Full context.
Before the agent answers.
Cartograph indexes repositories into a local knowledge graph of files, symbols, imports, calls, inheritance, communities, and execution flows.
Built for codebase exploration and security research: fast local indexing, ranked flow discovery, precise symbol context, plugin-fed datasets, and permissive MIT licensing.
curl -sSfL https://onixhdz.github.io/cartograph/install.sh | sh
Agents need system context, not isolated file reads.
Cartograph gives the agent a graph it can traverse: structure, dependencies, call paths, communities, processes, and search results grounded in the indexed repository. It ranks what matters so investigations start from relevant flows and important symbols, not a flat list of text matches.
- • Read files one at a time
- • Grep for symbol names without relationships or ranking
- • Guess which calls, imports, and files are actually important
- • Miss execution flows, communities, and relevance signals
- ✔ Files, symbols, imports, calls, and inheritance in one graph
- ✔ Hybrid BM25 + vector search when embeddings are enabled
- ✔ Ranked flows, communities, and impact context for focused answers
- ✔ Communities, process traces, file trees, and symbol impact connected in one local graph
What Cartograph connects.
KNOWLEDGE GRAPH
Maps folders, files, symbols, imports, calls, inheritance, communities, and execution processes.
206 LANGUAGES
Tree-sitter detection across 206 languages, with full extraction for 13 Tier 1 languages and inferred extraction for 56+ Tier 2 languages.
HYBRID RETRIEVAL
BM25 full-text plus optional vector search via reciprocal rank fusion for better semantic recall.
FLOW DETECTION
Discovers execution processes from entry points through call chains, then surfaces ranked flows instead of isolated matches.
SKILLS + MCP
Install Cartograph skills for AI coding agents, then expose the warm graph through MCP for Claude Code, Cursor, OpenCode, and other tools.
PLUGIN DATASETS
Embedded Go plugins can register external datasets directly through the public API and query them as local graph-backed datasets.
Purpose-built for exploration and security research.
Many code-intelligence tools stop at visualization, chat, or oversized tool catalogs. Cartograph focuses on the investigation loop that matters: index the repo, retrieve the right symbols, follow execution flows, understand impact, and keep the agent grounded in real code relationships.
Find where behavior lives with semantic query, file-tree views, and symbol context.
Follow ranked execution flows, callers, callees, processes, and related files to understand how behavior moves through the codebase.
Use plugins and graph context to widen investigations with external datasets, references, and domain-specific signals.
Keep graphs hot through the background service so multi-query research sessions avoid repeated startup and reload costs.
Index. Resolve. Query. Context.
Cartograph builds the graph in stages: walk the repository, parse symbols with tree-sitter, resolve imports, calls, inheritance, and references, cluster related symbols, trace execution processes, build BM25 plus optional vector search indexes, and serve the result to CLI and MCP clients.
cartograph │ ├─ structure ──► repositories, folders, files, ignore rules ├─ parse ──► tree-sitter symbols, imports, calls, heritage ├─ resolve ──► cross-file calls, imports, inheritance, references ├─ cluster ──► communities of related symbols (Leiden) ├─ processes ──► execution flows from entry points through call chains ├─ search ──► BM25 + optional vector retrieval with RRF ├─ plugins ──► external nodes, edges, and agent references └─ serve/mcp ──► warm graph cache, HTTP API, editor tools
Install it, then ask your agent.
Cartograph is meant to sit in the agent loop. Install the binary, install the skills, then ask your coding agent to index, inspect, and reason through the graph.
# or
$ curl -sSfL https://onixhdz.github.io/cartograph/install.sh | sh
Homebrew is recommended on macOS. The shell installer supports Linux and macOS, or check out the latest Releases .
# targeted install
$ cartograph skills install --agent claude,opencode
Adds Cartograph routing, prompts, and workflows to AI coding agents so they know when to use the graph.
then map the auth flow and impact
for validateUser."
The installed skill guides the agent to start the service when useful, index first, and use graph-aware commands like context, impact, and query.
Teach your agent how the codebase works.
Cartograph maps repository context into one local graph so agents and researchers can explore code structure, trace behavior, and investigate risk.