Skip to main content
Use a context bundle when you need to assemble the right context for a specific run — gathering relevant knowledge and memory into one place before an orchestration task starts. Think of it as packing a briefcase: you decide which knowledge stores and memory stores matter for this job, and the bundle gives the task a single, consistent view to query against. A context bundle is a prepared set of grounded context drawn from knowledge stores and memory, assembled for a specific run. It provides a consistent, queryable view of the information a task needs, and can be refreshed as underlying sources change.

Operations

Create a context bundle

string[]
Knowledge stores to include in this bundle.
string[]
Memory stores to include in this bundle.

Response

string
Unique bundle identifier.
string[]
Included knowledge stores.
string[]
Included memory stores.
string
ISO 8601 creation timestamp.

Get a context bundle

Query a context bundle

Retrieve grounded context from within the bundle using semantic search.
string
required
Semantic search query. Length: 1–10,000 characters.
integer
default:"10"
Number of results. Range: 1–100.

Query response

array
Ranked results from across the bundle’s stores, each with content, score, source_type, and source_id.

Refresh a context bundle

Updates the bundle to reflect current content in its underlying stores. Call this after ingesting new documents or writing new memory items.

Errors

Security

Production recommendations

  • Scope each bundle tightly to the sources a run actually needs — unrelated stores dilute retrieval.
  • Refresh on a cadence that matches how often the underlying content changes; a bundle reflects its sources at assembly time.
  • Reuse one bundle across related runs for consistent, reproducible context.

Rate limits

10,000 requests per organization per hour.
A bundle reflects its sources at assembly time. Refresh after:
  • New documents are ingested into a referenced knowledge store.
  • A connector sync completes.
  • New memory items are written to a referenced memory store.
Include only the stores a specific task needs. Over-broad bundles (many unrelated stores) dilute retrieval relevance. Create separate bundles for different task types.

Knowledge Stores

Sources for bundle content.

Memory

Recallable context included in bundles.

Orchestration

How bundles support task execution.

Data Platform Overview

The full data model.

Next steps

Explore Orchestration to put knowledge and memory to work in tasks.