Basis

Basis stands out for its commitment to open, non-commercial AI research at the reasoning frontier. Projects like Pact—a formal coordination language combining game theory, distributed systems, and cryptography—show genuine technical depth. ExoPredicator's use of abstract world models for robot planning is novel, and the 14-day agent-built compiler (despite incomplete proofs) demonstrates multi-agent orchestration. AutumnBench offers a rigorous benchmark for world model learning. Strengths include a clear mission, academic-quality publications, and a collaborative community through fellowships and partnerships. Weaknesses are the lack of any product, API, or customer support—this is a research organization, not a SaaS. Funding relies on donations, so long-term sustainability is uncertain. Where it fits: university labs, independent researchers in neuroAI/formal methods, and donors who value open science. Where it doesn't: any commercial application, quick prototyping, or user-facing AI tool. If you need a reasoning engine to integrate into a product, Basis is not ready—try DeepMind, OpenAI, or Anthropic instead.

Basis is a nonprofit AI research organization developing a universal reasoning engine to tackle complex challenges across domains like multi-agent coordination, robot planning, and formal verification. Unlike commercial labs, Basis prioritizes open research and societal impact—releasing papers, benchmarks, and open-source code rather than products. Recent work includes Pact (a coordination language for trustworthy multi-agent systems), ExoPredicator (robot planning via abstract world models), an unverified 93k-line JS-to-WASM compiler built by four agents in 14 days, and AutumnBench for world model learning. Basis offers fellowships, partnerships, and donation-based support. For researchers seeking foundational AI reasoning, Basis provides open publication and mission-driven collaboration, but does not offer production-ready APIs or consumer tools.