Umbra

Umbra provides an innovative platform for accessing open-weight language models via an OpenAI-compatible API. Key features include: • Access to a diverse range of open-weight, community GGUF models • OpenAI and Anthropic API compatibility for seamless integration • Private-prompt inference with on-device decryption • Hosted by independent Apple Silicon providers • Pay-per-token usage with free starter credit This platform addresses the challenge of running useful open-weight models that often lack hosted API access. By leveraging a distributed network of Apple Silicon machines, Umbra makes these often-uncensored and specialized models readily available without the need to rent GPUs or maintain local infrastructure. Your existing client applications can connect by simply swapping the `base_url`, ensuring a smooth transition. Umbra prioritizes privacy and security, encrypting prompts so that machine owners cannot read them. Hardware attestation is actively deployed, and code identity is being rolled out, strengthening the integrity of the inference process. This design fosters trust and enables users to work with sensitive data without compromise. Ideal for developers, researchers, and content creators who need access to a broad spectrum of specialized or uncensored language models for diverse applications, from text generation to data analysis or creative writing, while maintaining strict data privacy.