Zenith Logic Foundry

zlf

Zenith Logic Foundry

Forge Grounded Wisdom, Mapped To You.

解答世间万物

zlf is a WAM-backed Prolog graph database for building AI-native knowledge bases. It stores graph facts persistently, evaluates queries and rules through a single Prolog runtime, and supports retrieval over the resulting knowledge graph.

What it provides

• Persistent property-graph storage backed by RocksDB
• Prolog facts, rules, and queries executed by a single WAM runtime
• Durable logical facts (arity >= 3) and stratified negation with tabled negation
• BM25 retrieval, optional vector similarity, temporal queries, and code retrieval
• Tree-sitter code indexing: 10 languages, zero-config discovery, canonical code graph, bounded traversal, BM25 symbol search, CodeQuery AST (Prolog + JSON/HTTP)
• JSON-over-STDIO CLI, HTTP/SSE server, and interactive Prolog REPL
• Compiled-rule persistence and storage-backed fact writes

Quick start

Build the CLI:

cargo build --release

Open a database in the Prolog REPL:

target/release/zlf repl ./zlf-db

Documentation

The usage guide is the directory page for the docs:

• Build and configuration
• Code indexing — Tree-sitter ingestion, canonical code graph, BM25 symbol search, CodeQuery predicates, benchmark
• Prolog: basics, storage & fact store, tabling & negation
• Database: property-graph model, retrieval (BM25, vector, temporal, code)
• IndexProfile
• Interfaces: REPL, JSON-over-STDIO, HTTP server
• Internals: WAM runtime, RocksDB layout, index algorithms

Track and design records live under docs/track/; outdated material lives under docs/archive/.

Architecture

CLI / HTTP server / REPL
  -> WAM Prolog runtime
  -> storage-backed facts, compiled rules, and indexes
  -> RocksDB

zlf-codeindex (code indexing)
  -> Tree-sitter parsers (10 languages)
  -> discovery / blob store / lifecycle
  -> canonical code graph (provenance, certainty)
  -> bounded traversal + BM25 symbol search
  -> CodeQuery AST (Prolog + JSON/HTTP)

Development

cargo fmt --all
python3 scripts/check-rust-size.py
cargo clippy --workspace --all-targets -- -D warnings -W clippy::too_many_lines
cargo test --workspace

License

MIT OR Apache-2.0