Get started
What is Kyalulu?
Understand Kyalulu, a local-first Character AI runtime that stores your characters, conversations and memories, and how it connects to models and phones.
At a glance
- Kyalulu is a local-first Character AI runtime that stores characters, conversations and memories.
- Connect a model provider such as LE, LM Studio or Ollama to generate replies.
- It is a public alpha. Mobile access requires a running PC or personal server.
On this page
Keep a story going with your characters#
Kyalulu is an open-source Character AI runtime for continuing stories with your characters. It keeps conversations, character settings, personas, worlds and memories separate, so you can reuse a configuration in a conversation or a comparison experiment.
You can choose a character or bring settings from SillyTavern and Character Cards. Kyalulu tracks which character revision and memories a conversation uses, alongside sending text to a model.
The runtime and the model have different jobs#
| Component | Responsibility |
|---|---|
| Kyalulu | Store conversations, characters and memories; assemble prompts; display responses. |
| Model provider | Generate responses through LE, LM Studio, Ollama or an OpenAI-compatible API. |
| Web / Desktop | Interfaces to the same runtime. Desktop can also supervise API and LE processes. |
A different model changes writing, instruction following and latency. Installing Kyalulu does not automatically make a large model run on your PC. Start by checking the interface with Mock Echo, then connect a model provider.
Start on PC, then connect a phone#
- PC: run the local version from source and connect your own model or an external API you subscribe to.
- Phone: keep your PC or personal server running and access it over an HTTPS PWA or Remote.
- Evaluation: use Studio / Research to inspect settings and measurements. The independent CharacterBench is a separate evaluation track.
Using the public alpha#
The public repository records implementations and checks for local inference, imports, Memory Lab and comparison experiments. Packaged distribution, long-run model quality, physical Android acceptance and Remote release approval still have open requirements.
Memory is off by default. Enabling it does not guarantee every promise will be remembered. You can inspect and correct memories yourself.
Read release status and supported scope before installation, and data storage and privacy for how information is handled.
Sources for this article
Edited from public GitHub materials. Links are pinned to the reviewed commit.
eea271bReport a correction ↗