Chat Channel

Chat Channels define the conversation with AI Agents. Each channel can have multiple Chat Messages.

AI Agents

The Tygron Platform currently has several default AI Agent channels:

• WIKI Assistant: Searching the Tygron Wiki for information on features, configurations and guides.
• TQL & Combo Generator: Creates TQL queries and formulas for Combo Overlays.
• Project Coder: Can edit data in your Project, by executing TQL queries or interact with project API endpoints.

LLM

List of open source Large Language Models available to the Chat Channel Agents:

• Gemma4 (Speed): Gemma4 model optimized for fast reactions.
• Gemma4 (Quality): Gemma4 model optimized for quality needed for coding and TQL.

Privacy

To protect your data, the following measures are in place:

• Chat messages are stored within your Project and follow the same data and privacy terms described in your license agreement.
• All models run exclusively on Tygron-owned GPU infrastructure.
• Zero Data Retention (ZDR): no data or metadata is stored by the AI model and conversations are not used for AI training.

MCP Tools

List of available Model Context Protocol (MCP) Tools for the Chat Channel Agents:

• get-summary: Get AI Summary on a WIKI subject.
• search-page: Search the WIKI Pages.
• get-page: Get specific WIKI Page.

• validate-endpoint: Validate Tygron API Endpoint.
• validate-query: Validate Tygron TQL Query.

• execute-query: Execute TQL Query in your connected Project session.
• execute-endpoint: Execute API Endpoint in your connected Project session.

Attributes

Chat Channel attributes can be used to guide the behavior of the underlying Large Language Model (LLM).

Attribute	Unit	Description	Default value
TEMPERATURE	number	Temperature controls the randomness of text that is generated by LLMs during inference. A value >1 is more creative, <1 is more strict behavior.	1
TOP_P	number	Top-P sampling selects a group of LLM tokens whose cumulative probability is at least P.	0.95
TOP_K	number	Top-K sampling limits the LLM to choosing from the K most likely tokens at each step.	64
NUM_PREDICT	number	The maximum number of tokens the LLM can generate before stopping.	5000
TIMEOUT_SEC	sec	The maximum time allowed for the LLM before it is stopped.	600
THINK_MODE	boolean	With think-mode enabled the LLM returns both the reasoning and final answer.	1
CACHEABLE	boolean	Debug value, when deactivated the LLM Tools can no longer cache wiki pages.	1
PROJECT_INFO	boolean	Whether to include some basic info about the project that is accessible by the LLM.	1

API Endpoints

• Api session items chatchannels
• Api session event editor chatchannel

References