What is Luker
Luker is a deeply refactored roleplay chat platform based on SillyTavern. It retains SillyTavern's mature character card ecosystem and data format compatibility while introducing extensive innovations in data transfer architecture, extensibility, and built-in tooling to deliver a more efficient and powerful roleplay experience.
Luker is fully compatible with SillyTavern data — character cards, world info, and presets can be used directly with zero migration cost. If you decide to stop using Luker, you can downgrade back to SillyTavern at any time without data loss.
Why Luker
SillyTavern is an excellent roleplay frontend with an active community and a rich character card ecosystem. Building on that foundation, Luker introduces systematic improvements in the following areas:
More Efficient Data Transfer
Most save operations in SillyTavern use full-payload transfers — a message edit, settings toggle, or world info change sends the complete data to the backend. For cloud-deployed users, this means significant bandwidth consumption.
Luker introduces an incremental sync mechanism that uniformly uses patch endpoints compliant with the RFC 6902 standard. Toggling a plugin setting that previously transferred 3 MB of data now takes less than 200 bytes. Save operations also support debounced triggering and conflict detection, fundamentally preventing data corruption caused by concurrent writes.
Stronger Extensibility
Luker provides richer infrastructure for plugin developers: character/preset state APIs, managed regex providers, unified extension injection settings, a function call runtime, and more. Plugins can conveniently reuse the user's existing API and chat completion presets without redundant configuration.
Built-in Professional Tooling
Luker ships with several professional tools designed for roleplay scenarios — Memory Graph, multi-agent orchestration, character card editing assistant, search plugins, and more — all ready to use out of the box without installing third-party extensions.
Core Features Overview
Memory Graph
A character memory system built on a knowledge graph. Events, characters, locations, and plotlines from roleplay sessions are distilled into typed nodes connected to each other. Before a reply is written, a recall pass finds the nodes most relevant to the current scene and injects them into the creative context — when the protagonist returns to a place they visited earlier, an early character who has long been off-stage is recalled. Recall methods include: LLM Recall (the default — the model selects relevant nodes from the store, with multi-round exploration) and RAG Recall (vector retrieval over an embedded store, with optional cross-encoder rerank and optional LLM query rewrite). Extraction runs on its own after a reply, either through the built-in pipeline or through an orchestrator sub-agent.
Orchestrator
Before the creative LLM generates a response, a team of agents plans the scene: one distills the recent context, another drafts what this turn should accomplish, a reviewer checks the plan, and the result is packaged into a short briefing for the writer. Execution modes are available: Spec (a fixed Stage → Node pipeline, the default), Single Agent, Agenda (a planner dispatches agents through tool calls), Loop (one agent runs an iterative tool loop until it finalizes), and Director (a main agent and a sub-agent team explore the context and draft the message body). All agent actions are retained as a runtime trace you can inspect. Orchestration configurations can be bound to character cards and imported/exported along with them.
Skills
Reusable knowledge packs that agents read on demand, instead of one giant system prompt. Skills support the Anthropic Claude Skills format. The default Director profile ships with bundled skills, and skills can travel with a character card or a preset, so distributing a card distributes the writing rules that make it work.
Character Card Editing Assistant (CEA / CardApp Studio)
An AI-assisted character card editing tool with an integrated CodeMirror 6 code editor. It edits character cards, world info, and CardApp code through natural language conversation, committing real changes via structured tool calls, with diff-based approval — rejected changes never take effect. When you replace or update a card, it detects world info changes and offers to import the new book, keep the old one, or merge them with AI help. Regular cards use the popup editor; cards with an embedded CardApp open in the more capable Studio.
→ CEA Overview · Popup · CardApp Studio
Preset Assistant
An AI assistant for chat completion presets. It reads the current preset's real values, explains what each parameter does, edits prompt entries as well as sampling values, and compares against a reference preset — returning changes as diffs you approve. The same workflow covers presets whose downstream consumer is the orchestrator.
→ Preset Assistant Documentation
Search Tools
Provides web search capabilities for AI through the provider configured in the Search Tools plugin. It works as a callable tool for the creative LLM, or as a pre-request agent that automatically searches before generation and writes results into world info.
