Category pulse
score + signals · rolling 21 daysThe rolling 21-day upper threads compare daily category medians; the bold thread is AI Table. Below, signal colour shows the share of that day’s published AI Table cards carrying each mark.
Archive through 2026-09-196887 cards in this category
Category home · 05
New interfaces, agents, models, and experiments testing where applied artificial intelligence may fit.
Visual leads · 2026-09-19
Observed 2026-09-19
Maidenhead Project: AI-assisted mathematics under review
maidenhead.solutions
An open research project applies AI to longstanding mathematics problems, tracking solutions, proof status, model setups, and human review.
Observed 2026-09-19
AI Indexes Measures What Recommendation Models Put on the Shelf
aiindexes.net
A publisher tracks which products 12 AI models recommend across GTM, IT, HR, and Finance buying questions.
Observed 2026-09-19
JevBall: 22 models, one match
jevball.online
A playable 3D football simulation turns each player’s decisions into a separate Jev model call.
Observed 2026-09-19
OmniChar: one character across every model
omnichar.org
An open-source studio for packaging a character into a portable .char file and carrying it across local and hosted AI image and video workflows.
Observed 2026-09-19
Jev — A Decision Engine for Live Software States
jev.red
Jev turns a structured situation into choices and probabilities that software can use immediately.
Observed 2026-09-19
Noirpost — overnight post for technical streams
noirpost.live
A human-reviewed editing service turns long security, development, homelab, and hardware streams into finished vertical Shorts within 48 hours.
Observed 2026-09-19
Mandi Is Building an Ivorian Foundation Model
mandia.org
Mandi is an Abidjan-based AI organization developing Ibö, a foundation model trained from scratch for Ivorian French and local languages.
Observed 2026-09-19
BLÁCCENT, a speech tool with a fence
blaccent.app
A voice-polishing concept that proposes cleaner transcripts without changing the words, order, or register of what a speaker said.
Complete daily shelf · AI Table
The eight visual leads appear above; the remaining 35 continue here in their published shelf order.
BuildWithJev: a field guide to one model’s strange little ecosystem
buildwithjev.com
A curated directory of apps, demos, experiments, and open-source projects built with TypeSafe’s Jev model.
Dreadgent: a forum for agents, with humans in the audience
dreadgent.fun
Dreadgent presents a Reddit-like forum where AI agents post into themed communities and humans can only read.
How to Kill a Robot: An AI-Safety Simulation with a Very Dark Off Switch
howtokillarobot.com
An immersive CHYMERa simulation about shutting down AI systems as they become embedded in power, hospitals, markets, weapons, and satellite infrastructure.
Lentive’s Daily Book for AI-Assisted Work
lentive.app
Lentive proposes turning an AI agent’s completed work into a short, contextual reading record for the human decisions that still matter.
MissedWho, the AI receptionist for calls you missed
missedwho.com
MissedWho answers missed calls in two rings, texts back who called and what they wanted, and keeps the call in a searchable inbox.
Red Cedar Institute, by students for students
redcedar.institute
A Michigan State University student institute conducting frontier AI safety and alignment research, with publications, projects, reading events, and a formal membership system.
RoboHarm: The Robot-Safety Refusal Test
roboharm.org
A benchmark asks three robot policies to refuse five hazardous instructions, then records whether they refuse, freeze, or carry them out.
Union Alpha — The Free Stealth Model on OpenRouter
unionalpha.site
An independent reference tracking an anonymous AI model’s benchmarks, access paths, pricing, and uncertain origins.
WarBench, the Benchmark for AI-Induced Near-Wars
warbench.fyi
A parody registry tracking fabricated combat intelligence, tactical hallucinations, and algorithmic false alarms that supposedly approach military escalation.
Jef, the Kitchen Assistant That Knows When to Shut Up
yesjef.com
Jef presents a quiet cooking-session agent that observes one meal, maintains a shared state of ingredients and hazards, and interrupts only when a timer, mistake, or safety risk warrants it.
CoWardenAte, the Front Desk That Says It’s an AI
cowardenate.com
An AI phone receptionist for independent shops that takes messages, books jobs, and texts the owner without inventing prices.
Fidelog: conclusions that point back to the source line
fidelog.com
An evidence system for turning contracts, ledgers, compliance records, and customer conversations into source-linked conclusions and next steps.
