Clients and providers

Development builds

Protocol support is integrated; there is no public installer yet. Specific client versions and tool capabilities still need release-specific validation.

Tools and models

LiteGate implements OpenAI Chat Completions, Responses and Anthropic Messages, including streaming and token-counting boundaries. Codex and Claude Code are client use cases; their compatibility depends on the protocol, enabled features and tested version, not their name alone.

The client sends requests to LiteGate, policy selects an eligible destination, and an external engine or cloud provider executes the model. Local engines must be installed separately. LiteGate stores endpoint metadata rather than user-model weights.

Access and context

Provider API credentials and permissions determine access. A paid chat plan is not an API credential. Subscription connections are a separate workstream with provider-specific authorization and delivery gates.

Messages and tool results can travel between compatible providers; unsupported conversion is reported. Private reasoning and internal caches are not portable. Fallback stops at the first visible response event.

For chat, LAN, credentials, subscription plans and privacy boundaries, see Product and status. Authorized development users can consult Configuration.