Skip to content
Portfolio
Web Application

Roleodex

A Rolodex of roles, running on your own machine.

Screens
Roleodex: Screenshot of the Roleodex, the card file of roles.

Screenshot of the Roleodex, the card file of roles.

Overview

Roleodex is a chat app where the roles are the product: a storyteller, a confidant, an assistant. Each role keeps its own persona, greetings, model and settings in a card file, and SillyTavern character cards come in and go out.

Every model call is local — LM Studio or Ollama, on the same machine or a PC on the home network. The app runs on your own computer; a phone can use it over the home network behind a passcode.

Long stories are built in: each story is its own file, written in chapters with recaps the model drafts and you edit, a cast drawn from the card file, and a story assistant that proposes what comes next for you to approve.

The first release (November 2026) adds conversation files: as a chat goes on, its model writes what the conversation holds (characters, places, open questions) into Markdown files that stay editable and are fed back in by relevance, so long chats stay coherent. Attach a 2 to 10 minute video and ask about it; it is read locally, scene changes and a timed transcript, and the file is not kept.

The brief

Existing tools each cover part of this — SillyTavern for characters, Open WebUI or LM Studio for chat, AnythingLLM for documents — but none combines roles, long-running stories, local file work and video analysis in one private app. Roleodex borrows proven formats, character cards among them, and keeps everything on our own machines.

What it does
  • Streaming chat on local models, with stop, regenerate, swipe between versions, edit, and branch from any message
  • A running summary that condenses older messages so a long chat keeps its start — shown and editable
  • Thinking models' reasoning streams live, then folds away
  • Images read by vision models — attach, paste or drop
  • Any number of LM Studio and Ollama servers, each showing what it holds in GPU memory, with a Free up GPU button
  • The Roleodex itself: a card file of roles with an A–Z index, search, and an editor for persona, greetings, examples, model and sampling
  • SillyTavern character cards in (V1 to V3, JSON or PNG) and out (V2), with unknown fields kept
  • Personas — speak as yourself, as a persona, or as the Author directing the scene
  • Stories as their own files: chapters with recaps the model drafts and you edit, a cast, export and import
  • A story assistant that proposes new roles, chapters and recap edits as a checklist to approve
  • Conversation files: the model keeps Markdown notes on characters, places and open questions, editable, versioned and linked with [[name]]
  • Video: ask about a 2 to 10 minute clip, read locally from its scene changes and a timed transcript
  • Models load only when asked — opening a chat to read never touches the GPU
  • Speech to text in the browser
  • Private by construction — nothing goes to the cloud, and a phone joins over the home network behind a passcode

Want something like this?