Anonymized, verified, and checksum-pinned model weights designed for the Self-Alignment Framework appliance and autonomous agent nodes.
View Repository manifest.jsonPrimary reasoning and orchestration model for SAFi agents. Engineered with multi-token prediction speculative draft heads for ultra-fast generation and uncensored, vendor-neutral analysis.
High-throughput Mixture-of-Experts architecture optimized for rapid tool calling, shell operations, and Vulkan/RADV acceleration with minimal compute overhead.
Dedicated policy auditor and alignment guardian. Runs in an isolated runtime loop to supervise, evaluate, and verify tool commands and agent decisions against organizational constitution standards.