Tool Infrastructure
One authority for reusable vector infrastructure. These settings are consumed by multiple agent subsystems rather than belonging to Tool Retrieval alone.
Embedding
configured ·
text-embedding-qwen3-embedding-4bNative vector store
healthy · native-sqlite
Vector collections
9
Config
cf54e77001da51edEmbedding service
Shared OpenAI-compatible embedding authority used by tool search, procedure search, context/chat recall, semantic entities, and memory stores.
Embedding save build:
embedding-save-r2-20261014. The Save button verifies the persisted Base URL before reporting success.Last successful embedding:
— · dimensions:
—
Native vector storage
The shared runtime uses an in-process SQLite vector store for context, tool, procedure, chat, and semantic-entity vectors. This page reports the persistent local topology.
Shared vector collections (9)
artifact_4ebe9306c3798242 artifact_e7126af019a08e3d context_4ebe9306c3798242 context_e7126af019a08e3d procedure_4ebe9306c3798242 procedure_e7126af019a08e3d utility_tool_4ebe9306c3798242 workflow_run_4ebe9306c3798242 workflow_run_e7126af019a08e3d
Memory stores intentionally retain isolated vector databases under
/context/memory_stores/<store_id>/vectors. They use this shared embedding service but do not
share one memory-vector collection.
Consumers
Changing the embedding model changes the vector space used by these consumers. Derived indexes rebuild or migrate according to their own contracts.
| Consumer | Embedding | Vector storage | Notes |
|---|---|---|---|
| Tool Retrieval | shared | shared vector store | semantic candidate generation; deterministic rank remains on Tool Retrieval |
| Procedure Search | shared | shared vector store | verified procedure semantic index |
| Context Retrieval | shared | shared vector store | indexed context chunks |
| Workflow runs / semantic recall | shared | shared vector store | sealed WorkflowRunRecord and semantic-entity vectors; raw chat is not indexed |
| Memory stores | shared | isolated per-store vector store | memory settings enforce model/dimension compatibility |