FlosForge Reader for macOS

Private content reader and manager for macOS

A native RSS reader and content manager and mapper. Pull full articles, set up custom keyword alerts, and freely tag or annotate your reading. No accounts, no tracking, just your feeds.

No subscription
No account
Runs locally

Coming soon

Email me when it is out See it working

FlosForge Reader on macOS, with the sidebar, the article list and one article open in Reader

Requires: macOS 15 or later
Feed formats: RSS, Atom, JSON Feed and OPML
Feed gallery: 440 feeds in 62 categories
Accounts: None, and no sync service

Reader mode and full webpage mode

Reader shows the article stripped to its writing. Web shows the publisher’s page with the trackers blocked, and says so. When a feed sends only a summary, the article’s own page is fetched as you open it.

Reader, Web and Get full text on every article
Trackers blocked in the web view
Extraction rules you can edit per source
Read aloud, find in article, and a text size control

The same article shown in the Web view, with a trackers-blocked badge beside the tab

Track and get notified on watch-words

Name the terms worth interrupting you for. Each one carries its own matching rules, lives in a folder, and produces one notification per refresh rather than one per article.

Whole words, case sensitivity, or a regular expression
Example lines that show what a term would catch
Hits highlighted in the article and counted per row
Blocked words for the opposite case

The Watch Words editor, showing a term with its matching rules

Notes, tags, comments and linked articles

Notes, comments and links each get their own row in the sidebar, because they are yours rather than the publisher’s. A comment on an article outlives the article: pruning what arrived does not take what you wrote about it.

Your own comment on someone else’s article, kept when the article is pruned
Link two articles and say why they belong together
Collections are shelves, not folders — an article stays where it arrived
Search across the library, grouped by kind, with the line that matched

The Comments tab of the article inspector, with the empty comment box visible

Apply AI prompts to articles and lists

Run a prompt against whatever is open. The result carries which provider and which model produced it and a badge saying it happened here, and it is one button away from becoming an ordinary note that stays after the article is gone.

Apple Intelligence first, Ollama second, a cloud provider only with your own key
Automatic tagging chooses from tags the workspace already has, never invents one
Language detection safeguards prevent local models from hallucinating on unfamiliar languages
Two separate acknowledgements: running here, and sending elsewhere, are different bargains

The AI panel open beside an article, showing a prompt result labelled as on-device

Reading analytics and visual content mapping

Visualize your reading habits and library growth over customizable time periods. The interactive Content Map plots connections between linked articles and research notes, revealing topic relationships without relying on algorithmic recommendations.

Configurable reading metrics and volume charts over days, weeks, or months
The complete statistics and reading history view exports cleanly
Interactive node graph mapping manual research links and topic clusters
Fetching polls only the active workspace, with configurable caps per source

Content Statistics, showing headline numbers and charts for a chosen period

Every feature, in full

A list of features you can find in the application.

FlosForge Reader feature categories

Everything52Sources & workspaces8Reading12Watch words7Notes, tasks and links10Sharing6AI & privacy9

