The Vector Store Inventory Tool
Inspect every vector store for a user — per-source counts, collections, sample items, and storage tables — so agents and operators know exactly what knowledge is available before they query it.
You can’t trust a RAG answer you can’t audit
When an agent returns "I couldn’t find anything," is the knowledge missing, or just not indexed? Without visibility into the vector store, retrieval becomes a black box and teams stop trusting it.
Silent gaps
A source that was never indexed looks identical to a source with no relevant content.
No ground truth
Operators have no quick way to confirm what’s actually searchable per source.
Debugging is guesswork
When retrieval underperforms, there’s nothing to inspect to find out why.
Onboarding new sources
After connecting a system, teams need to confirm the index populated correctly.
A live readout of your retrieval layer
Inventory
Per-source counts and collections
Know exactly what’s indexed.
The tool reports how many items each source holds, which collections exist, and the underlying storage tables — turning the vector layer from a black box into something you can verify.
- Item counts per source
- Collection and table listing
- Populated vs empty sources
- Scoped to one tenant
Jira · GitHub · Confluence
Samples
Representative sample items
Eyeball the content, not just the numbers.
For each populated source the tool returns a few real sample items so you can confirm the right content was embedded — and spot mis-ingested or stale data at a glance.
Verify content quality
Operations
Diagnostics agents can act on
Built for self-checking workflows.
An agent can call inventory first to decide whether retrieval is even worth attempting, or to tell a user which sources are available — all scoped per user and run on-premise.
Before you query
Parameters
The all_vectors_inventory tool accepts these inputs when an agent calls it. Required inputs are flagged.
default: true Optional Include representative sample items for populated sources.
default: 3 Optional Maximum sample items to return per source (1–5).
Where inventory pays back
RAG health checks
Confirm every connected source is indexed and populated before trusting answers.
Onboarding verification
After connecting Jira, GitHub, or Confluence, verify the embedding job actually populated.
Coverage reporting
Show stakeholders exactly what knowledge the assistant can and cannot see.
Retrieval debugging
When answers are weak, inspect the index to separate "missing data" from "bad query."
Agent self-check
Let an agent confirm sources exist before it promises a cross-system answer.
Capacity planning
Track index growth per source over time to plan storage.
Assigned to agents, orchestrated as networks
On VDF AI, an industry’s use cases map to agents, and you assign tools like this one to those agents. Compose multiple agents into a governed, on-premise network.
What changes after you assign it
Questions about the Vector Store Inventory tool
What does the vector store inventory tool do?
It inspects the Jira, GitHub, and Confluence vector stores for a given user and returns per-source counts, the collections and storage tables present, and optional sample items. It is how you verify what is actually indexed before relying on retrieval.
Why would an agent call inventory?
An agent can run inventory as a pre-flight check — to decide whether retrieval is worth attempting, to tell a user which sources are available, or to self-diagnose when an answer comes back empty.
Can I limit which sources it inspects?
Yes. Pass the sources array to inspect only a subset (jira, github, and/or confluence), and use include_samples and sample_limit to control whether and how many sample items are returned.
Is it safe to run in a shared environment?
Yes. Every call is scoped by user_id for multi-tenant isolation and runs on-premise, so one tenant can never see another tenant’s index.
How does this relate to federated search?
Inventory tells you what exists; federated vector search queries it. They are frequently assigned to the same agent so it can confirm coverage and then retrieve.
Tools that work well alongside this one
Where this tool delivers value
Make your retrieval layer auditable
See the vector store inventory tool give agents and operators a live readout of what’s indexed — on-premise.