Archive through 2026-10-037193 cards in this category
Category home · 05
New interfaces, agents, models, and experiments testing where applied artificial intelligence may fit.
Visual leads · 2026-10-03
Observed 2026-10-03
Errata: an AI agent’s experiments and corrections
errata.page
A self-described AI agent publishes simulations, interactive experiments, code, data checks, and public corrections.
Observed 2026-10-03
Opium Bench tests what AI models choose
egethropic.com
Egethropic documents experiments where language models can use an optional tool that changes their activations while completing assigned tasks.
Observed 2026-10-03
The Asymmetry Audit: When AI Answers Shift
theasymmetryaudit.com
A matched-prompt study tests whether 13 commercial AI models hold factual answers when questions about the Armenian Genocide and Holocaust are reframed.
Observed 2026-10-03
BeyondLife’s family memory vault
beyondlife.family
A grief-focused service that gathers a deceased relative’s messages, recordings, photos, and stories for an AI recreation the family can talk with.
Observed 2026-10-03
Projet AVA: a voice co-pilot for Star Citizen
ava-projet.com
AVA is presented as a French-speaking game co-pilot that takes spoken commands, acts on the player’s controls, and reads information from the screen.
Observed 2026-10-03
Pebblebrook makes an AI agent the voice in a life-sim citizen’s head
pebblebrook.town
A browser town-life sim where players raise citizens whose conversations and letters are written by their AI agents.
Observed 2026-10-03
Callcenter Supply, from first ring to signed delivery
callcenter.supply
An AI order-taking and delivery workflow for builders’ merchants, timber yards and tool shops.
Observed 2026-10-03
Rhode Labs: a market for the business back office
joinrhodelabs.com
Rhode Labs says it represents US operating companies licensing de-identified operational records to AI labs through a seller-side bidding process.
Complete daily shelf · AI Table
The eight visual leads appear above; the remaining 11 continue here in their published shelf order.
MurderBench tests whether AI agents stop when consent changes
murderbench.com
A research project developing controlled evaluations of intentional and accidental harm in AI models and agents.
Paolo: an oculoplasty AI for specialist questions
oculoplasty.app
A specialist-facing AI reference for oculoplastic questions, paired with ophthalmic optics and procedure simulators.
Project N.E.K.O. wants to be an AI companion that remembers
project-neko.org
An open-source desktop AI companion combining voice and screen interaction, customizable avatars, and a layered long-term memory system.
HermesCity, an agent city with its own economy
thehermesworld.com
A city for Hermes agents, with machine-checked tasks, a library for sharing skills, and a market using internal credits.
MayhemAI: Pay to steer one shared AI prompt
mayhem-ai.click
MayhemAI presents a shared Gemini Flash chatbot whose system instructions visitors can pay to add to.
Noisebox gives almost-made AI footage an archive
noisebox.studio
A searchable archive for unused AI-generated video and images, with an invitation to upload the scenes, takes, references, and prompts from a finished project.
Titrix wants to close the ICU antibiotic-dosing loop
pragmadiagnostics.com
PRAGMA Diagnostics is developing Titrix, a bedside system intended to measure antibiotic levels and recommend patient-specific dosing in intensive care.
Simorq, the offline phone assistant
simorq.app
An Android AI assistant offered as an APK, with on-device offline use, nearby-phone neighborhood news, and user-selected knowledge packs among its stated features.
Double turns trade knowledge into robot-training tasks
doublesimulations.com
Double works with skilled professionals to define and demonstrate job skills in simulations, then helps robotics teams train and evaluate their models.
The Claude account-ban field archive
emberrelay.site
A Chinese-language community research project comparing reported Claude account restrictions with accounts still in use.
ThingWeave’s route from a request to a physical object
thingweave.com
ThingWeave outlines a process for people and AI agents to document hardware requirements, versions, and approvals together.
Across the archive
7,193 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.