Multiple workspaces
Each keeps its own sources, tags, prompts and schedule, and overrides the global settings.
Folders for sources
Workspaces, folders and individual sources can each override the fetch schedule.
Fetch on your schedule
On a timer, on launch, or only when you ask — and only so many sources at once.
Background workspaces
Off by default: only the workspace you are in is polled, and the others catch up when you switch.
A ceiling per source
At most so many new articles per source per check, newest kept — an overnight wire service does not become the day.
Blocked words
Terms that stop an article arriving: refused as the feed is read, never stored, never counted.
Collections
Belong to a workspace, hold articles and notes, and can hold other collections.
Overview rows
Unread, Today, Watch Words, Starred, With Notes, With Comments, Linked, All Articles and Archived, each with its own count.
Reader and Web
The extracted writing, or the publisher’s page rendered in place.
Trackers blocked
The web view carries the badge, on the page itself.
Get the full text
One request per article you actually read, and a button that re-fetches when extraction gets it wrong.
Extraction you can tune
A container selector, things always removed, the tags the body is rebuilt from, and a link ratio above which a block is navigation rather than writing.
Stop at “read also”
Cuts the article where the rail of other stories begins, on a list of ending phrases you can edit.
Identify as
Which browser the fetch claims to be.
Read aloud
The article, spoken, from the toolbar over it.
Find in article
Search inside what is open, with a match count and next/previous.
Marking as read
When you open it, so arrow-keying a list clears everything you pass — or on your own terms.
List density and paging
Row density, what the sidebar badges count, and how many rows build before you scroll.
Starred and archived
Set from the article’s own toolbar; each has a row in the sidebar that collects them.
Text size
Per your eyes, not per the publisher’s stylesheet.
A term, a rule
Whole words only, case sensitive, or a regular expression.
Folders
Terms group into folders, and each folder counts what it caught.
Scoped to one workspace
A term belongs to the workspace it was written in.
Preview before saving
Example lines under the rules show what the term would and would not catch.
One notification per refresh
Naming the terms that hit — not one notification per article.
Highlighted in the article
Matches are lit up in the text, and each row carries how many times the term hit.
Menu bar and Dock
Either can count unread watch-word hits, across every workspace rather than the open one.
Notes
Written against the article and listed in the workspace, not filed in a separate app.
Comments
Kept even if the article is later pruned.
Tags
A vocabulary per workspace; an article takes tags from the list that workspace already has.
Automatic tagging
An on-device model picks from the existing tags. It never invents one, and it is never a hosted model.
Languages it may label
Tick only what your model reads — asked about a language it does not, a small model answers with plausible nonsense rather than declining.
Tasks
Status, priority, and an optional due date; without one the task never appears on the calendar.
Linking articles
Link one article to another in the same workspace, with an optional line about why they belong together.
Both directions
Linked from and Links to are listed under the article itself.
Content Map
One node per article, one line per link, pointing at what was linked to.
Content Statistics
Charts over a period you choose, each switchable, and the whole thing exportable.
Share cards
A card is drawn on this Mac out of an article you already have, and goes nowhere until you export it.
The publisher’s image
Off by default — putting it on your own post is republishing it, so that is your decision rather than the app’s.
Size and format
The aspect ratio a card is drawn at, and the file it leaves as.
Your handle and link
Filled in once, and optionally appended under the copied post text.
AI instructions
What the sparkle button offers when rewriting a card’s text or drafting the post beside it.
Off per workspace
Worth doing for a workspace holding reading you would rather not be a keystroke away from posting.
Apple Intelligence first
On device by default. The article never leaves the machine.
Ollama second
Your own local models, free and private, on hardware you control.
Gemini only on request
Off until you enable it and supply your own key — and the pane says plainly that the article then leaves this Mac.
Two acknowledgements
One for models running here, one for providers you send prompts to. Until each is accepted, tools of that kind will not run.
One switch for all of it
Nothing in this app contacts a model provider while AI features are off.
Custom prompts
Write your own, choose where each may run — article, list, collection, note, task or project — and give the model separate instructions.
Save an answer as a note
An AI result becomes an ordinary note on the article, or nothing at all.
Keys stay in the keychain
A backup carries which providers you had set up, never the key itself.
Passcode
Locks on launch, and after an idle time you set.

Multiple workspaces
Each keeps its own sources, tags, prompts and schedule, and overrides the global settings.
Folders for sources
Workspaces, folders and individual sources can each override the fetch schedule.
Fetch on your schedule
On a timer, on launch, or only when you ask — and only so many sources at once.
Background workspaces
Off by default: only the workspace you are in is polled, and the others catch up when you switch.
A ceiling per source
At most so many new articles per source per check, newest kept — an overnight wire service does not become the day.
Blocked words
Terms that stop an article arriving: refused as the feed is read, never stored, never counted.
Collections
Belong to a workspace, hold articles and notes, and can hold other collections.
Overview rows
Unread, Today, Watch Words, Starred, With Notes, With Comments, Linked, All Articles and Archived, each with its own count.

Reader and Web
The extracted writing, or the publisher’s page rendered in place.
Trackers blocked
The web view carries the badge, on the page itself.
Get the full text
One request per article you actually read, and a button that re-fetches when extraction gets it wrong.
Extraction you can tune
A container selector, things always removed, the tags the body is rebuilt from, and a link ratio above which a block is navigation rather than writing.
Stop at “read also”
Cuts the article where the rail of other stories begins, on a list of ending phrases you can edit.
Identify as
Which browser the fetch claims to be.
Read aloud
The article, spoken, from the toolbar over it.
Find in article
Search inside what is open, with a match count and next/previous.
Marking as read
When you open it, so arrow-keying a list clears everything you pass — or on your own terms.
List density and paging
Row density, what the sidebar badges count, and how many rows build before you scroll.
Starred and archived
Set from the article’s own toolbar; each has a row in the sidebar that collects them.
Text size
Per your eyes, not per the publisher’s stylesheet.

A term, a rule
Whole words only, case sensitive, or a regular expression.
Folders
Terms group into folders, and each folder counts what it caught.
Scoped to one workspace
A term belongs to the workspace it was written in.
Preview before saving
Example lines under the rules show what the term would and would not catch.
One notification per refresh
Naming the terms that hit — not one notification per article.
Highlighted in the article
Matches are lit up in the text, and each row carries how many times the term hit.
Menu bar and Dock
Either can count unread watch-word hits, across every workspace rather than the open one.

Notes
Written against the article and listed in the workspace, not filed in a separate app.
Comments
Kept even if the article is later pruned.
Tags
A vocabulary per workspace; an article takes tags from the list that workspace already has.
Automatic tagging
An on-device model picks from the existing tags. It never invents one, and it is never a hosted model.
Languages it may label
Tick only what your model reads — asked about a language it does not, a small model answers with plausible nonsense rather than declining.
Tasks
Status, priority, and an optional due date; without one the task never appears on the calendar.
Linking articles
Link one article to another in the same workspace, with an optional line about why they belong together.
Both directions
Linked from and Links to are listed under the article itself.
Content Map
One node per article, one line per link, pointing at what was linked to.
Content Statistics
Charts over a period you choose, each switchable, and the whole thing exportable.

