Skip to main content

Configuration: AI Integration

Embeddings configuration keys​

The following configuration keys apply to integrating AI-powered embeddings generation into your RavenDB server.

  • Embeddings can be generated from your document content via AI-powered tasks and stored in a dedicated collection in the database.
  • When performing vector search queries, embeddings are also generated from the search term to compare against the stored vectors.

Ai.Embeddings.MaxBatchSize​

The maximum number of documents processed in a single batch by an embeddings generation task.
Higher values may improve throughput but can increase latency and require more resources and higher limits from the embeddings generation service.

  • Type: int
  • Default: 128
  • Scope: Server-wide or per database

Ai.Embeddings.MaxConcurrentBatches​

The maximum number of query embedding batches that can be processed concurrently.
This setting controls the degree of parallelism when sending query embedding requests to AI providers.
Higher values may improve throughput but can increase resource usage and may trigger rate limits.

  • Type: int
  • Default: 4
  • Min value: 1
  • Scope: Server-wide or per database

Ai.Embeddings.MaxFallbackTimeInSec​

The maximum time (in seconds) the embeddings generation task remains suspended (fallback mode) following a connection failure to the embeddings generation service. Once this time expires, the system will retry the connection automatically.

  • Type: int
  • Default: 60 * 15
  • Scope: Server-wide or per database

AI Agents configuration keys​

Ai.Agent.Tools.TokenUsageThreshold​

The recommended token threshold for a tool response to the LLM.
If the response exceeds this threshold, a notification will be raised.

  • Type: int
  • Default: 10000
  • Scope: Server-wide or per database

Ai.Agent.Trimming.Summarization.SummarizationResultPrefix​

The text prefix that precedes the summary of the previous conversation.

  • Type: string
  • Default: "Summary of previous conversation: "
  • Scope: Server-wide or per database

Ai.Agent.Trimming.Summarization.SummarizationTaskBeginningPrompt​

The instruction text that precedes the serialized conversation when requesting a summary.

  • Type: string
  • Default: @"Summarize the following AI conversation into a concise, linear narrative that retains all critical information. Ensure the summary:
    • Includes key identifiers, usernames, timestamps, and any reference codes
    • Preserves the original intent of both the user and the assistant in each exchange
    • Reflects decisions made, suggestions given, preferences expressed, and any changes in direction
    • Captures tone when relevant (e.g., sarcastic, formal, humorous, concerned)
    • Omits general filler or small talk unless it contributes to context or tone Format the output in a structured manner (such as bullet points or labeled sections) suitable for fitting into a limited context window. Do not discard any information that contributes to understanding the conversation's flow and outcome."
  • Scope: Server-wide or per database

Ai.Agent.Trimming.Summarization.SummarizationTaskEndPrompt​

The user-role message that triggers the conversation summarization process.

  • Type: string
  • Default: "Reminder - go over the entire previous conversation and summarize that according to the original instructions"
  • Scope: Server-wide or per database
In this article