TTS NPC Dialogue Attribution
With TTS enabled, every quoted line in a reply can play with its own speaker's voice. A background pass works out who says each quote — NPCs it discovers appear in the voice map automatically, and names you add in advance are recognized from the first playback.
→ TTS NPC Dialogue Attribution Documentation
Chat Merge and Split
Split a chat at any turn, or merge two chats into one; branch history stays consistent on both sides.
→ Chat Merge and Split Documentation
LAN Sync
Pair two Luker instances on the same network and keep chats, cards, world info, and settings in step between them — nothing is pushed to a cloud.
Standalone App
Run the complete server and interface as a desktop or mobile app for macOS, Windows, Linux, Android, or HarmonyOS. The app runs everything itself — no Termux, no manual Node install, no port forwarding.
→ Desktop App Guide · Android App Guide · HarmonyOS App Guide
Change the Model, Keep Your Prompt Setup
In SillyTavern, API presets and chat completion presets are switched together, so changing the model can silently overwrite your API address, key, or prompt configuration. Luker separates the connection fields from the generation parameters, letting you freely combine different LLM backends with different prompt presets.
Nothing Lost to a Closed Tab
Generation runs on the backend and streams to the UI, and saves happen on the backend as data changes reach the server. Reload the tab, close the laptop lid, or lose your Wi-Fi connection mid-reply: the generation keeps running on the server and reconnects to your UI when you return. Data changes are written with incremental patches rather than full overwrites, which also prevents concurrent writes from clobbering each other across tabs and devices.
→ Backend Storage and Lifecycle
Function Call Runtime
A unified function call / tool call runtime supporting these modes:
- Native tool calls: Compatible with native tool call formats from OpenAI, Claude, Gemini, and other APIs
- Plain-text function calls: Implements tool calls through a text protocol, suitable for models that don't support native tool calls
CardApp
An embedded application runtime within character cards. Allows character cards to carry custom application logic, providing context APIs and lifecycle management.
→ CardApp
Card-Bound Presets and Personas
Character cards can bind dedicated chat completion presets and user personas. Bound presets and personas are independent of the global list, won't pollute the user's global configuration, automatically disappear when the character card chat is closed, and can be imported/exported with the character card. Card creators no longer need to ask users to manually import dedicated presets.
→ Card-Bound Presets and Personas
Request Inspector
A per-user generation request diagnostic tool that can trace request details for all backends (including image generation and vector embedding / rerank calls), making debugging and troubleshooting easy.
Authentication and Quotas
Supports GitHub / Discord OAuth login. Administrators can configure storage quotas for each user. Discord login can additionally require users to be members of a specific server or hold specific roles.
More Features
- Backups, storage, and migration — back up and restore by data category, inspect server-side and browser-side storage, and move a whole setup to a new device over the LAN.
- Structured data for card authors — per-message variables and the state system.
- Troubleshooting — a log viewer with frontend and server sources, plus a one-click debug export.
- Day-to-day comforts — immersive mode, mobile background keep-alive, desktop notifications, lazy media loading, undo toasts, personas carried by Character Cards, collapsible groups for presets and prompts, model lists loaded live from the provider, World Info activation tracing, preset-associated World Info, and plugin-registered regex rules.
Compatibility
Luker maintains full data format compatibility with SillyTavern:
| Data Type | Compatibility |
|---|---|
| Character Cards (PNG/JSON) | ✅ Fully compatible, bidirectional import/export |
| World Info / Lorebook | ✅ Fully compatible |
| Chat Logs | ✅ Fully compatible |
| Chat Completion Presets | ✅ Fully compatible |
| Third-party Extensions | ✅ Compatible, with isomorphic-git fallback support |
| User Settings | ✅ Fully compatible |
Bidirectional Migration
You can migrate from SillyTavern to Luker at any time, and vice versa. Data generated by Luker-exclusive features (such as Memory Graph, orchestration configs, etc.) is stored in separate state files and won't affect SillyTavern's core data structures. However, it's still recommended to back up your data before migrating.
Next Steps
Ready to get started?
→ Getting Started — Install and deploy Luker