Pokee AI has unveiled Pokee-Isaac 28B, a significant new 28-billion parameter, text-only foundation model featuring an unprecedented 10-million token context window. What truly sets Isaac apart is its design philosophy: it's built to operate entirely within the customer's security perimeter, whether deployed in a Virtual Private Cloud (VPC), on-premises, or directly on-device. This addresses a critical challenge for long-horizon AI agents, which rapidly accumulate context but have historically relied on cloud endpoints, making them unsuitable for regulated industries, public sector institutions, and applications with strict data residency requirements.
This innovative approach allows organizations to leverage advanced AI agent capabilities without compromising data privacy or security. Pokee-Isaac is offered through a licensed, OpenAI-compatible developer API, enabling seamless integration into existing workflows. By keeping data within the customer boundary, the model empowers sensitive applications to harness the power of extensive context understanding and coherent reasoning, a capability previously almost exclusive to external cloud services.
For developers and researchers, Pokee-Isaac 28B represents a breakthrough in deploying powerful, long-context AI agents in data-sensitive environments. The model demonstrates robust performance, achieving 93.3% on RULER at its full 10M-token context and showing parity with strong cloud baselines on agentic benchmarks like BFCL v4 and τ³-bench. Crucially, it's engineered for efficiency, capable of running on a single GPU (such as an RTX 4090 or B200-class), with Day-0 support for vLLM and SGLang, making high-performance, in-boundary AI agents a practical reality.