causa

causa™ is a reasoning orchestration platform for on-device inference that manages and runs AI models entirely offline. As foundation models become commoditized, fast and scalable AI deployment across edge devices is becoming essential. One-time purchase. Lifetime access. No subscriptions. causa™ is available for individual users, as well as for enterprises and educational institutions through bulk licensing for deployment at scale, per employee or device (via Apple Business Manager). Seamless, secure and at scale. Enjoy! Key Highlights • NEW (v1.3): Active Inference. causa™ can now reason over policies - at inference time! This enables patterns to be defined on the fly using regular expressions and allows multiple policies to be evaluated together to identify overlaps, inconsistencies, and compliance drivers in real time. Bringing policy reasoning, compliance, and adaptability together in a single runtime. - Cognitive Policies / Decision Rules. Reasoning over goals, beliefs, and uncertainty. - Behavior Policies. Task execution and sequencing — determining which actions to take to complete defined tasks. - Adaptive Policies. Learning from data and generalizing for control deployment and operational optimization. - Meta-Policies (Policies over Policies). Policy coordination, arbitration, and conflict resolution. • Adaptive Reasoning. causa™ enables seamless hot-swapping between Large Language Models (LLMs), allowing users to select the most suitable model for each task. One LLM may excel at natural language understanding, while another is fine-tuned for domain specific problems. Our model-agnostic orchestration system is capable of instantly switching between multiple models on-the-fly, ensuring the right one is used at the right moment. This allows to dynamically assemble the correct reasoning path from multiple experts (models) and validate it with Chain-of-Thought. • Prompt Engineering. Guide LLMs to think out loud! As LLMs grow in size and complexity, effective prompting is critical for achieving transparent, logical, and verifiable outputs. The Chain-of-Thought prompting guide enables users to articulate intermediate reasoning steps, by encouraging models to “show their work” mirroring human problem-solving. This approach not only enhances the accuracy of responses but also makes them more interpretable and trustworthy. • Built-in AI Safety. At Black Forest AI, creators of causa™, we strongly advocate for human-in-the-loop oversight in AGI systems. By keeping humans involved at critical stages, we minimize risks of uncontrolled autonomy or metacognition. This ensures AI behavior remains trustworthy, accurate, fair, transparent, and ethically aligned. Core to this framework are guarding mechanisms, allowing users to classify prompts, responses and policies within a structured safety risk taxonomy. • Trusted Execution Environment a. 100% Privacy by Design - causa™ runs entirely offline! No data or prompts are collected or transmitted, a user retains full ownership and control over data and deployed models. Native support for Apple Silicon ensures outstanding performance and reduced power consumption, making it super efficient. b. Smart Memory Management - To handle the significant memory footprint of LLMs, only selected model weights are downloaded and stored locally. Strict memory management prevents clogging, ensuring smooth operation. Depending on the device and available RAM, some models are intentionally disabled. For optimal experience, we recommend a device with 16GB RAM. • causa™·vision, featuring integrated Vision Language Models (VLMs), is currently in beta development. All causa™ license holders will be automatically granted access to this upcoming feature.