Knowledge

Knowledge your agents can actually use.

Build virtual databases from your documents and sources, and let agents retrieve from them. Answers come from what you own, and you can see exactly how it was prepared.

Start free Book a demo

No credit cardBring your own model key

app.cadrixflow.com / knowledge
A Cadrix knowledge base page showing counts of sources, chunks and characters and the four pipeline stages
Virtual databases

One place for what your agents should know.

A virtual database is a named knowledge base: a set of sources, split into chunks, embedded and stored in a vector store. Workflows point at it instead of at loose files.

  • Add sources with document loaders and see each one processed
  • Status at a glance: whether the store is upserted and how many sources, chunks and characters it holds
  • See how many workflows use the store before you change it
  1. 1SourcesDocument loaders
  2. 2ChunkingText splitter
  3. 3EmbeddingsYour model key
  4. 4Vector storeUpserted vectors
Upsert pipeline

From documents to a searchable store.

Every knowledge base follows the same four stages, shown on its page so nothing about preparation is a black box.

  • Sources: load files, web pages or app content with a document loader
  • Chunking: a text splitter cuts content into chunks
  • Embeddings: your chosen embedding model turns chunks into vectors
  • Vector store: vectors are upserted into the store you configure
app.cadrixflow.com / knowledge
The status, counts and pipeline stages of a Cadrix knowledge base
Retrieval for agents

Agents answer from your knowledge, not from memory.

Attach a knowledge base to an agent in a workflow. Retrieval (RAG) runs as part of the workflow, and the run trace shows what the agent was given.

The agent gets a question

A message, an email or an API call starts the workflow and reaches the agent.

It retrieves the best chunks

The agent searches the knowledge base you attached and pulls the most relevant chunks.

It answers from your content

The answer is grounded in what you own, and every step is recorded in the run.

You stay in control

Check the data before you commit it

Preparing knowledge should not be guesswork. Cadrix gives you the tools to inspect each step.

Preview chunks

See how a loader and splitter cut your content before anything is embedded.

Upsert history

Review past upserts and their details, so you know what is in the store and when it changed.

Query the store

Run a query against the vector store and read back the chunks it returns.

FAQ

Questions, answered plainly

Cadrix includes document loaders for common files such as PDF, Word, Excel, PowerPoint, CSV and plain text, plus web pages and several business apps. Check the loader list in the app for the current set.

Yes. Cadrix is bring-your-own-model-key: you connect your embedding provider as a connection and it bills you directly. You also choose where vectors are stored.

Knowledge lives in your workspace and access follows its roles and permissions. See governance for how roles work.

Each knowledge base shows how many workflows use it, so you can see the impact before changing it.

Yes. Open the run trace to see the inputs and outputs of each step, including the retrieval.

Give your agents something real to know.

Create a knowledge base on the free plan and connect it to a workflow in minutes.

Start free Book a demo