Sixgree blog

Inside the Galaxy physics lab, and the replay of your network growing

2026-09-29T00:00:00Z

The Galaxy is the first thing Sixgree shows you: you in the middle, your connections on rings around you, the ones who can open the most doors closest in. It's drawn by a force simulation, the same idea behind graph views in tools like Obsidian: every dot pushes the others away, lines pull connected dots together, and the picture settles where the forces balance.

The physics lab hands you the controls. It's experimental, and it lives in the Galaxy's Filters panel, under Physics.

The forces

Five sliders set how the Galaxy lays itself out, under the names Obsidian gives its graph's forces:

Center force (0 to 3): a pull in towards you. Higher keeps the whole network a tighter, rounder disc.
Repel force (0 to 300): how hard dots push each other apart, the space between clusters.
Link force (0 to 15×): how tightly a line holds: your connections to you, a circle to its connection, and someone in several circles to each of them.
Link distance (0.05 to 6×): how long the lines are: how far a circle sits from its connection, and your connections from you.
Rings (0 to 5×): how hard each tier holds its ring. At the default, S sits nearest and D furthest out. At 0 the rings let go, and the Galaxy finds its own shape.

A sixth, Orbit, sets the whole map slowly turning round you, every circle with it, so nothing loses its shape. At 0 it holds still.

Two more set the look: node size and link thickness. Dots can be sized by power score, by their lines, or by how many people hang off them, and names can show for the hubs, the S tier and catalysts only, or for all your connections. A Names switch in Filters turns them off altogether.

Everything moves the Galaxy in place, and Reset to today's layout puts it back.

Clusters

Layout picks Rings, Orbit or Clusters, the fastest way to see the lab do something. Clusters draws your network the way Obsidian draws a graph: it lets the rings go (Rings 0), sets Center force 0.8, Repel force 6, Link force 6× and Link distance 0.15×, and sizes dots by their lines. Each connection with a scanned circle becomes a big hub with their circle round them, anyone in two circles ties them together, and your network turns from rings into neighbourhoods.

Tick Light up a branch on hover and hovering a dot lights up its branch: everyone behind it, and the chain back to you. The rest dims.

Clusters, on an invented network.

Colour, find and names

Colour by repaints the Galaxy five ways, each with a legend:

Tier, the default, from S to D;
Heat: power as a thermal map, the higher someone's score among the people showing, the hotter;
Degree: your connections, the people they know, and anyone further out;
Company: the eight most common companies in view, and everyone else in grey;
Warmth: how recently you've been in touch, from LinkedIn's export read in Outlink → Messages & follow-ups (experimental).

Find takes a name, a company or a role from two letters. Everyone who matches lights up and the rest dim; Enter flies to the best match and opens their card.

The replay

Replay plays your network growing. Each connection appears on the day you connected, and their circle arrives with them. There's a time slider, a length of 5 seconds, 15 (the default), 30 or a minute, and Loop. If you've read LinkedIn's export in Messages & follow-ups, your job starts are marked along the timeline in gold and your posts in blue, and the job you were at shows as the replay passes each date.

The replay needs "connected on" dates, which come from a scan or from LinkedIn's own export. The invented sample network has none, so there's nothing to replay until you bring your own network in.

Saving it

Save a picture makes a PNG of the Galaxy at twice its size on screen, with its legend.
Record the replay saves it as a video: MP4 (H.264) where your computer's window supports it, otherwise WebM, at 12 frames a second. The button only appears where the window can record.
Saved layouts keep a set of slider settings under a name, so a layout you like is one click away.

Beside Save a picture, the lab reminds you that it shows real names, so check it before you post it. Nothing leaves your computer unless you share the file yourself.

Why a physics lab?

Because rings are a good first picture and a poor only picture. Letting the rings go shows who gathers people around them. Colouring by company shows where your network clusters. And the replay shows something no list can: how your network grew, and which chapters of your working life it came from.

The layout code is in lib/galaxy-lab.js, with its tests.

Local-first: why Sixgree has no account, no server and no telemetry

2026-09-29T00:00:00Z

A map of your network is a map of other people: their names, their jobs, who they know. That's a good reason not to send it anywhere. So Sixgree doesn't. There's no account to create, no server that holds your data, and no telemetry. The app, and everything it knows, runs on your own computer.

This post explains what that means in practice, and where the edges are.

One folder

Everything Sixgree knows about your network lives in one hidden folder in your home folder, .six-degrees:

the network itself, in a single SQLite database file;
backups/, your network with its photos, backed up once a day while the app is open, before each new version first opens it, and before an import or a restore, each one checked as it's made;
avatars/, the profile photos a scan saved;
chrome-profile/, the scanner's signed-in Chrome, if you scan;
small files for the scanner's budget, cooldown and progress.

