About Services Tools Blog Contact us

AI Skill and Model Search

A directory of AI skills, tools and models you can run yourself, searched by meaning. Model entries show memory needs and download links.

  • Semantic Search & RAG
  • Cross-sector
  • Website
  • Product
  • Hybrid
  • 2025

The problem

Useful skills and models are scattered across the web. Keyword search misses a close match that uses different words. A model entry that gives no memory figure and no download link does not help someone choose.

What it does

Entries are listed by type and opened from stored notes. A search phrase is turned into a meaning-based search on the spot and matched against the stored entries. Model entries add task tags, a memory estimate and the Hugging Face page the files can be downloaded from. Public pages show a summary and leave out the raw source text.

Outcomes

  • A task phrase finds nearby skills and models even when the words differ.
  • A model entry shows a memory estimate before anyone downloads files.
  • Filters keep skills, tools and models apart.

Simplified mockup

Interactive preview with sample data. Nothing is sent to a server.

Find a card

Unranked until you ask
tool–
Local recorder bridge

Hands a recording from the desk to the transcriber on this machine.

model–
Meeting speech, small

Short model for British English in a quiet room. 8 GB memory, 3.4 GB weights.

skill–
File a meeting on the project

Picks the project folder from the title and who was in the room.

skill–
Action lines from a transcript

Pulls decisions, owners and dates out of a long transcript.

model–
Desk note embedder

Small embedder for notes already on disk. 4 GB memory, 1.1 GB weights.

skill–
Decisions in a rambling call

Finds the decision when the room talked around it.

Meeting speech, small

model
Memory
8 GB
Weights
3.4 GB
Fits beside
16 GB
if the browser is quiet
Summary from stored markdown

Turns a meeting recording into text on this machine. Tuned for British English, two to six speakers, and a quiet room.

A poor fit for a noisy hall or a two-hour lecture. The public page shows this summary, not the raw card from upstream.

speech meetings british-english short-form

8 GB of a 16 GB machine.

Weights
local-store/meeting-speech-small
Nothing is pulled until you queue it

On this machine

Weight size, GB
ModelMemoryWeightsState
Meeting speech, small8 GB3.4 GBnot on disk
Desk note embedder4 GB1.1 GBon disk
Long note summariser16 GB7.8 GBtoo large here

How it works

  1. Store skills, tools and models as entries with a short summary and tags.
  2. Turn a search phrase into a meaning-based search when the request arrives.
  3. Rank entries by how closely they match.
  4. Narrow the list with type and tag filters.
  5. Add a memory estimate and download hints to a model entry.
  6. Build the detail page from the stored notes.

Pipeline

flowchart LR
  Phrase[Search phrase] --> Embed[Embed the query]
  Embed --> Rank[Rank stored cards]
  Rank --> Filter[Type and tag filter]
  Filter --> Card[Summary card]
  Card --> Detail[Memory and download notes]