Knowledge

The knowledge system provides contextual data to your AI pipelines. It includes document sources, conversation threads, semantic memory, generated artifacts, and vault secrets — all organized within namespaces for clean isolation.

Namespaces

A namespace is an isolation boundary within an app. Each namespace has its own threads, memory entries, sources, artifacts, and vault secrets. This enables multi-tenant scenarios where each customer, project, or workflow gets its own context without cross-contamination.

Creating a Namespace

To create a namespace, navigate to an app and open the Namespaces tab. Click Create Namespace and fill in the dialog:

Namespace Tabs

Once created, a namespace detail page has five tabs:

• Threads — conversation entities with message history
• Sources — uploaded documents processed into searchable chunks
• Memory — semantic, vector-indexed knowledge entries
• Artifacts — generated content from pipeline executions
• Secrets — namespace-scoped vault secrets and credentials

Sources

Sources are documents that you upload or paste for your pipelines to reference. When ingested, they go through an automated processing pipeline that converts raw content into searchable, vector-indexed chunks.

Creating a Source

Click Add Source on the Sources page. The creation dialog offers two input methods:

After selecting your input method, fill in the remaining fields:

Advanced Options

Expand the advanced options section for additional controls:

• Smart Chunking — toggle for semantic-aware document splitting (enabled by default)
• Indexing Level — controls the trade-off between recall and precision:
  • Standard — balanced performance and recall
  • High Recall — maximize document retrieval at the cost of precision
  • High Precision — prioritize relevance over recall

Processing Pipeline

When you add a source, it progresses through a series of processing stages:

Each stage updates in real time in the UI. If processing fails at any stage, the source shows a Failed status with an error message.

Supported File Types

Source Scopes

Sources can be scoped at two levels:

• Workspace — shared across all apps in the workspace
• Namespace — isolated to a specific namespace within an app

Threads

Threads represent ongoing conversations. Each thread has a message history with user and assistant messages, along with metadata and timestamps.

• Threads are scoped to a namespace
• Each thread tracks message count, creation time, and last update
• Message history is available to inference steps for conversational context
• Pipeline steps can read from and write to threads using thread_update and thread_query step types

Memory

Memory entries are semantic, vector-indexed knowledge that persists across conversations. Unlike threads (which store raw messages), memory stores distilled knowledge that the AI can reference for long-term context.

Memory Types

Memory entries have a status (active or archived) and a scope (namespace or thread). They are vector-indexed for semantic retrieval during pipeline execution.

Artifacts

Artifacts are content generated by your pipelines during execution. They can be code files, documents, reports, images, or any other output. Artifacts have three scope levels that determine their lifetime and visibility:

Artifacts are viewable in the namespace Artifacts tab and in Cortex's scratchpad.

Vault Secrets

Each namespace has its own Secrets tab for managing namespace-scoped credentials. Secrets stored here are isolated from other namespaces and can be referenced by tools and pipeline configurations. See Security for more on vault scoping and access control.

Using Knowledge in Pipelines

Pipeline steps interact with knowledge through dedicated step types:

• source_injection / source_query — query and retrieve document chunks
• source_ingest — add new sources programmatically
• thread_update / thread_query — read and write thread messages
• memory_update / memory_query — store and retrieve memory entries
• artifact_create / artifact_query / artifact_get — manage generated artifacts