Settings → Your data shows where the folder is, what each part takes up, and the backups in it. Delete the folder and Sixgree knows nothing. A LinkedIn CSV you import is kept there too, in csv-network.json, apart from any network you scan, until you remove it.

What the app contacts, and when

Nothing goes online by itself. The app contacts:

LinkedIn, only while you scan, or when you press Auto to send someone a connection request. Each profile photo is saved on your computer as the scan reads it, and the app shows photos only from there, so browsing your network never contacts LinkedIn.
GitHub, only when you click Check for updates, to read the newest version number, and in the Mac app when you then click Install and restart, to download it.
PyPI and GitHub, only without the Mac app (with npx sixgree or from source), once, when you click Set up the scanner: for the scanner's Python packages and, if your computer has no Python it can use, a private copy of Python. Every file is checked against a checksum built into Sixgree.

One experimental feature runs on a timer, and only if you turn it on: the daily messages sync in Outlink → Messages & follow-ups opens LinkedIn once a day, in the daytime, while the app is open. It's off until you tick it.

There is no telemetry, no analytics and no crash reporting. The web framework the app is built on has anonymous usage statistics of its own; they're switched off. And the app never checks for updates on its own.

Local, and locked to your computer

The app is a small web server that listens only on your own computer, at 127.0.0.1, and opens in a window (or your browser, on Linux). Running on localhost brings its own risks, and the app guards against them:

Other websites can't drive it. Any web page you visit can try to send requests to a local server. Sixgree refuses writes whose browser headers say they came from another site, including another port on your own computer, and answers requests addressed to any other hostname with an error, which blocks a trick called DNS rebinding.
If it's ever exposed, the dangerous routes lock. The ten routes that delete, replace or export your data, start the scanner or update the app need an ADMIN_TOKEN whenever the server is bound to any address other than your own computer. Unset, they refuse every caller.

The full threat model is in SECURITY.md.

Moving it yourself

Because there's no account, there's no sync. To take your network to another computer, Settings → Your data → Export backup file… makes one file that you move yourself. It's built from an allow-list: your network, your settings, the scanner's progress and its LinkedIn budget, and photos if you want them. Your LinkedIn sign-in never goes in.

The other computer treats the file as untrusted. It checks the database's integrity and its structure, and each file's SHA-256, before anything changes, and applies it the next time the app starts. What was there before goes into backups/, and stays for at least 30 days.

Updates you can check

When the Mac app updates itself, it checks the download against the release's published SHA256SUMS file and verifies its code signature, bundle and version before putting it in place. If the new version won't open, the old one is put back. The one-line Terminal install does the same checksum check. Since 1.0.0 the Mac app is also signed by Blake Burford and notarized by Apple, which macOS checks when you open it.

Open, so you can check

None of this has to be taken on trust. Sixgree is open source under the MIT licence, and every line described here is on GitHub, with tests. That's the other half of local-first: a tool that reads your network should be one you can inspect.

Settings → Your data.

Scanning slowly and safely: pacing, budgets and the honest risk

2026-09-29T00:00:00Z

Sixgree can show you who your connections know, but only if it can read their connection lists. LinkedIn's own data export doesn't include them: it lists the people you're connected to, and nobody else. So the app has an optional scanner that reads them from LinkedIn, in a Chrome window you sign into yourself.

This post is about how that scanner behaves, and why. Start with the part that matters most.

The risk, in plain words

Automating LinkedIn may break its User Agreement, and LinkedIn can restrict accounts that do it. The scanner reads slowly, keeps to a budget and stops at the first push-back, but the risk is yours.

Accounts have been restricted for this, and no amount of pacing makes the risk zero. That's why scanning is optional, and why the two other ways in carry no risk at all:

The sample network: 150 invented connections and the 598 people they know, built into the app, for trying every view.
LinkedIn's own export: Settings & Privacy → Data privacy → Get a copy of your data → Connections. It's read on your computer and shows the people you know, though not who they know.

If you do scan, everything below is there to keep it slow, easy to follow and easy to stop.

The scanner drives your own installed Google Chrome, with a profile of its own that the app keeps in its data folder. You sign into LinkedIn in that window yourself, once. The app never asks for, sees or stores your password. That signed-in profile is the one thing in the data folder that grants access to your account, so the app never puts it into a copy of your network, and the docs ask you never to sync or share it.