Share cards
A card is drawn on this Mac out of an article you already have, and goes nowhere until you export it.
The publisher’s image
Off by default — putting it on your own post is republishing it, so that is your decision rather than the app’s.
Size and format
The aspect ratio a card is drawn at, and the file it leaves as.
Your handle and link
Filled in once, and optionally appended under the copied post text.
AI instructions
What the sparkle button offers when rewriting a card’s text or drafting the post beside it.
Off per workspace
Worth doing for a workspace holding reading you would rather not be a keystroke away from posting.

Apple Intelligence first
On device by default. The article never leaves the machine.
Ollama second
Your own local models, free and private, on hardware you control.
Gemini only on request
Off until you enable it and supply your own key — and the pane says plainly that the article then leaves this Mac.
Two acknowledgements
One for models running here, one for providers you send prompts to. Until each is accepted, tools of that kind will not run.
One switch for all of it
Nothing in this app contacts a model provider while AI features are off.
Custom prompts
Write your own, choose where each may run — article, list, collection, note, task or project — and give the model separate instructions.
Save an answer as a note
An AI result becomes an ordinary note on the article, or nothing at all.
Keys stay in the keychain
A backup carries which providers you had set up, never the key itself.
Passcode
Locks on launch, and after an idle time you set.

See it working

Take a look at the user interface of the application.

The article list filtered to one watch word, with matches highlighted in the article

The “Link an article” dialog, searching the workspace for the article to link to

A collection selected in the sidebar, with the list narrowed to its members

The search overlay open over the window, with results grouped by kind

The Content Map, drawing the links between articles as a graph

The Article Text settings pane, showing the extraction rules

The Fetching settings pane, showing the schedule and the per-source limits

The AI settings pane, listing the providers and the acknowledgements

The Tagging settings pane, showing automatic tagging and the languages it may label

Opt-in AI that runs on your Mac first

FlosForge Reader answers with models running on your own hardware. A cloud provider is used only if you switch it on and supply your own key.

First

Apple Intelligence

On-device. Nothing leaves your Mac — no key, no account, no cost.

Default
Then

Ollama

Your own local models, on your own machine or a box you control.

Local
Only if you say so

Gemini API

Off until you enable it and enter your own API key. Your key, your account, your billing — never proxied.

Opt-in

Not in the app

Here is exactly what Reader does not do:

No sync, no cloud, no account. Nothing leaves the Mac except requests to the publishers you chose and, if you switched it on, the model provider you chose. The single-file backup is the answer to a second Mac.
No related-article suggestions and no ranking. Links between articles are made by hand, always, and what you subscribed to is what you get, in the order it arrived.
No posting to social networks from the app. Share cards are exported as images and captions for you to post.
No custom themes: sixty-five built in, none of your own.
No podcast or audio player. A Podcasts category exists in the gallery; its feeds arrive as articles.
No AI providers other than Apple Intelligence, Ollama and Google Gemini, no cost or currency figures anywhere in AI usage, and Ollama is reached on localhost only.

How every FlosForge app works

No telemetry

No analytics, no crash reporting, no usage pings. Not configurable — simply absent.
Local storage only

Your data stays on your own device, stored in private application sandboxes under your control.
Open exports

Everything exports to standard formats you own (CSV, GPX, JSON, Markdown), with zero lock-in.

Coming soon

FlosForge Reader for macOS is in active testing in real-world scenarios, and will be released in the near future.

Development (done) Features being built
Active Testing (current stage) In daily use, tested in real-world scenarios
Release prep Store listing and final checks
Released Available to download

Private content reader and manager for macOS

Reader mode and full webpage mode

Track and get notified on watch-words

Notes, tags, comments and linked articles

Apply AI prompts to articles and lists

Reading analytics and visual content mapping

Every feature, in full

Multiple workspaces

Folders for sources

Fetch on your schedule

Background workspaces

A ceiling per source

Blocked words

Collections

Overview rows

Reader and Web

Trackers blocked

Get the full text

Extraction you can tune

Stop at “read also”

Identify as

Read aloud

Find in article

Marking as read

List density and paging

Starred and archived

Text size

A term, a rule

Folders

Scoped to one workspace

Preview before saving

One notification per refresh

Highlighted in the article

Menu bar and Dock

Notes

Comments

Tags

Automatic tagging

Languages it may label

Tasks

Linking articles

Both directions

Content Map

Content Statistics

Share cards

The publisher’s image

Size and format

Your handle and link

AI instructions

Off per workspace

Apple Intelligence first

Ollama second

Gemini only on request

Two acknowledgements

One switch for all of it

Custom prompts

Save an answer as a note

Keys stay in the keychain

Passcode