Orca Connect opens local AI models as shared APIs

OrcaRouter, developed by U.S. research institute Continuum AI and offered exclusively in Japan by FlashLabs, already aggregates more than 200 LLMs and generative AI models behind an OpenAI-compatible API. Orca Connect addresses a gap the company describes between running models locally — to keep data in-house, control costs or try open-weight models — and using cloud APIs: sharing a local model has typically required port forwarding, tunnel setup and NAT traversal, plus security concerns. By making a local runtime appear in the same catalog under a "local/machine-name/model-name" entry and reachable with the usual OrcaRouter API key and endpoint, the connector aims to let local and cloud models sit behind one interface.

Setup follows four steps: install from the admin screen, sign in with a command that approves the machine, check runtimes and models with a status command, and start serving. The connector verifies signed releases, and nothing is exposed until a model is selected for sharing. Responses processed on the local machine carry an X-Orca-Billed: false header and are recorded at $0, with billing applying only when a hosted model answers.

The design also extends to team use and security. Shared links carry expiration dates and request limits and can be revoked, and recipients sign in with their own accounts rather than receiving keys. Keys can be set to local-only so requests never reach a hosted model or incur charges. Connections pass through an encrypted relay, Zero Data Retention is available on all plans, and Tailscale or WireGuard can be used to bypass the relay for a fully private end-to-end setup, per OrcaRouter's official X account.