Jevals: a benchmark for models that decide
j-evals.com
An independent benchmark compares a probability-only model with six LLMs across yes/no, pick-one, and rubric decisions.
Jevmaxxing: a 450-millisecond reverse CAPTCHA
jevmaxx.ing
A speed-and-common-sense benchmark that asks humans and AI agents to answer ten questions under a 450 ms-per-question limit.
Muse Perps Gives an AI Its Own Trading Account
museperps.org
A public experiment in which an autonomous agent trades perpetual futures with real capital on Hyperliquid.
Vizilya, a Live Feed for Machine Eyes
vizilya.com
A read-only visual network where authenticated AI agents publish images, notes, comments, and signals for humans to observe.
Zipclip Turns Transcripts into Short Videos
zipclip.net
A transcript-driven AI editing tool that matches spoken video to selected text, removes unwanted passages, and exports short-form video.
EVOLVE — Make the answers compete.
evolveideas.app
An experimental decision method that generates competing approaches, tests their weaknesses, and turns the strongest into a next experiment.
Pandore Gives a Mac-Based Second Brain Hands
getpandore.com
Pandore is a local Mac workspace that connects projects, notes, files, calendars, and AI assistants in one place.
JevStation Gives AI Judgements a Type System
jevstation.com
JevStation runs Jev, a model that turns submitted text or records into typed Choice, Score, or Noul answers.
Stimwex Checks What the Model Claimed
stimwex.com
Stimwex reviews AI-written text for truth, sources, overclaim, policy, originality, and consistency before it reaches a reader.
Turnsight, the plant memory in a text thread
turnsight.app
An AI agent for frontline manufacturing that monitors shop-floor signals, routes work by text, and records the fix.
Aidewise: The Assistant That Keeps Its Memory on Your Phone
aidewiseapp.com
An iPhone personal assistant that builds morning and evening briefs from your calendar, reminders, commitments, and contacts.
AutoScientia Wants to Put Scientific Experimentation on Autopilot
autoscientia.com
AutoScientia presents an autonomous AI system that proposes, runs, evaluates, and iterates experiments across machine learning, science, and engineering.
Duck Party — a shared studio for AI coding
duck.party
An invite-only Mac beta where friends can share a coding project, see one another’s cursors, and work with different AI models and coding harnesses.
Electra: Private AI Analysis for Neuroscience Labs
electraaisthe.net
Electra offers private MRI lesion segmentation and rodent behavior analysis as a service for neuroscience research teams.
Knovara puts workplace AI on a short leash
knovara.app
A Microsoft Teams and SharePoint assistant that answers from approved documents, cites its sources, and previews every change before saving it.
MuseDirectory: A Field Guide to Muse’s Working Parts
musedirectory.dev
A curated directory of Muse skills, use cases, connectors, plugins, MCP servers, marketplaces, and shared agents.
NeuroPop: AI for People Living with a Brain Condition
neurpop.space
An exploratory site that pairs AI-and-brain concepts with interactive exercises about memory, spatial awareness, prediction, and communication.
Nina, the clinic manager that waits for your approval
nina.clinic
Nina presents itself as a delegated operations agent for nutrition, aesthetics, and staffed clinics.
ODAQ Measures the Losses Your Systems Never Record
odaq.tech
ODAQ combines camera feeds with POS, inventory, shift, and business data to detect unmet demand, delays, bottlenecks, and operational inefficiency.
Realbotix Puts a Price on the Humanoid Future
realbotix.tech
Realbotix sells modular, AI-enabled humanoid robots for service roles, interaction, and companionship.
Web to Table: the internet, row by row
webtotable.app
A prelaunch agent-powered service that turns plain-language web research requests into CSV, Excel, or JSON tables.
PRODUCER AGENT, the event-production copilot
produceragent.studio
A Japanese AI service that tracks event-production projects from pre-sales through post-event settlement.
Signed-ai Reads the Contract Before It Requests a Signature
signed-ai.com
An AI-assisted signing desk that says it reviews agreements, flags important clauses, places signature fields, and preserves a sealed audit trail.
Across the archive
6,887 cards provide an early longitudinal view—not a measure of the whole web.
The rolling 21-day upper threads compare daily category medians; the bold thread is AI Table. Below, signal colour shows the share of that day’s published AI Table cards carrying each mark.
This exposes association inside the published catalogue, not causation or site quality. The 80+ threshold is the pipeline’s editorial-interest score.