Everything Synaptic does.

Parses 30+ languages with tree-sitter, resolves call, import, and reference edges, and clusters the result into communities so the structurally important pieces surface on their own. A code-only corpus never makes a network call.

Incremental rebuilds, file watching, and git hooks keep the graph current as you work; an on-disk AST cache makes re-runs fast.

A text query returns a relevance-ranked subgraph (each node scored, best-first). path finds the shortest route between two symbols; explain shows a node and its neighbours.

SYNQL matches on kind, visibility, LOC, fan-in/out and variable-length paths with count(...) aggregation, --explain, saved queries, and a library of named patterns: singleton, factory, observer, service-locator, god-class. Not text search.

affected reports everything that transitively depends on a symbol. predict forecasts a change's blast radius, public APIs at risk, at-risk tests, new cycles, and a risk score, before you write the edit.

speculate then applies the change in a throwaway git worktree and actually runs the at-risk tests plus a build and type-check, reporting real pass/fail: the ground-truth half of prediction.

Event buses (Node EventEmitter, DOM CustomEvent, C# events) and Electron IPC link a publisher to its subscriber through a channel node, so a handler reached only across the bus is not a phantom 0-caller.

Reflection and dispatch that can't be resolved statically (by-name lookups, dispatch tables, eval, dynamic import, .NET/Python/JVM reflection) are cataloged, and affected attaches a caveat when a symbol is reachable only dynamically.

Each refactor emits a confidence-scored execution plan (plan.json + plan.md) with the exact edit sites, for an AI agent to apply.

Afterwards, refactor verify rebuilds and checks the graph held: the definition moved or renamed, no references were lost, and no new cycles were introduced.

Reports added and removed dependencies, removed public APIs, architectural drift, new cycles, and hotspots between two revisions (or --since <date>), as a Markdown or self-contained HTML report.

Built graphs are cached per commit SHA, so a repeat diff returns immediately and only the working-tree side is rebuilt.

Flags row-level-security gaps, over-broad grants, likely SQL injection, missing indexes on filter and foreign-key columns, SELECT *, non-sargable predicates, N+1 patterns, and missing primary keys.

sql advise --query "<sql>" critiques a candidate query before you write it, cross-referenced against the graph's tables, indexes, and RLS policies.

A workspace federates many repositories into one graph with real cross-repo edge resolution (export surfaces plus import, tsconfig paths, and module-federation aliases) so architecture stays visible across repo boundaries.

The cross-repo global store and merge-graphs compose several graphs into one namespaced view.

Over stdio or HTTP (protocol 2025-11-25): graph navigation, reverse impact, dynamic-dispatch hazards, PR and working-tree blast radius, change forecasting, predictive test selection, structural search, time-travel diff, plan-only rename, and SQL audit/advise.

Plus MCP prompts, argument completions, resource templates and subscriptions, and a small REST surface. synaptic install wires it into Claude or Codex.

One run writes graph.json, a Markdown GRAPH_REPORT.md (god nodes, surprising connections, import cycles), an interactive 2D explorer, a 3D force graph, and a static SVG layout.

Re-emit GraphML, Cypher, DOT, an Obsidian vault, or a Markdown wiki from an existing graph, or push live to Neo4j / FalkorDB.