are distributed through Apple TestFlight. Read this product guide first, then visit the Beta page for the current invitation status and the complete installation steps.

Four stages
  1. 01Import
  2. 02Read and organise
  3. 03Analyse
  4. 04Take it with you

Before you start

Two things, and one of them is optional.

A local folder

SourcePath does not ask you to move files into a directory of its own. You point it at a folder you already have; the originals stay where they are and are never modified silently. Disconnecting a source removes the index inside the app, not the files on disk.

An AI provider account (optional)

Import, parsing, reading, search, favourites and collections need no AI at all. A connection is only required to run an analysis, so you can build the library and live with it for a while before deciding.

Four stages

There is a lot on screen, but only one real path through it. On a first run, go through these in order rather than jumping straight to analysis — that is where new users most often go wrong.

  1. Import

    What happens
    Point at a local folder. SourcePath discovers the Markdown, TXT, HTML, PDF and DOCX files inside it and parses them on your device.
    What you see
    Progress is reported as separate counts for discovered, processed, failed and skipped. Parsed documents then appear in the library list.
    Where people get stuck
    Not every file becomes analysable. Scanned PDFs, and DOCX files with embedded objects, with body text mostly inside text boxes, or using a non-standard namespace, are marked partially parsed — still readable and searchable, but they do not enter AI analysis.
  2. Read and organise

    What happens
    Read through the material and use favourites, collections and search to give the batch a structure of your own.
    What you see
    Library and source tree on the left, the document in the middle, conversation and generated artifacts on the right. Reading position and unsent drafts are preserved, so you return where you left off.
    Where people get stuck
    This is the stage people skip. But the scope of an analysis is exactly what you select here — organise first, or the analysis has nothing specific to work on. Running an analysis across hundreds of documents on day one produces vague output.
  3. Analyse

    What happens
    Choose a scope, then run one analysis. The scope can be the whole library, one collection, or a handful of documents you pick by hand.
    What you see
    The result is a generated artifact with its source list, separating documents cited by this result, documents analysed but not cited, and documents retrieved but not matched.
    Where people get stuck
    The six analyses differ a lot in purpose, and the wrong choice produces nonsense. The next section is the comparison table.
  4. Take it with you

    What happens
    Export the artifact into an open format that stays useful outside SourcePath.
    What you see
    Markdown, HTML, CSV and DOCX; PDF export as well on Mac.
    Where people get stuck
    If a source document changes later, artifacts built on it are flagged as possibly stale rather than quietly serving you an old answer. When you see the flag, re-run it.

Six analyses, and when each one applies

This is the step people stall on. The names and subtitles below match the wording inside the app exactly. Read the middle column before choosing.

AnalysisWhen to use itWhat you get
GuideAgreements, disagreements, evidenceThe sources disagree and you need to see where they line up and where they clash.A comparison organised by claim, each item carrying its source location.
SummaryKey conclusions and takeawaysYou already trust this material and just want a conclusion you can use directly.Core conclusions plus a list of takeaways.
ExtractionPull targeted informationYou want structured fields — pulling figures, dates or names out of dozens of reports into one table.Entries matching the target you specified, exportable as CSV.
Topic MapThemes, members, and intersectionsThere is too much material and you do not yet know what is in it. See the shape before choosing what to read.Themes, the documents in each, and where themes intersect.
DiagnosisAI Library DiagnosisYou want to know whether the material itself is good enough — what it covers, what is missing, where it contradicts itself.A judgement about the library as a whole rather than an answer to one question.
CustomYour own analysis instructionNone of the five above fits and you know what you want to ask.Output shaped by your instruction, with the same source and coverage reporting.

What "coverage" and "citations" mean

A core design choice in SourcePath is that it tells you what it did not manage to do. These ideas are unfamiliar at first, but they are exactly what lets you judge whether a result is usable.

Coverage
How much of the material an analysis actually read. When the volume exceeds what one pass can handle, SourcePath does not pretend otherwise — it states how far it got. Incomplete coverage is your signal to narrow the scope and re-run.
Citations
Claims in the result carry the source position where possible, and clicking one jumps back to that passage. This is where the product name comes from: tracing a conclusion back to its source.
Three source states
The result separates sources into cited by this result, analysed but not cited, and retrieved but not matched. Not being cited does not mean a document was irrelevant — only that this particular conclusion did not draw on it.
Staleness
An artifact remembers which revision of the source it was built from. If the source changes, the artifact is flagged as possibly stale instead of silently handing you an outdated answer.

About the AI connection

SourcePath ships no model of its own and does not resell model credit. You connect your own account and spend your own quota.

Usable without one

Import, parsing, reading, search, favourites and collections all work with no AI connection. A key is only needed to run an analysis.

Keys stay on the device

API keys live in the macOS Keychain. They are not written to logs and never land in the library directory.

Requests do not pass through us

Once you confirm an analysis, the request goes from your device straight to the service address shown in the interface. There is no relay server.

Billing is yours

Model calls are billed to your own provider account, and every analysis needs your confirmation before a request goes out.

Applies to SourcePath for macOSContent last verified September 2, 2026

The iPad and iPhone builds have different capability boundaries; see the platform section on the product page.

After reading

If this workflow fits the material you have, you can install the beta through Apple TestFlight. Review the build expiry, backup and update notes on the Beta page before installing.

See the TestFlight guide