Ripwire maps codebases for AI agents without a cloud index

What this is about

Ripwire is an open-source tool that gives a codebase a machine-readable map. Its goal is to stop coding agents from reading large files, search-result lists, and documentation before they can understand a change. The project runs locally as a single command-line program and can alternatively expose its functions through an MCP server.

Version 0.6.0 was released on September 11, 2026. It adds Kotlin and Dart, improves analysis of Ruby projects, and speeds up several operations on very large repositories. These changes matter in practice because agents consume substantial context when entering an unfamiliar codebase and can easily miss dependencies.

What Ripwire actually does

Ripwire scans a repository, recognizes symbols, and builds a relevance-ranked call and dependency graph. A query such as ripwire . --for="change the session check" is intended to surface the affected locations. Other commands show callers, uses, likely blast radius, relevant tests, or the context around a Git diff. Agent integrations include Codex, Claude Code, Cursor, Windsurf, Gemini, OpenCode, and Aider.

The tool works without an API key, embeddings, a database server, or a persistent daemon. A cache speeds up later runs. According to the project, output identifies truncation, uncertain mappings, and declined resolutions rather than presenting an incomplete answer as complete. Version 0.6.0 expands support to 24 grammars and connects calls across the Kotlin and Java language boundary.

The project's measurements are documented with unusual openness. Ripwire publishes commands, raw evidence, counterexamples, and cases where an approach lost. However, these figures come from the project itself and are not an independent certification.

Why it matters

A coding agent pays for every unnecessary file it reads through time, context-window space, and often API cost. A local map can therefore be useful in large monorepositories, parallel-agent setups, or security reviews. Ripwire also tries to connect two phases of work: locating the relevant code before a change and checking likely tests and neighboring areas afterward.

Local operation is another advantage. Source code does not have to be transferred to an additional search service for indexing. That does not remove the privacy risk of the coding agent being used, but it reduces the number of external data recipients. The Apache 2.0 license lets teams inspect, adapt, and integrate the tool into their own workflows.

In plain language

Imagine entering an unfamiliar city without a street map. An agent could walk every road and ask people for directions. Ripwire is more like a local map marking important intersections, connections, and possible detours. The map does not replace looking at the real street, but it helps you look in the right place first.

A practical example

A team wants to add a second factor to authentication in a repository containing 8,000 files. The coding agent starts with a Ripwire query about session validation. The output names five central symbols, three direct callers, and six test files likely to matter after the change.

The developer asks the agent to read those locations first instead of opening 40 search hits in full. After the patch, Ripwire checks the Git diff again and flags one caller that was not updated. The six suggested tests and the normal CI pipeline then run. This example is fictional; it illustrates a sensible use as a navigation and checking aid, not a replacement for tests or review.

Scope and limits

• The call graph is partly name-based. Dynamic dispatch, macros, and callbacks can hide connections or make them ambiguous.
• Published speed and token comparisons are project-run measurements. Teams should repeat them with their own language, repository, and agent.
• Ripwire does not automatically understand business logic. A structurally relevant location can still be functionally wrong, so human review, compilers, tests, and security checks remain necessary.
• Prebuilt x86-64 binaries for version 0.6.0 require an x86-64-v3 CPU. Older systems need a suitable source build or another environment.
• The first run on a repository is a cold analysis. Very large or unusually structured projects may still take noticeable time.

The next sensible test is small: install Ripwire in a familiar repository, prepare one real change with --for, and compare the listed files and tests with your own knowledge. Only add it to an automated agent workflow once the results are reliable.

SEO & GEO keywords

Ripwire, coding agents, code graph, code analysis, MCP server, local developer tool, repository navigation, change impact, test selection, Codex, Claude Code, open source