A budget, not a race

Every page of someone's connections is a LinkedIn search, and every circle scan opens their profile once. The scanner counts both, against limits you set on the Scan page:

Searches: 50 a day and 250 a month by default. The month follows LinkedIn's own, from midnight Pacific time on the 1st.
Profile views: 50 a day by default, and at least 60 seconds between any two, even across separate scans. There's no "no limit" choice.

When a budget is used, a scan saves what it read and stops. The next one carries on from the same page. Picking more than 100 searches a day, or no monthly cap, makes the page ask first, and a note stays under the budget, with a button back to 50 and 250, for as long as it's that high.

Slow on purpose

Between the budgets, the pacing is deliberately unhurried:

20 seconds before each page of results, and a minute more after every 10 pages;
about two minutes between one person's circle and the next;
so a long list can take around 55 minutes for one person.

The waits are fixed. They're never randomised to look like a person: the scanner doesn't pretend to be anything it isn't. It reads each list to the end by default (or 10, 25 or 50 pages if you choose), saves every 10 pages, and a read that stops carries on from the same page next time. An In progress list on the Scan page shows everyone whose scan stopped partway, with Resume beside each and Resume all.

It stops at the first push-back

The scanner reads what's on the page, and it stops the moment LinkedIn pushes back: a page that won't open, an unusual-activity warning, restricted viewing, the search limit, a security check or a sign-in wall. It saves what it had read, and keeps a copy of what the page said in the data folder, so you can see exactly what happened.

Then a cooldown lock blocks anything that searches, for a day, or until the 1st for the monthly limit. The Scan page shows when it lifts, and you can lift it early if you choose.

And a scanner that can't read its page says so. It never reports "you have no connections" because a page failed to load.

Out of your way

The scanner's Chrome works in the background. Once you've signed in, its window stays out of sight (hidden, on a Mac), so it doesn't cover the app or keep your typing, and Chrome doesn't stop it to ask about saving a password, translating, notifications or restoring pages. It comes forward only when LinkedIn needs you: to sign in again, or to finish a check it asks for, and the notch says so on every page while it waits. If you'd rather watch it work, Show the scanner's Chrome window in the Scan page's Fine-tune keeps it in front.

What it's doing, on every page

While a scan runs, the notch, a small bar hanging under the header's tabs, says on every page what it's doing and how far it's got, and a dot collects along the header's line for each page read. Hover or click the notch to open it: who it's scanning, your searches over the last 24 hours against your budget, the latest line, and Stop. Stop saves what was read, and Resume on the Scan page carries on from that page.

The one thing it sends

The scanner reads. The one thing it sends is a connection request, without a note, and only when you press Auto on someone: one press is one person, never a batch and never on a timer. Auto stops at 15 requests in any 24 hours and 80 in any 7 days, opens their profile once like a scan does, and waits while a scan runs.

The experimental parts

One newer option is marked experimental on the Scan page, and off by default:

All-day pacing (Auto-Bridge), with an Auto scan button once it's ticked. Up to 8 pages in a sitting, then an hour's rest; searches only between 09:00 and 18:00 on your computer's clock; never more than 40 searches in a day or 200 in a week, however high your budget; and at the daily budget it waits for the budget to free up instead of stopping. A monthly limit or a cooldown still stops it. It hasn't yet been tried on a live account, which is why it's labelled the way it is.

There used to be a second, Hide the Chrome window while scanning, which ran Chrome with no window at all. Every scan now runs out of sight in a real Chrome window instead, which comes forward when LinkedIn needs you, so that option is gone.

Why go to this trouble?

Because the people your connections know are the most useful part of your network, and the one part no official export contains. The scanner exists to get that part, for your own use, on your own computer, as gently as possible, with the risk said out loud rather than buried.

The full details, down to what each page of the scanner reads, are in docs/SCRAPING.md.

The budget, on the Scan page.

Who can introduce you? How Sixgree ranks your way in

2026-09-29T00:00:00Z

The question Sixgree is built to answer is simple: who can introduce me? Not "who do I know", which a list of connections already tells you, but who your connections know, and which of them is the way to each person.

This post walks through how the app works that out: how it ranks the people two steps away, how it counts the ways in to each of them, and how it finds the people only one of your connections can reach. Every example comes from the sample network that ships with the app: 150 invented connections and the 598 invented people they know.

Two steps away

Your 1st degree is the people you know. Your 2nd degree is the people they know and you don't. Sixgree gets the 2nd degree from circle scans: when it reads one of your connection's own connections, each person on that list becomes someone you can reach through them.

A person can turn up in more than one circle. In the sample, the 14 scanned circles hold 723 rows, but only 598 different people, because some people are known by two or three of your connections. That difference is the whole point of the next section.

