Your network as a graph in Venom People
See you, every one of your people and their companies as a graph, with the links between people, why two people are connected, the People list's filters, colour and size by derived fields, a panel for what you select, focus, freeze, clusters, and people like someone.
People → Graph (/people/graph) draws your network on the same canvas as the Notes graph: you in
the middle, your people around you, and the companies they work at.
| What | How it is drawn |
|---|---|
| You → a person | a thin line (your connection; from LinkedIn it is dated) |
| Person ↔ person link you made | a coloured line, one colour per kind (colleague, introduced by, family…) |
| Person ↔ person relationship Venom found | a thinner coloured line, one colour per kind (same company, same school…); see Relationships |
| Person → company | a dashed line |
| Person → a company they left | a dotted line, fainter the longer ago: left within 2 years, 2 to 5 years ago, longer (the list's recency shading) |
| Person → group | a dotted line to a school, place or topic hub when you group by one |
| You → your employer | a bright dashed line |
Drag the background to pan, scroll to zoom, drag a node to pull it. The search box lights matching people and dims the
rest; choosing one selects them. It finds people by name, role, company or tag: type a tag plain or with its #
(investor, #investor), and words combine (ana #investor). Tags are not in the graph the page loads; the first
time you click into the search box, People fetches the tags on your own people (never a colleague's), and from then on
the hover card shows them too. Hover a node for its card (role, circle, strength, last touch, the first line of
their context card, their chips) and why they are connected to the others shown. An employer whose dates are
unknown (a profile line with no start date) is drawn as a past employer and its line says "dates unknown", as in the
list. The legend lists only the kinds
the graph shows.
Colour and size
Colour colours people by circle (the default), seniority, warmth, outreach, country, region, sector, function or
gender, with a legend and each value's count. Seniority, warmth and outreach use fixed scales; for country, region,
sector and function the nine most common values get a colour each and the rest are Other. Size is strength (the
default) or seniority. Changing either repaints the picture where it is; the layout does not start again. Both are kept
in the URL (colour=, size=), so a link opens the same view.
Select, focus, freeze
Click a person, company or group to select it: its panel opens on the right (a bottom sheet on a phone) with the card, the last touch, and chips for their companies now and before (shaded by how long ago, as in the list), country, seniority, warmth, outreach, sector and tags. Each chip opens the list filtered by it. Nothing selected, no panel: the canvas takes the whole width. In the panel:
- Focus shows only them and their connections and fits the view to them. Press it again (Show everyone) or Esc to see everyone again; the layout is not run again.
- Open profile (or Open company) goes to their page. It is the only way the graph opens a profile: a click selects, it never navigates.
Click an edge to select it: its two ends light up and the panel says why they are connected (every relationship and link between them, with each reason); selecting an edge leaves focus. A click picks a line only when it is the only one under the pointer: in the thick of thousands of lines, zoomed out, a click is a click on empty canvas, and zooming in parts the lines so each can be picked. Your own line to each person is never picked. Click empty canvas or press Esc to deselect (Esc leaves focus first). While something is selected the layout holds still, so nothing drifts while you read; it carries on when you deselect. Freeze in the toolbar holds it still at any time, Unfreeze lets it go on. A new picture (a filter changed, or People like them) always lays out once and then holds still if something is still selected or Freeze is on. The panel's buttons are reachable with Tab.
Relationships
Venom links two of your people when they share something it already knows about them. Each link has a kind, a reason in words, and a strength:
| Kind | From | The reason reads |
|---|---|---|
| Same company | their company now, past employers you recorded, and the experience in their profile | "Both at Acme 2019–2021" (overlapping years, strongest), "Both at Acme now (since 2020)", "Both worked at Acme, at different times" |
| Same school | the education in their profile | "Both studied at NUS 2014–2015" |
| Same event | the event you met them at, and events your interactions were at | "Both at GTC 2024" |
| Same city, same country | where they live | "Both live in Singapore" (the country only when the cities differ) |
| Worked in the same country | where their jobs were, and where they live | "Both have worked in the UK" |
| Topic from notes | your notes, how you met and your interactions, read against your topic list | "Your notes on both are about fintech" |
| Same tag, same circle | your tags; your inner or active circle | "Both tagged #fintech" |
| Mentioned together, introduced | @ mentions in your notes and interactions; the introducer you @ mention in how you met |
"Mentioned together in your notes on Kai", "Sarah introduced you to Zoe" |
| Same deal, same sequence | Funnel deals you may see, your Outreach sequences | "Both on the deal “Renewal”" |
Two people who share several things have one line with every reason; its colour is the strongest one's kind, and its strength combines them. These relationships are worked out, never saved as links: add a link yourself from a person's page when you want one. A company, school or place more than 30 of the shown people share draws no line between each pair; it shows as its hub instead (the company node, or a group when you group by it).
Topics from notes
Your topic list is your 30 most used tags. Venom first reads topics with plain rules (the tag's words in your notes).
