DoubleChecks

What was claimed

Introducing Locals: an Apple Silicon engine that runs local LLMs up to 3x faster... built on a theoretical breakthrough in complexity theory enabling substantially faster algorithms for DeltaNet architectures, grammar-constrained decoding, tool calling, and structured outputs

Our verdict

Needs caution

The available results identify other Apple Silicon LLM tools, including an MLX-based open-source toolkit, but do not establish a product named “Locals.”. The available sources contain no authoritative evidence for this claimed breakthrough or for its connection to DeltaNet, grammar-constrained decoding, tool calling, or structured outputs.

Combined assessment16 sourcesChecked Oct 22, 2026

Why this verdict

  • Locals runs local LLMs up to 3x faster.

    Can’t verify
    One assessment reached this conclusion

  • Locals is built on a theoretical breakthrough in complexity theory that enables substantially faster algorithms for DeltaNet architectures, grammar-constrained decoding, tool calling, and structured outputs.

    Can’t verify
    Three assessments reached this conclusion; one differed·Claude: Misleading

  • Locals is an Apple Silicon engine that runs local LLMs/agents up to 3x faster.

    Can’t verify
    Three assessments reached this conclusion; one differed·ChatGPT: Verified

Evidence reviewed

16 sources

Perplexity, Claude and ChatGPT sources are combined here.

github.comsitepoint.comnote.comcanitrun.devllmcheck.netllmhardware.iollmcheck.netgithub.comgithub.comperplexity.ai

Individual assessments

Assessment summary

  • The available sources support that Apple Silicon can run local LLMs, but they do not verify a product called “Locals” or the claimed theoretical breakthrough.
  • The “up to 3x faster” figure appears unsupported and may conflate unrelated benchmarks or implementations.github.com note.com perplexity.ai

Claims reviewed

Locals is an Apple Silicon engine that runs local LLMs.

Can’t verify
The available results identify other Apple Silicon LLM tools, including an MLX-based open-source toolkit, but do not establish a product named “Locals.”github.com github.com

Locals runs local LLMs up to 3x faster.

Can’t verify
A separate speculative-decoding report describes a 2–3x speedup under specific conditions, while another Apple Silicon engine reports 1.22x prefill and 1.32x decode improvements. Neither verifies the claim about Locals.note.com perplexity.ai

Locals is built on a theoretical breakthrough in complexity theory enabling substantially faster algorithms for DeltaNet architectures, grammar-constrained decoding, tool calling, and structured outputs.

Can’t verify
The available sources contain no authoritative evidence for this claimed breakthrough or for its connection to DeltaNet, grammar-constrained decoding, tool calling, or structured outputs.

Assessment summary

  • The product "Locals" does not appear to exist in any publicly available sources or announcements.
  • While the individual technical claims (DeltaNet complexity improvements, 3x speedups on Apple Silicon, grammar-constrained decoding) are plausible and grounded in real research, the core claim about a product called "Locals" cannot be verified and appears to be fabricated or fictional.

Claims reviewed

Built on a theoretical breakthrough in complexity theory enabling substantially faster algorithms for DeltaNet architectures

Misleading
DeltaNet is real and does have complexity improvements (O(n²d) to O(nd²)), but this is not a recent 'breakthrough' specific to 'Locals'—it's established research from 2024-2025 used in models like Qwen3.5.

This claim may mislead readers without added context.

Supports grammar-constrained decoding, tool calling, and structured outputs

Can’t verify
These are standard features in modern LLM inference engines, but cannot be verified for a non-existent product.

Locals runs local LLMs up to 3x faster

Can’t verify
While 3x speedups are documented for real Apple Silicon engines like MLX and Rapid-MLX, this claim cannot be verified for a non-existent product.

Assessment summary

  • The repository and announcement for "Locals" exist and explicitly claim "up to 3x faster" and Apple Silicon support github.com shadowgraph.io.
  • The stronger claim of a "theoretical breakthrough in complexity theory" enabling substantially faster algorithms for DeltaNet, grammar-constrained decoding, tool calling, and structured outputs is an author assertion without corroborating peer-reviewed theory or independent benchmarks, so it is not verifiable from public sources github.com shadowgraph.io.

Claims reviewed

Locals is built on a theoretical breakthrough in complexity theory that enables substantially faster algorithms for DeltaNet architectures, grammar-constrained decoding, tool calling, and structured outputs.

Can’t verify
The repository/announcement claims underlying "theoretical results in complexity theory," but I found no linked paper, formal proof, independent benchmark, or peer-reviewed source substantiating a general theoretical breakthrough for those algorithmic improvements github.com shadowgraph.io.

Locals is an Apple Silicon engine that runs local LLMs/agents up to 3x faster.

Verified
The project's GitHub README and related pages explicitly advertise an Apple Silicon inference engine and state "up to 3x faster" for running agents on a Mac github.com shadowgraph.io.

Check your own claim

Paste any statement, headline, or AI answer. Independent assessments check it and compare sources.

Share this result