What PremAgentic is

What PremAgentic does, the claim it makes about the network and how that claim is tested, what setup needs from the internet once, where the retrieval core came from, and what the product does not do.

On-premises, for agents: the name says what it is. Governed knowledge retrieval that runs entirely inside a customer's own network. Point it at the documents an organization already has, and its people and agents get cited answers out of them, with the organization's own permissions enforced before anything is retrieved.

the customer's document systems -> connector -> chunk -> embed (locally)
-> plain PostgreSQL -> access, lifecycle, trust and freshness gates
-> lexical search in PostgreSQL + exact vector search in process, fused
-> cited answers
-> a golden set that proves it works on THIS corpus

At runtime, with the default local embedding provider, there is no outbound call and no API key, and no document leaves the network. That is the product, not a configuration option.

That claim is tested, not assumed. scripts/clean-install/run.sh --offline installs and runs the whole quick start (setup, ingest, a signed-in search over HTTPS, an agent's call over MCP) in a container whose only network interface is loopback. It first shows that the same checks do reach out from a container that has a network, then traces every connection an ingest and the API open: each one goes to the database. The embedding runtime is one dependency that can report on its own: some builds of ONNX Runtime upload telemetry from a long-running process, and PremAgentic switches that off itself (ORT_DISABLE_TELEMETRY=1, unless you have set the variable) before the runtime loads.

Setup is different, and it is stated here so the claim above is not read as more than it is. Installing downloads the embedding model from Hugging Face, the PostgreSQL container image and the .NET packages, once. A site with no internet access has to bring those in first. The optional openai embedding provider sends passage text to a third party and is off unless selected.

PremAgentic is built and maintained by Agave Information Solutions. The retrieval core was developed and measured on a single-user knowledge base of about 1,300 Markdown documents before the authorization model and the connector abstraction were added for multi-user organizations.

01What it does not do

• It does not sign people in through a company directory yet. PremAgentic keeps its own accounts, signs people in with a password over HTTPS, and checks agent tokens; a customer whose proxy already signs people in can pass a PremAgentic sign-in name in a trusted header. OIDC and Windows integrated sign-in are not built.
• It does not write to the source. Ingest is one directional. The customer's systems stay authoritative and this store is rebuildable from them at any time.
• It does not generate prose. It returns passages and citations. What reads them is the caller's business.