API Connections

Luker doesn't include AI models itself — it connects to external Large Language Model (LLM) services via APIs to generate responses. This page explains how to configure and manage API connections.

Supported API Types

Luker supports a wide range of mainstream LLM APIs:

Chat Completion

API Provider	Description
OpenAI	GPT series models (e.g., GPT-5, etc.)
Anthropic	Claude series models
Google AI Studio / Vertex AI	Gemini series models
OpenRouter	Aggregation service routing to multiple model providers
Custom OpenAI-compatible API	Any service compatible with the OpenAI API format (e.g., various proxy/relay services)

Text Completion

API Provider	Description
KoboldAI	Locally-run open-source models
Ollama	Local model management and inference tool
llama.cpp / TabbyAPI	Local inference backends
Text Generation WebUI	Oobabooga's web interface

INFO

Chat Completion and Text Completion are two different API modes. Most commercial APIs (OpenAI, Claude, Gemini) use Chat Completion mode; local models typically support both. If you're unsure, Chat Completion is the more common choice.

Connection Manager

Luker provides a Connection Manager to manage multiple API connection profiles.

Creating a Connection Profile

1. Open the settings panel and find the Connection Manager
2. Click "New Profile"
3. Enter a profile name (e.g., "Claude Sonnet 4.5," "GPT-5")
4. Select the API type and fill in the connection parameters
5. Save the profile

Switching Connections

Select a different profile from the Connection Manager dropdown to switch instantly. Switching connections does not affect your current chat completion preset.

Managing Multiple Connections

You can create any number of connection profiles, for example:

• A low-cost model for everyday conversation
• A flagship model for high-quality creative writing
• A profile for local models

The Connection Manager lets you switch between them quickly without re-entering API addresses and keys.

API Key Configuration

Obtaining API Keys

Each API provider has its own way to obtain keys:

• OpenAI: Create an API Key at platform.openai.com
• Anthropic: Create an API Key at console.anthropic.com
• Google: Get an API Key at aistudio.google.com
• OpenRouter: Register and get a Key at openrouter.ai

Entering Keys

Enter the corresponding API key in the connection profile. Keys are securely stored on Luker's server side and are never exposed on the frontend.

TIP

If you're using a self-hosted Luker instance, API keys are stored on your own server. If you're using someone else's Luker instance, be mindful of key security.

Model Selection

After configuring an API connection, you need to select the specific model to use. Luker dynamically loads the list of available models based on the API type.

For APIs like Claude and Gemini, Luker supports dynamic model lists — automatically fetching the latest available models from the API without manual updates. You can also customize the model list for each API source. See Other Improvements for details.

Gemini History Caching on OpenRouter

Under OpenRouter Gemini Prompt Caching, enable Cache stable chat history to explicitly cache a prefix containing the system prompt, summaries, and older conversation messages. This is opt-in and applies only to OpenRouter Gemini models that advertise explicit cache support. Direct Google AI Studio/Vertex connections and other models keep their existing behavior.

Keep recent turns uncached defaults to 2 completed turns; the current user input is also always excluded. Raise it if your regexes or extensions rewrite older messages as their depth changes. The boundary is selected from the final outbound messages, so consecutive user messages count as one turn and tool results do not count as new user turns. Message text, roles, summary placement, tool calls and media are preserved.

Gemini uses only the last explicit breakpoint for normal message content. Luker therefore holds one breakpoint at the same history position across requests instead of advancing it on every turn. It checks the prefix before each reuse. After five minutes, or when an edit, swipe, summary replacement, context trimming, system instruction or tool definition changes that prefix, it selects a new boundary. New messages remain outside the fixed prefix until the next refresh. A larger uncached-tail setting also takes effect on the next request.

The existing Cache system prompt toggle remains useful for short requests without enough history. With history caching active, Luker places one history marker instead of adding a competing system marker. Explicit markers supplied by another extension are preserved and take precedence.

These settings are saved with connection profiles and can also be read or changed using /gemini-enable-history-cache and /gemini-cache-keep-recent-turns. Backend callers that omit UI settings can use gemini.enableHistoryCache and gemini.cacheKeepRecentTurns in config.yaml.

Checking the result

In Request Inspector, inspect the final outbound request: the selected history message's last text block should contain cache_control: { "type": "ephemeral" }. The server log also reports created, reused, refreshed, no-history, or external-breakpoint for the local boundary plan. A reused boundary is not proof of a provider cache hit. Check OpenRouter's usage.prompt_tokens_details.cached_tokens, cache writes and total cost across several turns.

OpenRouter documents a five-minute Gemini explicit-cache lifetime and cache creation/storage charges. A larger cached prefix can cost more when rebuilt frequently or used only once. Long pauses, upstream routing, minimum cache sizes and server restarts can reduce reuse. Luker stores only bounded, user/credential/conversation-scoped hashes and positions in memory, not provider cache objects; it cannot guarantee hits or control the upstream cache lifetime. See OpenRouter's Gemini caching documentation.

Proxy Settings

If you need to access APIs through a reverse proxy, you can configure it in the connection profile:

• Proxy address: The relay service URL
• Proxy password: The relay service authentication password (if required)

Proxy settings are part of the connection profile — different profiles can use different proxies.

Relationship with Preset Decoupling

In Luker, API connections and chat completion presets are completely independent concepts:

• Connection profiles manage "which API, which model, and through what address"
• Chat completion presets manage "which prompts and which sampling parameters"

You can freely combine them. For example:

• Use the same Claude API connection with different presets to switch writing styles
• Use the same carefully tuned preset to compare results between OpenAI and Claude

This decoupled design lets you optimize connections and presets independently without interference.

See Preset System and Preset Decoupling for details.

Slash Commands

Luker's Connection Manager provides slash commands for power users:

Command	Description
/profile [name]	Switch to the specified connection profile, or view the current profile name
/profile-list	List all connection profiles
/profile-create <name>	Create a new profile with current settings
/profile-update	Update the currently selected profile

Request Inspector

Luker includes a built-in Request Inspector that lets you view detailed information about generation requests, including the complete request content sent to the API and the returned response. This is very useful for debugging connection issues or optimizing prompts.

Next Steps

• Learn how the Preset System controls AI response behavior
• Learn the basics of Character Cards
• Learn the basics of Chat Management