Structured data
Relational database records and schemas, such as the product data your agents need to answer a specific question.
Preview documentation
ServoKB documentation.
Start with the concepts. Setup guides and the API reference will follow as the product and integrations become available.
A knowledge layer for your agents.
ServoKB is being built to connect structured databases and unstructured documents in a shared, queryable knowledge layer. Its purpose is to give an AI agent the information needed for a task without placing entire source documents into the prompt.
Bring the pieces together.
Documents and files
PDFs, spreadsheets, and other documents that hold the details missing from a database.
Supported connectors and their setup requirements will be listed as they are validated.
Find what matters for the question.
- Retrieve
Find the information relevant to the current task.
- Prune
Remove unrelated content from the selected context.
- Summarize
Condense details where appropriate before delivering the context.
The savings section is prepared for measured aggregate results. It shows an awaiting-results state until eligible usage is published.
What a saved token means.
We compare the full authorized source text relevant to a completed request with the context ServoKB actually delivers, including its citations and formatting. Both are counted with the selected model’s tokenizer. The comparison does not include your entire library or assume every uploaded file would be sent to a model.
Net tokens avoided equals full-source input tokens minus delivered context tokens. Requests that increase context count against the total. Failed requests, retries, test traffic and baselines that would exceed the model’s context window are excluded.
Cost equivalents multiply that difference by the model’s published standard input rate. They are estimates, not invoice savings or a measure of answer quality. Output tokens, cache and batch discounts, embedding costs and Servo fees are excluded. Token counts may differ between models; model comparisons are alternatives, not additive savings.
Public results will use rounded batches from opted-in workspaces, with no documents, queries, identities or individual usage exposed. Small cohorts are withheld. Each published batch shows its timestamp; there is no continuously simulated counter. Usage collection and automatic batch publication are being built with the retrieval backend.
A shared connection to your agent.
Model Context Protocol (MCP) provides a common way for AI applications to connect with tools and data. ServoKB is designed to expose focused knowledge through an MCP endpoint.
Knowledge sources → ServoKB → MCP → Your agent
Connection URLs, authentication details, and framework examples will be published with working integrations. There is no live public MCP endpoint to configure yet.
What’s available today?
ServoKB is in development. You can explore the product direction and join the waitlist. Account creation, live retrieval, and integration setup are not open yet. ServoAI is coming soon.