v0.1.0-beta.1

llmfp v0.1.0-beta.1

First public beta of the llmfp command-line tool.

Highlights

• OpenAI Responses and Anthropic Messages adapters
• One-time Interactsh/OAST callbacks with exact-path correlation
• Tool-infrastructure attribution against provider-published CIDR registries
• Versioned JSON report schema and documented exit-code contract
• Linux, macOS, and Windows archives for amd64 and arm64
• SHA-256 checksums, CycloneDX SBOM, third-party notices, and GitHub build provenance

Evidence boundary

A matching callback can attribute the infrastructure that executed the web fetch. It does not prove which provider generated the model response. A missing callback is inconclusive.

The default public oast.pro service can observe callback metadata. Do not put secrets in probe URLs or prompts.

Project documentation

• Evidence model
• Roadmap and fingerprint integration TODO
• Propose another fingerprint method

The next planned evidence layer evaluates the CC-BY-4.0 Zenodo dataset Single-token output distributions as behavioral fingerprints of large language models.