Embeddings and aggregates over a profile's public posts. Read-only; nothing is written.
A whole conversation in one page — every reply in order, the author's own chain kept unbroken instead of interleaved with the crowd, images and quoted posts inline, and a wall of everything the thread quoted.
A thousand of their posts, indexed in your tab and searchable instantly with no key at all. Add your own API key and the same retrieved posts answer questions — or become a dossier of themes, arcs and a personality read, every claim cited to a post.
Phrases only they've ever posted — every two- or three-word phrase used exactly once by a handle that no one else on all of Bluesky has ever posted. Their hapax legomena, verified against platform search.
Personal memes + co-memeticists — the phrases a handle repeats that no one else on Bluesky ever posts, plus the small in-group who echo the same private language. The mirror of unique: repeated instead of once, shared instead of solitary.
The dense core of your mutual follows, found before the graph finishes loading. A thousand mutuals is ~20,000 requests to know who follows whom; this reads rows in the order most likely to matter and shows a genuine k-core from a fraction of them — every intermediate answer is true, so stopping early is a legitimate way to finish.
A mood ring around the profile picture — the last ten posts and replies read for tone by jev, the decision-only model, and drawn as a band of colour each around the avatar. The colour is the mood, the haze is how sure the model was. Asked as two ordered axes and one unordered one, because a single angry-to-happy scale would average half fury and half delight into calm.
Six lines read off a whole repository — how evenly you arrive, whether you sleep, how wide you draw, how well-formed you are, whether you are still the same writer, whether you answer. Placed against a curated mutual-follow cluster and drawn as a shareable radar, with one of 30 named archetypes from the pair of lines that run furthest toward the machine and the animal. The whole reference pool is browsable as tiles. A percentile, not a verdict: it is stylometry in aggregate, not an AI detector.
See Bluesky as anyone sees it — type a handle, get their reconstructed timeline, network, and blocks.
Posting profile — personality traits and valence from post embeddings, topics via k-means.
Semantic novelty trajectory — how an account's posting drifts or repeats over time.
Three lusts, one chart — a ternary plot of posting temperament: flesh, knowledge, argument.
Post density head to head — two handles side by side on volume and patterns.
What they liked before they followed — every post a follower had liked before the follow, a hint at what won them over.
A dyadline — every recorded interaction between two accounts (likes, reposts, replies, quotes, mentions, follows, blocks) on one timeline, both directions.
Who blocks whom — paste a set of accounts and a set of posters, get a matrix of which accounts directly block which posters. Public block records, no login.
Mutuals of X who block Y — X's mutual follows (or all followers) that block a third named party, off public follow + block records.
Everyone who blocks a named user — the full count + list, from the firehose-built block index. Also a clean JSON
API for agents.
Post brevity, measured — filter posts by exact word count, ranked by character length.
Group novelty — how wild is your corner? Semantic novelty across a whole list against the group centroid.
These write records: votes, voice posts, answers, edits. Sign in with Bluesky; the data lands on a PDS, not our servers.