Supermodel maps every file, function, and call relationship in your repo and writes .graph files next to your source. Agents read them automatically — 40%+ fewer tokens, no prompt changes.
Every time an agent needs to understand your codebase, it reads files — in a large repo, that's thousands of tokens per task just for orientation. Run supermodel once and it writes .graph files next to every source file. Any agent that can read files picks them up automatically. No configuration.
"In a monorepo spanning TypeScript, Python, and Go, an agent asked to trace a data flow was reading 47 files to orient itself. With Supermodel, it reads the .graph file — same answer, 40%+ fewer tokens."
Your agent thinks in files. Your codebase thinks in dependencies. Supermodel normalizes every language into one dependency graph — so cross-language edits don't break, and your agent knows exactly which files to touch.
"'Add a created_at field to the user model' — a one-liner data change that touches 11 files across 4 languages. The graph tells your agent exactly which ones."
Static linters read files. Supermodel reads your codebase as a graph of what calls what, across every language. When your agent needs to remove a function or rename a type, it has ground truth — not guesses.
"Deleting a 'dead' utility caused a silent runtime failure three services away. The call graph caught it before the PR."
Install the CLI and your codebase gets a persistent graph. Any agent that reads files can use it — no config, no workflow changes, no new tools to learn.
Run supermodel in your repo. It writes .graph files next to every source file and watches for changes. Any agent reads them automatically via grep and cat.
Full API access via typed SDK. Integrate graph queries directly into your agent workflows.
Language-agnostic HTTP API. Use from CI, scripts, or any custom agent.
Memory systems rely on agents subjectively selecting what to remember. A code graph is an external structural model of your system, like a blueprint or a map. It's incrementally updated as code changes, accurate because it's derived directly from your source, and useful across agents and sessions without asking any agent to decide what was worth remembering.
No. We process your code to build the graph and then immediately remove it. Your source code is not retained after processing. What we keep is the structural graph: function signatures, call relationships, dependency edges, domain classifications, and metadata.
No. Because your source code is removed immediately after processing, it is never available to train on. We only retain the structural graph, not the code itself.
TypeScript, JavaScript, Python, Java, Go, Ruby, C++, Kotlin, Swift, Rust (experimental), Elixir (experimental), and more. 10+ languages, one API.
The graph updates incrementally. Each file change triggers a partial update, only re-processing changed files and their affected relationships, not a full rebuild. Agents always query an accurate model of the current state of your code.
Install our CLI. It drops map files next to your code. Any agent that reads files can use them. Full docs at docs.supermodeltools.com.
Static analysis tools are built for specific checks at a point in time. Supermodel builds a persistent, queryable multi-layer graph combining call relationships, dependency edges, domain classifications, and parse structure. The graph is designed to be queried by agents at runtime, stays current as code changes, and works across languages and repositories.
If your business needs special compliance, send us a message at [email protected].