Skip to content

fialr

AI-native archival infrastructure. Local vector intelligence, cryptographic integrity, and semantic discovery for files that matter.
version 1.0 platform macOS / Linux inference local llm / vectors integrity BLAKE3-verified

fialr treats files as long-term digital assets with local AI at the foundation. It assigns cryptographic identity, classifies sensitivity, generates vector embeddings for semantic discovery, and structures your corpus into an adaptive, machine-native archive.

Every file is enriched by local AI that learns from your corpus. Every operation is logged to an append-only ledger. Every file is hash-verified before and after modification. Nothing moves without a dry-run review first.

Content hash as identity

BLAKE3 hash is the stable identifier. Filenames and paths are mutable metadata. Dedup, rename, and reorganization stay coherent regardless of path changes.

Local AI, adaptive corpus

Enrichment and vector embeddings run locally via Ollama. Each file indexed improves semantic search, deduplication, and metadata quality across the entire corpus. Cloud providers available as opt-in for non-restricted files.

Safety by default

Every module has dry-run mode on by default. Execution requires explicit confirmation. Jobs generate a pre-flight manifest before touching anything.

Infrastructure, not organizer

A structured, vector-indexed corpus is a reliable input to future applications: AI agents, semantic search, automation pipelines, and tools that do not yet exist.

fialr /FYL-er/ — from filar, Latin filum (thread).

A filar micrometer is a precision instrument used in astronomy and surveying: fine threads stretched across a lens, positioned to measure the exact location of objects in a field of view. The observer does not move the stars. The instrument provides the coordinate system that makes them addressable.

fialr does the same for files. It does not decide what your files mean or where they belong. It provides the infrastructure — cryptographic identity, sensitivity classification, structural schema — that makes a corpus addressable, verifiable, and machine-native. The threads run through the collection without altering it.

The transposed vowel is deliberate. In a system where content hash is identity and filenames are mutable metadata, the surface representation of a name is the least stable thing about it. The word shifts; the referent remains.