causa

causa™ enables rationale through reasoning orchestration. As foundation models evolve into commodities, rather than relying on the performance of a single Large Language Model (LLM), orchestrate causal reasoning with the logical consistency of multiple ones. Combined with Chain-of-Thought (CoT) prompting technique, one can achieve a true coherent reasoning. Tackle uncertainties and concept drift with rules of inference and a rational thought process.

This is a PERPETUAL license, not subscription! Users are free to purchase causa™ for personal use, while businesses and schools can purchase licenses in bulk and deploy them at scale per employee or per device (via Apple Business Manager). Seamless, secure and at scale. Enjoy!


Key Highlights

• Adaptive Reasoning: causa™ empowers you to assign specific reasoning tasks across self-hosted models following the challenge at hand. Common scenarios are counterfactual reasoning, dynamic problem-solving and data-driven analysis. Eliminating dependence on a single generalist brain for everything, allows you to dynamically assemble the correct reasoning path from multiple experts and validate it with CoT. Ultimately, this strategy aids in obtaining sovereignty from overbearing enterprise solutions and helps you to become fast and more accurate in context-aware decision making.

• Swap AI Brains: With this feature, you can effectively “swap out” the core thinking process as needed. 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 seamlessly hot-swapping between multiple models, ensuring the right one is used at the right moment. This introduces a new level of sophistication to workflows that require nuanced, multi-step reasoning.

• Built-in AI Safety: We at Black Forest AI (creators of causa™) strongly advocate for human-in-the-loop approach in AGI system's lifecycle. This means incorporating human involvement at various decision-making stages, basically eliminating, and constraining any potential for uncontrolled autonomous agency and metacognition. On par with safety measures, this is also to guarantee trustworthy behavior of AI algorithms that exhibit exceptional accuracy, fairness, transparency, and adherence to ethical standards.
The crucial first step is to incorporate guarding mechanism, allowing every user to robustly attain prompt/response classification within a safety risk taxonomy.

• Trusted Execution Environment
a. 100% Privacy by Design! causa™ operates entirely offline, utilizing on-device inference with neither data nor prompts are collected or transmitted. You retain full ownership and control over data and deployed models. Additionally, native support for Apple Silicon accomplishes outstanding low latency performance and reduced power consumption.
b. Implemented strict memory management prevents memory clogging. Given that LLMs have a significant memory footprint, only selected model weights are downloaded and stored offline on the device. To experience full potential of causa™, a device with 16GB of RAM is recommended.

• Dedicated Reasoners & VLMs (Roadmap 2025): Integrate custom reasoners to leverage hybrid search using in-house RAG and fine-tuned LLMs. Support for Vision Language Models (VLMs) is also planned for a phased rollout. The goal is to achieve always-on decision intelligence for error-free operations and excellence.


Note on Our Code of Conduct:
As we are committed to deliver the latest releases with in-demand feature sets and incorporating top-tier lightweight models that leverage extensive knowledge sources, we continuously seek feedback from organizations and advanced users regarding their deployment use cases, AI safety, and the practical application of foundation models in both multilingual and monolingual business environments.
Please, send your feature requests and feedback to hi@blackforest.ai