AI Models Overview · Documentation

PacketSafari can use a hosted provider or an OpenAI-compatible model endpoint operated inside the customer environment. Codex remains the Agent engine; external models supply the reasoning and tool-call responses.

Under the standard On-Prem Enterprise scope, the customer supplies and operates the compatible AI endpoint, credentials, compute, storage, and monitoring. PacketSafari supplies model configuration, qualification, secure egress, and runtime integration.

Saving a provider does not provision a model server, fund provider usage, or open a new outbound network path.

Choose the next page

Supported models summarizes current PacketSafari test results.
Configure a custom model explains endpoint, parser, profile, role mapping, and qualification setup.
Egress approval covers access to another VM, customer subnet, or public provider.
Proxies and private CAs covers corporate proxy chaining and internal HTTPS certificates.
Troubleshooting provides a short diagnostic sequence.

Qualification boundary

Finding a model through Refresh models proves only discovery. A connection's Health → Test model sends one small inference request and does not qualify packet-tool or report behavior.

For models with Require passing validation, the exact provider, endpoint, model, server/parser profile, Codex version and model revision must pass the synthetic qualification before the permitted Agent stages run. Enable without validation is an explicit administrator access policy; it does not create a passing result or prove analysis quality.

Qualification makes three short synthetic calls and sends no packet capture. Results are reused while their fingerprint remains unchanged. Follow the provider onboarding and acceptance guide to set access, defaults, reasoning and deployment limits. Every customer-operated model still needs representative PCAP tests; hosted test results do not establish local compatibility, memory capacity or report quality.