Knowledge Base for Handles
Use the handle knowledge base to store your business context and retrieve relevant chunks in app flows and MCP-assisted workflows.
What it is
The knowledge base lets you attach structured or unstructured content to a handle, then query it using semantic retrieval. The platform supports:
- document ingestion and indexing
- semantic query with tenant-safe retrieval
- fallback behavior when vector retrieval is degraded
- managed Context Store connector replicas for read-only external/project context
This feature is documented as generally available for eligible tenants and handle operators.
Manual knowledge vs Context Store
Manual knowledge documents are operator-authored content that you create, update, and query directly. Use them for policies, runbooks, sales notes, and approved reference material.
Context Store sources are managed connector replicas. They read configured sources such as web pages, repository docs, or exported upstream resource records, then materialize those records as normal knowledge documents with connector metadata and source/entity status. Use Context Store when an agent needs searchable connector context, but use a direct function binding or provider tool when the workflow needs writes or real-time freshness.
For OpenAnalyst and Smartwork project workflows, use Company Context for OpenAnalyst to decide how to model handle-wide business context, project-scoped containerTags, 10Xmemory boundaries, and Context Store fallback behavior.
When to use it
Use this feature when you need fast, handle-scoped answers from your own docs:
- policy and procedure lookup
- onboarding material retrieval
- support and resolution guidance
- sales and enablement reference recall
Use-case catalog
1. Support assistant
Load troubleshooting guides, runbooks, and common-case resolutions into the handle knowledge base. Query during live support sessions to return precise response snippets.
2. Sales enablement
Ingest pricing playbooks, objection handling notes, and battlecards. Query in call-prep or proposal workflows for consistent messaging.
3. Onboarding copilot
Store onboarding checklists, implementation steps, and role-specific SOPs. Query by topic to guide new team members through first-week tasks.
4. Policy and compliance lookup
Ingest policy text, control summaries, and compliance checklists. Query by policy topic to quickly surface approved language and controls.
Role and access expectations
JWT (creator/operator)
Use JWT-authenticated handle routes for document management and direct handle query workflows:
POST /v2/handles/{handle}/knowledge/documentsGET /v2/handles/{handle}/knowledge/documentsPOST /v2/handles/{handle}/knowledge/rebuildPOST /v2/handles/{handle}/knowledge/query
OAuth connector grant (MCP tools)
Use hosted MCP connector access for assistant-driven retrieval:
https://{handle}.mcp.10x.in/mcp(canonical)https://ai.10x.in/mcp/{handle}/mcp(compatibility fallback)
Grant knowledge.query (plus discovery scopes) for the target handle/workflow.
For Context Store MCP tools, grant context_store.search for search/status and context_store.manage only when the assistant should be allowed to request source refreshes.
Retrieval behavior summary
Responses include retrievalMode so callers can understand how results were produced:
vector: vector retrieval succeeded in shared tenant modevector_dedicated: vector retrieval succeeded in dedicated tenant modehybrid_fallback: vector path ran but lexical fallback supplied the final resultslexical_fallback: vector path unavailable or failed; lexical fallback used
Also expect:
backendtookMsresults[]usage
Different from link embedding targeting
This feature is different from Vector Embedding Targeting:
- Handle knowledge base: document retrieval for question-answer and knowledge use cases.
- Link embedding targeting: content-aware link ordering/personalization for visitor-facing link experiences.
They serve different workloads and should not be treated as interchangeable.
Next step
Continue to the full setup and usage walkthrough:
Updated Jun 19, 2026
