|
<< Click to Display Table of Contents >> Raynet One > 2026.4 > User Guide > System management > Manage your data sources > Connection Details AI Chat Connection |
The AI Chat connection links Raynet One to the AI provider that powers the AI assistant Raynée. Without a configured and activated AI Chat connection, the AI Chat interface remains available but cannot generate responses.
For general information on setting up and editing connections, see Connection Details and Set up a connection.
Go to Configuration > Connections > AI+Chat and click Set up... on the AI Chat card, or Edit... if it is already configured. Enter the following settings:
Setting |
Required |
Description |
|---|---|---|
Service URL |
Yes |
The URL of the AI service endpoint. It must be a valid URL reachable from the Raynet One server. |
API key |
Yes |
The credential used to authenticate requests to the AI provider. |
Model |
Yes |
The name of the AI model used to process requests and generate responses, for example gpt-5.1. |
The following values are part of the Edit... dialog of the card. In the setup wizard, they appear in the Advanced settings step when you select Edit default settings. With Use default settings, the defaults below apply.
Setting |
Default |
Allowed values |
Description |
|---|---|---|---|
Timeout (ms) |
360000 |
300000 and higher |
How long Raynet One waits for a response from the AI provider before the attempt fails. |
Maximum attempts |
2 |
1 to 5 |
The maximum number of attempts made for an AI provider request. |
Advanced tracing |
No |
Yes / No |
When enabled, the activity timeline of a chat turn additionally shows the handoffs between agents and the skills they consult. When disabled, the timeline shows only the thoughts, queries and tool calls of a turn. |
Reasoning settings control whether the assistant displays its reasoning steps to users and, when enabled, how much reasoning effort the model applies.
•Reasoning Enabled (default: enabled): When enabled, users see a Thinking... indicator while the assistant processes a request, and can expand the resulting Thinking section to review the assistant's reasoning steps. When disabled, no reasoning output is shown and the model processes requests without applying extended reasoning.
•Reasoning Effort (default: Medium): Controls the depth of reasoning the model applies. Available values are Low, Medium and High. Higher effort levels produce more thorough reasoning traces but may increase response time. This setting has no effect when Reasoning Enabled is disabled.
The following parameters can be set in the environment section of the server service in your docker-compose.yml. They provide the default values of the connection; afterwards, the connection is managed in the user interface. Parameters without a default must be set for the connection to work.
Parameter |
Default |
Description |
|---|---|---|
AiConfig__Url |
- |
The AI service endpoint. |
AiConfig__ApiKey |
- |
The credential used to authenticate requests to the AI provider. |
AiConfig__Model |
gpt-5.1 |
The AI model used to process requests. |
AiConfig__Timeout |
360000 |
Request timeout in milliseconds. Values below 300000 are replaced by the default. |
AiConfig__MaxAttempts |
2 |
Maximum number of attempts per request. Values outside 1 to 5 are replaced by the default. |
The following optional parameters are only available as deployment settings:
Parameter |
Default |
Description |
|---|---|---|
AiConfig__CompactionEnabled |
true |
Compacts the context of a long conversation once it exceeds the compaction threshold, so that chats with many tool calls do not run out of context. |
AiConfig__CompactionTokenThreshold |
80000 |
The context size in tokens above which a conversation is compacted. |
AiConfig__RateLimitMaxWaitSeconds |
90 |
How long, in seconds, a request waits for the model's rate limit to reset before the error is returned. 0 disables the wait. |
AiConfig__MaxToolIterationsPerRequest |
120 |
The maximum number of tool-call rounds one request may take. |
AiConfig__IncludeDetailedToolErrors |
true |
Passes the actual error message of a failed tool call to the model, so that the assistant can report what failed. |
AiConfig__CheckpointIndexMaxEntries |
24 |
The maximum number of conversation checkpoints kept per chat. |
AiConfig__CheckpointTtlHours |
24 |
How long, in hours, conversation checkpoints are kept. |
For the effect of the AI Chat on users, see AI Chat and Chat Management and Context Isolation.