hirn-core

⚠️ Experimental: This project is under active development. APIs, on-disk formats, and behaviour may change without notice. Not recommended for production use.

Core types, traits, configuration, and error definitions for the hirn cognitive memory database. This is the leaf crate — all other hirn crates depend on it, but it depends on none.

Key Types

Type	Backing	Copy?	Description
`MemoryId`	ULID	Yes	128-bit universally unique memory identifier
`Namespace`	interned `u32`	Yes	Column-level memory isolation (pre-interns `"default"`, `"shared"`)
`AgentId`	interned `u32`	Yes	Agent identity (pre-interns `"system"`)
`Timestamp`	`DateTime<Utc>`	Yes	UTC timestamp with chrono backing
`Layer`	enum	Yes	Memory tier: Working, Episodic, Semantic, Procedural
`EdgeRelation`	enum	Yes	Graph edge types with `is_bidirectional()`

Configuration

HirnConfig — 40+ parameters controlling the entire cognitive pipeline:

let config = HirnConfig::builder()
    .db_path("./brain")
    .embedding_dimensions(768)
    .rpe_fast_path_threshold(0.3)
    .quality_gate_threshold(0.5)
    .build()?;

Builder validation at .build() enforces invariants (threshold ranges, template placeholders).

Traits

Embedder — Embeds text → Vec<f32> (sync + async)
LlmProvider — LLM completion for consolidation, causal discovery
EntityExtractor — Named entity extraction from text
McfaAuditSink — Security audit reporting interface

Error Handling

HirnError — #[non_exhaustive] thiserror enum. All crates define their own error types with From impls into HirnError.

Utilities