Reference
Source library
Every source cited on this site, in one place: official pages, license files, model cards, repositories, papers, and filings. Each entry links to the original and lists the records that cite it. Links go to the original publisher; documents are not copied here.
Key documents
Annotated primary documents from the subject hubs.
- Model Context Protocol specification (revision 2026-07-28) (external site: modelcontextprotocol.io)The normative MCP specification, including the consent, privacy, and tool-safety principles that implementers are expected to follow.
- Agent2Agent (A2A) Protocol README (external site: raw.githubusercontent.com)The A2A project's own summary of agent discovery through Agent Cards and its Linux Foundation stewardship and Apache 2.0 license.
- AGENTS.md README (external site: raw.githubusercontent.com)Shows what an AGENTS.md file contains and why it is a convention rather than software.
- AAIF projects (external site: aaif.io)The Agentic AI Foundation's list of hosted projects, useful for checking which agent projects are under neutral stewardship.
- OpenAI Agents SDK README (external site: raw.githubusercontent.com)A concise statement of the parts of an agent (instructions, tools, guardrails, handoffs, tracing) as one SDK defines them.
- Codex CLI README (external site: raw.githubusercontent.com)Shows that a locally running, open-source agent can still require a hosted model account or API key.
- goose README (external site: raw.githubusercontent.com)Describes an agent that can use different model providers and that is now part of the Agentic AI Foundation.
- AI can now model and design the genetic code for all domains of life with Evo 2 (external site: arcinstitute.org)Arc Institute's announcement of Evo 2, which states what was shared openly and which pathogens were excluded from the training data.
- Model card for esm3-sm-open-v1 (external site: huggingface.co)The ESM3 open model card, with its training-data removals, output filtering, and license statement.
- Boltz README (external site: raw.githubusercontent.com)The Boltz repository's statement that its code and weights are MIT-licensed for academic and commercial use, and its optional automatic MSA generation through an MSA server.
- Chai-1 README (external site: raw.githubusercontent.com)Chai-1's current license statement for code and weights, and its hardware requirements.
- Profluent-E1 Attribution Guidelines (external site: raw.githubusercontent.com)Profluent-E1's attribution guidelines, an example of conditions that go beyond a standard open-source license.
- ether0 model card (external site: huggingface.co)The ether0 card, which records the base model, training stages, and refusal training.
- MedGemma 1 model card (external site: developers.google.com)MedGemma's intended-use and limitations sections, which state what its outputs should not be used for.
- PaperQA2 README (external site: raw.githubusercontent.com)Shows how an open-source science agent depends on a separately configured model service.
- Glean Technologies, Inc. Terms of Service (June 5, 2026) (external site: assets.glean.com)An enterprise AI platform's terms, which describe permission-preserving connections and say that third-party models are governed by their own terms.
- Snowflake Cortex AI Functions (external site: docs.snowflake.com)Snowflake's documentation of which developers' models its AI functions use and where those models are deployed.
- Databricks Foundation Model APIs (external site: docs.databricks.com)Databricks' description of the open models it hosts behind its Foundation Model APIs.
- Model retirement policy (Databricks documentation) (external site: docs.databricks.com)A data platform's lifecycle policy for hosted models, with deprecation notice periods and past retirement dates.
- Models (external site: writer.com)Shows a platform that defaults to its own model while letting administrators enable models from other providers.
- Granite-4.2-8B model card (external site: huggingface.co)The model card for an open release from an enterprise software company, with its license statement.
- Palmyra-mini model card (external site: huggingface.co)A card that names its base model, an example of provenance to check before relying on a release.
- General MLPerf Submission Rules (external site: raw.githubusercontent.com)MLPerf's submission rules, including peer review between submitters and the requirement that submitted code can reproduce the results.
- ARC Prize Verified Testing Policy (external site: arcprize.org)ARC Prize Foundation's testing policy, which separates verified results from community submissions and discusses leakage through external APIs.
- Continuous Benchmarks (external site: tbench.ai)Explains how a benchmark can be versioned like software, and which changes require re-running or only re-grading.
- HLE rolling changes log (external site: raw.githubusercontent.com)A dated log of question changes, which shows that a benchmark's contents can differ between two results.
- Maintenance Mode Policy (HELM documentation) (external site: crfm-helm.readthedocs.io)What maintenance mode means for a harness whose tests depend on external model APIs.
- Language Model Evaluation Harness README (external site: raw.githubusercontent.com)The harness's own description of supported model types, and of which task types need log probabilities.
- Methodology (external site: unitedstatesofamericasuperintelligence.com)USASI's source standards, including its rule against publishing benchmark results.
- NVIDIA Corporation Form 10-K for the fiscal year ended January 25, 2026 (external site: sec.gov)Separates a chip designer's principal executive office from a manufacturing chain that runs through outside foundries and suppliers.
- CoreWeave, Inc. Form 10-K for the fiscal year ended December 31, 2025 (external site: sec.gov)Gives a cloud provider's principal executive office separately from its description of leased data centers in several countries.
- High-bandwidth memory (HBM) (external site: micron.com)A memory maker's own explanation of high-bandwidth memory and the workloads it is built for.
- AI networking (external site: arista.com)Shows the networking layer that connects accelerators in a cluster, as described by a switch vendor.
- Etched (external site: etched.com)An inference-hardware designer's statement that its silicon is fabricated by TSMC.
- MLPerf Inference: Datacenter (external site: mlcommons.org)Explains what MLPerf Inference measures, its closed and open divisions, and that results belong to specific submitted systems.