Then the background sweep asks Venom's decision model, venom-decide (TypeSafe's Jev behind /v1/systemone), one
yes/no question per topic about each person whose notes changed, and keeps topics at 0.7 or more. The call goes through
your organisation's gear, is booked like People's other AI calls and stops at the same daily People AI limit. Where
nothing serves venom-decide for your organisation, the rules stay. Each person's notes are asked about once per change.
What a viewer sees
A relationship only ever joins people the viewer may see. What you keep privately about your own people (events you met them at, notes and topics, mentions, the introducer, your circle, deals and sequences) relates two people only for you, and only when both are yours. A colleague who sees two of your shared people sees only what anyone who sees them sees: company, profile, where they live, tags. The graph itself draws only your own people.
Kept up to date
Relationships come from a small index of what each person shares, rebuilt for a person when they change (the next time you open the graph or ask an agent) and by a sweep that catches up the rest and anything a day old. Opening an unchanged graph recomputes nothing. A few thousand people stay fast: the index is joined on its keys, and the graph draws at most 6,000 of the strongest relationships.
Filters
The bar above the graph has the People list's filters with the same pills, counts and URL:
country (with + nearby), seniority, company (now or before, or Only where they work now), tag, warmth, outreach
and circle, plus any other list filter the URL carries (region, sector, school, the search words…). Each count is over
your own people, as the graph is. The list's Shared with me (owner=shared) matches nobody here, since the graph
draws only your own people; the graph says so, with Show my people and a link to the list. Open in list opens the list with exactly these filters, and the list's Open in
graph opens the graph with its own; Views opens one of your saved views as a graph.
Filters on the canvas holds what only the graph has: Relationships picks the kinds of relationship to draw (all,
none, or some); Group by clusters the layout around a hub per company (the default), school, city, country, home
region (Southeast Asia, Europe…) or topic; a strength floor; only people I have met (met at an event, a meeting or a
call; a LinkedIn connection alone does not count); Show companies and groups hides those hubs. Grouped by company,
the people who used to work at one of the big companies sit in a fainter outer ring around the people there now. These
are in the URL too (hub=, rel=, min=, met=1). Filters run on the server.
People like someone
Find a person in the search box and choose Like them: the graph shows them and the people they share most with, and a panel lists each of those people with every reason. Show everyone goes back.
Everyone, at a few thousand
The graph draws every one of your people who match the filters, and the companies two or more of them share; the corner says how many. A safety bound of 25,000 people stays far above any network seen so far; if it is ever reached, the corner says "n of N, strongest first". People at the same company (or the same group) start in the same sector of the layout, so clusters form. Only your own people are ever drawn; colleagues' shared people and their links are not.
The page sends each person's name, role, circle, strength and the fields colour by needs, as compact columns with every word sent once; a selected person's tags, context card line, last touch and every employer come when you select them. Past 3,000 people each pair keeps its two strongest reasons (no tags or notes at this size, says the corner); labels show only as you zoom in, and at most 6,000 derived relationships are drawn. Extras go before any person does. Colour and size change on your screen only: nothing is asked of the server. Measured with 6,500 people on a laptop (Chrome, local dev server): the graph's data takes about 0.4 s on the server and reaches the page as 1.6 MB (510 KB gzipped, from 6.0 MB and 1.2 MB as plain JSON), the first frame is drawn about 0.25 s after the page arrives (1.0 to 1.6 s from the click on the dev server), and panning and zooming hold 60 frames a second (median frame 16.7 ms, slowest 34 ms, none over 50 ms), as does the layout while it settles.
Agents
people_graph_get returns the same nodes and edges (top, all, cursor, filter, circle, tag, company,
min_strength, met, kinds, group, like). It returns the strongest 300 by default, to keep answers short (top
at most 2,000); all returns everyone, 2,000 people a page, strongest first: while there are more, the answer has
truncated: true and a next_cursor to pass as cursor, and each page's edges join the people on that page. filter
is the list's URL query (country=sg&nearby=1&seniority=5,6), the same as people_person_search's; owner=shared
matches nobody, as on the page. Each person carries seniority, warmth, outreach, country, region, sector,
fn, gender (left out when unknown) and jobs (employers, r 0 now to 3 long ago). A person's current employer is an
edge of kind company, even when it is also among their recorded employments (such an employer used to come as past);
a company they left is an edge of kind past with its recency. A relationship is an edge of kind shared with its
rel kind, weight and reasons.
GET /api/v1/people/graph is its REST twin.
people_link_list lists a person's relationships with everyone you may see, strongest first, with every reason
(kinds, limit): "people like X". Its REST twin is GET /api/v1/people/{person}/links.
people_know_who answers "who do I know at Acme, from NUS, in Singapore": at (a company, now or before), from (a
school), in (a city, a country or a home region) and worked_in (a country), every one given must hold, each person
with the facts that matched. Its REST twin is GET /api/v1/people/know-who. See the workspace MCP server.