Languages

ContextIQ parses code in tiers. Run tokengraph langs to list every language and its tier for your install.

Deep parse (full graph) — 25+ languages

A full call / import / inheritance graph, with doc comments carried across:

Python (via the stdlib ast) and, via tree-sitter: Java · Go · TypeScript · JavaScript · C · C++ · C# · Rust · PHP · Ruby · Kotlin · Swift · Scala · Lua · Bash · Solidity · Perl · Erlang · Julia · R · Haskell · OCaml · Nim · PowerShell · Dart · SQL.

Every deep-parsed language falls back to regex when its tree-sitter grammar isn't installed.

SQL is parsed as schema, not as text

.sql / .ddl files produce a real graph rather than a list of table names:

Extracted	Becomes
CREATE TABLE / CREATE VIEW / CREATE INDEX / CREATE FUNCTION	symbols
every column_definition	a symbol scoped to its table (schema.orders.user_id)
REFERENCES, inline or as a named CONSTRAINT	a REFERENCES edge between tables
the tables a view or SQL function selects from	REFERENCES edges (lineage)
ALTER TABLE … REFERENCES in a migration	REFERENCES edges from that file to both tables
the comment above a statement	the symbol's doc comment, so semantic finds it

So impact schema.users answers "what breaks if this table changes" — the tables whose foreign keys point at it, the views that read it, and the migrations that touch it — the same way it does for a function.

Vue and Svelte components

.vue / .svelte are markup on the outside and TS/JS on the inside, so the <script> block is handed to the TypeScript grammar and everything else is blanked out — blanked rather than stripped, so every symbol keeps the line number it has in the real file. An SFC frontend therefore gets the same call graph as any other TypeScript, including calls that cross into your API client, at the cost of no extra grammar.

Regex index — 30+ more languages

Lightweight symbol indexing for: Elixir · Clojure · F# · Groovy · Zig · Crystal · Haxe · Objective-C · Visual Basic · Tcl · Pascal · GDScript, and markup/config — HTML, CSS, YAML, TOML, XML, INI, GraphQL, Terraform, Protobuf, Markdown, Dockerfile, and more.

When a grammar is missing

Deep-parse languages fall back to regex when their tree-sitter grammar isn't installed, and that fallback is silent in the worst way: definitions still appear, so the index looks healthy, but no call edges are produced — an empty get_callers then means "not extractable here", not "nothing calls this". index and doctor now say so directly, naming the languages and the file counts, and langs --repo shows the full split. The fix is one install:

pip install 'contextiq[langpack]'

Coverage is measured against code

langs --repo reports call-graph coverage as a share of code files. Markdown and YAML can never carry a call graph, so counting them would measure the repo's docs-to-code ratio rather than the extractor — a repo shipping 52 docs and 55 modules read as 49.5% covered when every one of its code files had a full graph. Both counts are shown, so nothing is hidden by the narrowing.

Exact references: SCIP

The tree-sitter tier resolves cross-file calls by name, which is best-effort by construction: a same-named method in another file can absorb an edge that belonged elsewhere. Where that precision matters — a large polyglot repo, a refactor whose blast radius has to be right — run your language's SCIP indexer and feed the result in:

scip-typescript index && tokengraph import-scip index.scip.json   # or scip-java / scip-go

That adds compiler-exact REFERENCES edges, which context expansion and impact use directly. The built-in resolver stays as the zero-dependency fallback, so this is an upgrade, never a prerequisite. doctor points it out when a meaningful share of your code is resolving by name and no SCIP index has been ingested.

Doc comments everywhere

Symbols carry their doc comments across every language — godoc, rustdoc, Javadoc, JSDoc, TSDoc, not just Python docstrings — so semantic search finds code by what it does, even when the identifier is opaque.

Check your install

tokengraph langs                    # languages by extraction tier
tokengraph diagnose-extractors      # self-test every extractor (CI gate)

Next steps

• Find code by meaning: Retrieval → semantic
• Whole-repo language mix: tokengraph arch in the CLI reference