- Models overview (external site: platform.claude.com)A hosted developer's own model list, with a model ID for each platform the models are offered on and a retirement commitment for each model.
- Model deprecations (external site: platform.claude.com)Defines the active, legacy, deprecated, and retired stages for hosted models and the notice given before retirement.
- Model card for Olmo 3.1 32B Think (external site: huggingface.co)Shows a family split into many released checkpoints, with the training data and the license and intended-use statements for one release.
- gpt-oss-120b model card (external site: huggingface.co)The model card for an open-weight release from a developer that also runs a hosted API.
- Llama 4 Community License Agreement (external site: raw.githubusercontent.com)A custom model license with attribution, naming, and acceptable-use conditions that go beyond a standard open-source license.
- xAI Community License Agreement (Grok 2) (external site: huggingface.co)A community license that conditions commercial use and restricts training other models on the materials or their outputs.
- Gemma 4 license (external site: ai.google.dev)Google's statement of the license for Gemma 4, which is Apache 2.0.
- PyTorch Foundation Participation Agreement and Charter (external site: cdn.platform.linuxfoundation.org)The PyTorch Foundation's participation agreement and charter, which define it as a directed fund of the Linux Foundation and leave technical oversight to each project's own charter.
- The Agentic AI Foundation Charter (amended July 29, 2026) (external site: raw.githubusercontent.com)The Agentic AI Foundation's charter, covering its governing board, project steering committees, and where project trademarks are held.
- The LF AI & Data Foundation Charter (external site: lfaidata.foundation)The LF AI & Data charter, which uses the same directed-fund structure for AI, machine learning, and data projects.
- Business leagues (external site: irs.gov)The IRS description of a 501(c)(6) business league, the tax category the Linux Foundation gives for itself.
- Exempt purposes - Internal Revenue Code Section 501(c)(3) (external site: irs.gov)The IRS list of 501(c)(3) exempt purposes, for comparison with business leagues.
- Model Context Protocol LICENSE (external site: raw.githubusercontent.com)A hosted project's license file that changes on its own schedule, independent of the foundation's tax status.
- About EleutherAI (external site: eleuther.ai)A nonprofit research lab's own description of what it trains and releases.
- openpi README (external site: raw.githubusercontent.com)Physical Intelligence's description of its released checkpoints and its plain warning that they may not work on other robots.
- GR00T-N1.7-3B model card (external site: huggingface.co)The GR00T N1.7 card, with its weights license, how it was evaluated in simulation and on real robots, and who is responsible for safe integration.
- Isaac GR00T README (external site: raw.githubusercontent.com)The Isaac GR00T repository, whose license section separates code terms from weights terms.
- Cosmos3-Nano model card (external site: huggingface.co)A world-model card whose limitations section says what its outputs should not be used for.
- Isaac Lab README (external site: raw.githubusercontent.com)Isaac Lab's license section, which notes that full workflows depend on Isaac Sim and its proprietary components.
- Waymo Dataset License Agreement for Non-Commercial Use (external site: waymo.com)The Waymo Open Dataset license, which limits use to non-commercial purposes and excludes production systems.
- Digit 5 (external site: agilityrobotics.com)A humanoid product page with its own disclaimers about features still in development and remaining operational risk.
- llama.cpp README (external site: raw.githubusercontent.com)The project's own list of hardware backends and quantization formats. Use it to check whether your machine is supported before downloading a model.
- Cloud (Ollama documentation) (external site: docs.ollama.com)Explains which Ollama models run on Ollama's servers instead of your machine, and how to keep to local models only.
- llamafile README (external site: raw.githubusercontent.com)Describes the single-file packaging approach and its basis in llama.cpp and Cosmopolitan Libc.
- MLX README (external site: raw.githubusercontent.com)Apple's description of MLX and its unified memory model, which matters for how models use memory on Apple silicon.
- ExecuTorch README (external site: raw.githubusercontent.com)Shows the export-then-run workflow for phones and embedded devices, and warns that accuracy, memory use, and performance must be validated on the target device.
- AWQ: Activation-aware Weight Quantization (README) (external site: raw.githubusercontent.com)The authors' own description of the AWQ weight-quantization method, with the code and 4-bit kernels that implement it.
- gpt-oss-20b model card (external site: huggingface.co)A model card that ties a memory figure to a stated quantization and names local runtimes, a useful example of what to look for in a card.
- Terms of Use (external site: commoncrawl.org)Common Crawl's Terms of Use, which put the risk and the duty to respect third-party rights on whoever uses the crawled content.
- FineWeb dataset card (external site: huggingface.co)A dataset card that states its own license, the upstream terms that also apply, and its handling of personal information.
- Dolma dataset card (external site: huggingface.co)Dolma's card, including its license change and the statement that the original sources' terms still apply.
- RedPajama-Data-V2 dataset card (external site: huggingface.co)Shows a dataset released with quality signals and duplicate IDs rather than a single filtered version, with separate terms for data and code.
- NVIDIA Data Agreement for Model Training (external site: huggingface.co)An example of a dataset agreement that limits use to model training and disclaims any grant of rights to copyrighted content.
- The Common Pile v0.1: An 8TB Dataset of Public Domain and Openly Licensed Text (arXiv 2506.05209) (external site: arxiv.org)Explains what "openly licensed" text means and why unlicensed training data limits sharing datasets.
- CCBot (external site: commoncrawl.org)How website owners can identify Common Crawl's crawler and block it in robots.txt.
All cited sources
Support Us
Help keep USASI useful.
Optional. No USASI account required. Payment takes place on the linked provider’s website (Buy Me a Coffee).
About supporting this project