Crate avocado_core

Crate avocado_core 

Source
Expand description

AvocadoDB Core Engine

The deterministic context compiler that fixes RAG.

AvocadoDB replaces traditional vector databases’ chaotic “top-k” retrieval with deterministic, citation-backed context generation using span-based compilation.

§Key Features

  • 100% Deterministic: Same query → same context, every time
  • Citation-Backed: Every piece of context has exact line number citations
  • Token Efficient: 95%+ token budget utilization (vs 60-70% in RAG)
  • Fast: < 500ms for 8K token context

§Architecture

Query → Embed → [Semantic Search + Lexical Search] → Hybrid Fusion
      → MMR Diversification → Token Packing → Deterministic Sort → WorkingSet

Re-exports§

pub use types::Artifact;
pub use types::Citation;
pub use types::CompilerConfig;
pub use types::Error;
pub use types::Result;
pub use types::ScoredSpan;
pub use types::Span;
pub use types::WorkingSet;
pub use types::Session;
pub use types::Message;
pub use types::MessageRole;
pub use types::SessionWorkingSet;
pub use types::Manifest;
pub use types::ChunkingParams;
pub use types::IndexParams;
pub use types::ExplainPlan;
pub use types::ExplainCandidate;
pub use types::ExplainTiming;
pub use types::ExplainThresholds;
pub use types::IngestAction;
pub use types::GoldenQuery;
pub use types::EvalResult;
pub use types::EvalSummary;
pub use types::WorkingSetDiff;
pub use types::DiffEntry;
pub use types::RerankEntry;
pub use session::SessionManager;
pub use session::SessionReplay;
pub use session::SessionTurn;

Modules§

approx
compiler
Context compilation engine
db
Database operations using SQLite
diff
Working set diff module
embedding
Embedding generation with local and OpenAI support
eval
Evaluation module for retrieval quality metrics
index
In-memory vector index for similarity search
session
High-level session management API
span
Span extraction from documents
types
Core data types for AvocadoDB

Constants§

VERSION
The version of AvocadoDB