zvec
Vector, full-text and hybrid search on the embedded zvec engine — no server to run, a collection is a directory on the plugin's own host. Fuses a vector and a full-text ranking by RRF or weights, and can embed text via OpenAI, Gemini, Cohere or a custom endpoint.
- SDK
- Go
- Protocol
inflowv1- Actions
- 12
- Version
- v0.1.0
- Author
- Inflowenger
- Runs on
- Any
inflowv1host - Categories
- AI and agentsDatabases and stores
Reach for it when
Vector search where the deployment should carry as little as possible — no service, no port, no credential. A collection is a directory.
The use people don't expect
Hybrid search is the real reason to pick it. A vector ranking and a full-text ranking from one query text, fused by RRF — which matters because exact terms (an error code, a part number) are exactly what pure vector search is worst at.
A flow that uses it
Each one is a shape of flow, not a template to copy. Node names are the labels as they appear on the canvas; a name with a slash means either would do.
Support search that handles both kinds of question
Starts withA ticket arrives.
- ZVEC
Hybrid search on the ticket text. The full-text half catches the error code; the vector half catches "it hangs when I save".
- Contract
A confident match routes to the known fix. Everything else goes to a person.
- JIRA
The unmatched ones become tickets with the near-misses already attached.
What you getOne query covers both ways people describe a problem, and the whole retrieval layer is a directory on the plugin host — nothing extra to operate. One plugin process per data directory, since the engine takes an exclusive lock.
The reference, not the pitch
This page is about what the plugin is for. Every action, every connection field, the ports, the install one-liner and the caveats are in the catalog entry, which is the thing the maintainer updates.
Near it in the catalog
The task is "do this, here, on this machine" rather than "answer this" — and you want the work to stop and ask when it hits a decision nobody delegated.
CLICKHOUSEClickHouseMany rows, asked questions repeatedly — events, logs, metrics, anything you'll want a GROUP BY over.
CODEX AGENTCodex AgentSame shape as Claude Agent, over Codex — reach for it when you want the workspace fence enforced by the operating system rather than by the model behaving well.