Ägir

Janitor-inspired local roleplay frontend

Features and other cool things

• Standalone and (optionally) self-hosted
• Mobile-first
• No sync between devices (chats and scenarios stay on your device)
• No JLLM (as in real LLMs only)
• No mobile app (as in no regulations, moderation and censorship)
• Unlimited number of first messages
• Native import/export
• ⧖ rEmber (WIP): a customizable recap assistant that helps to keep track of roleplay state. Basically a "Chat Memory" from Janitor but not half-assed and actually usable.

Less cool things

• No pagination in chats and scenario library (yet). It still seems to work well in chats with 500+ messages though.
• No lorebooks (yet). I intend to add SillyTavern lorebooks support, but Janitor's ones can not be extracted.

Usage

You can use a github-hosted version. GitHub basically serves a static html-page, so there's no any other traffic going through it (or anyone else except your LLM provider). Instructions for self-hosted version are below.

Setup

1. You need an LLM provider to use with Ägir. Janitor AI calls them "proxies", in Ägir they are "providers". Refer to their guide to acquire one for free on OpenRouter.
2. Navigate to Settings > LLM providers, add your credentials into the form, and hit "Save provider".
3. Navigate to Settings > Personas and add a persona -- a character representing you in roleplay.
4. Navigate to Characters and import or create a scenario. The download option allows to fetch characters from janitorai.com and jannyai.com.
5. Once ready, pick a scenario from Characters library, hit play, select a persona and start the roleplay session.

You also can run an LLM locally with LM Studio on desktop, just figure out how to host a web server with it and add a provider with localhost connection.

You can also use Fenrir to add characters from Janitor -- it's a sniffer proxy that helps to extract character definition even if it's hidden.

Where do I get characters?

From JanitorAI, mostly. Use datacat.run to extract character definition, download the JSON file, then import it via Characters > Import.

Downloading Janitor character links directly is supported too, but characters from jannyai.com lack some important data like character's chat name and alternative openings.

Alternatively, you can check Armories, which are decentralized character repositories, native to Ägir. There is one added from the start which you can use for test purposes (or actually play with them).

How do I use ⧖ rEmber?

Navigate to an existing chat, then open rEmber dialog via chat menu ([☰]). Ägir will send chat fragments to the model you pick and will ask it to provide summary using the prompt you can customize in the same dialog. The provided summary will then be stored and attached to chat messages. Only the most recent summary will be included during the roleplay. You can edit summary text manually.

After the first summary is added, a counter will appear in the topbar showing how many messages aren't included in the summary. The counter will glow if unprocessed message count exceeds the stride size.

Only the last N chat messages (70 by default, customizable in settings) are sent to the model. By default the summary extends this tail: the summary and every message it does not cover are always sent, so the model keeps the full state. You can turn this off in settings, in which case the summary is only sent while the message it is attached to is still inside the tail. Note that the tail is a cost-saving measure and most providers cache prompt prefixes anyway, so a short tail rarely saves much; its main use is keeping small-context models from overflowing.

Self-host

Clone or download the repo on your device:

git clone https://github.com/MilesVII/agir
cd agir

Then just serve the page from root with whatever tool you prefer (i use serve for example). If you don't know how, you can ask any free use bot on Janitor for instructions. Or just ChatGPT or Gemini if you're lame and boring.

Credits

Maid Ägir placeholder: https://gelbooru.com/index.php?page=post&s=view&id=11145081

Send icon (in chat): https://www.vecteezy.com/vector-art/21909958-abstract-logo-business-shape-element-clip-art-illustration-vector-geometric-simple-modern-and-clear

Dependencies

• marked and dompurify for markdown rendering
• tokenx for token count estimation
• and rampike, it's just my own utility lib for rendering elements because the whole thing works on vanilla