Loctree Architecture

Doctrine

Loctree is the structural perception and context-compiler layer for agentic software work. It tells agents what the code is, how it is connected, what runtime surface it participates in, and what context is safe to act on now. The parser is treated as a sensor, not the product; ground truth and derived views are kept explicit; and the agent context pack is the core UX.

Layer 1 — Sensors

OXC (JS/TS)
regex parsers (shell, makefile, zig, dart, go, css)
planned: tree-sitter for selected languages (Cut 2.5, conditional)
artifact metadata: source maps, gitignore, repo identity

Layer 2 — Symbol Model

types.rs::FileAnalysis
export/import edges
node ranges + sensor tags

Layer 3 — Runtime Semantics

See docs/semantic-spec.md for idiom catalog.

ShellSemantics (Cut 3A)
MakeSemantics (Cut 3A)
planned: PythonRuntimeSemantics, RustRuntimeSemantics, TauriSemantics (Cut 3B)
contract: trait RuntimeSemanticAnalyzer, types SemanticFacts/IdiomTag/etc.

Layer 4 — Agent Context Pack

planned: loct context (Cut 4)
composition: structural + runtime + risk + action + memory + authority slices
planned: --with-aicx overlay (Cut 5)

Cross-cutting

Snapshot identity & rebuildability (Cut 2)
Doctor surfaces (Cut 2)
Reports / Leptos cockpit (Lane 4)
Note: AICX integration depends on stable interfaces; live memex-sync drift must be resolved before use.

Architecture