Ways in

For each person two steps away, a way in is one of your connections whose scanned circle contains them. The app matches people by their profile address (ignoring any query or trailing slash), so the same person found in three circles counts once, with three ways in.

In the sample network:

503 of the 598 people have exactly one way in;
65 have two;
30 have three.

So 84% of the sample's 2nd degree hangs on a single connection each. Real networks vary, but the pattern is common: most of the people you can reach, you can reach one way.

Separation: the ranked list

The Separation tab lists everyone two steps away. It ranks them by their own score first, the same power score every card shows (worked out from their title and their company), and then by how many ways in they have. Ties on both share a rank, shown as "#8=", so no one is ranked above someone identical by accident.

Each row says who can introduce you, and how many others could. A slider reorders the list: 50, the default, is by power; slide towards 100 for the most mutual connections first, the list to work from when you want the warmest route rather than the most senior person, or towards 0 for the rarest ways in. It only reorders: scores, tiers and ranks stay as they are.

Beside each person's tier sits their rarity: how many mutual connections lead to them. It uses LinkedIn's own mutual count when a scan saved one, and the number of ways in otherwise, in five bands: only way in (1), rare (2 to 3), uncommon (4 to 10), common (11 to 30) and warm (31 or more). Rarity never changes anyone's score; it just tells you how easy they are to reach.

Who only they can reach

Now turn it around. Instead of asking "how do I reach this person?", ask "what would I lose if I lost touch with this connection?"

For each of your connections with a scanned circle, Sixgree counts:

total: everyone in their circle who isn't already your connection;
only: the people in their circle no other connection of yours reaches;
reach: their share of everyone, where a person with k ways in counts 1 ÷ k to each of those k connections.

In the sample, the connection with the largest exclusive reach is Bodhi Pereira, a managing director at an invented company called Northwind Labs. Their circle holds 69 people, and 51 of them, 74%, are reachable only through Bodhi. The card says it plainly: only through Bodhi. Show them opens Separation filtered to exactly those 51.

The reach number has a name in network research: it's ego betweenness (Everett and Borgatti, 2005). On a network shaped like this one, you in the middle, your connections around you and their circles beyond them, it gives the same answer as Brandes' betweenness centrality, in one pass over the data. The app ranks your connections by it: the rank on a card's Insights panel is by reach, not the plain "only" count.

Opening the same doors

Two connections can reach mostly the same people. A card shows which of your other connections overlaps theirs most, measured as the share of either circle that both reach (the Jaccard index: both ÷ (their total + the other's total − both)). It only says so when the overlap is real: 15% or more, and at least 3 people in common. "Opens the same doors as…" is a hint that one introduction may do the job of two.

Network health

Network health, under Insights on your profile (the round level button at the top right), puts it together for your whole network:

Reached two or more ways: of everyone you reach through a circle, the share with more than one way in. It's 1 minus (the sum of every connection's only, divided by the sum of every connection's reach). In the sample it's 16%. Higher is safer: lose touch with one connection and those people are still reachable.
Effective reach: Ronald Burt's effective size of your own network, N − 2t ÷ N, where N is your connections and t the ties between them that circle scans have kept. It discounts connections who all know each other. The sample has no ties between its connections, so its effective reach is simply its 150.
Your five connections who reach the most people no one else does, each with Explore to open their circle.

With fewer than five circles scanned, it says the numbers are an early estimate. And it never shows a percentile against other people: there's no one else's data to compare with, and that's by design.

What the numbers are not

A score in Sixgree estimates network position, not what anyone is worth as a person. The model is open, in lib/scoring.js, and the brokerage counts are in lib/brokerage.js, with tests beside them. If a number looks wrong to you, the working is there to check.

Separation, on an invented network.

Sixgree blog

Inside the Galaxy physics lab, and the replay of your network growing

The forces

Clusters

Colour, find and names

The replay

Saving it

Why a physics lab?

Local-first: why Sixgree has no account, no server and no telemetry

One folder

What the app contacts, and when

Local, and locked to your computer

Moving it yourself

Updates you can check

Open, so you can check

Scanning slowly and safely: pacing, budgets and the honest risk

The risk, in plain words

You sign in, not the app

A budget, not a race

Slow on purpose

It stops at the first push-back

Out of your way

What it's doing, on every page

The one thing it sends

The experimental parts

Why go to this trouble?

Who can introduce you? How Sixgree ranks your way in

Two steps away

Ways in

Separation: the ranked list

Who only they can reach

Opening the same doors

Network health

What the numbers are not