repomap

repomapGive your agent the map before it edits.

A map of your codebase for AI agents: code graph, search, call paths and change impact. No API key.

Get started
Live demo
GitHub
npx -y @sylphx/repomap setup
excalidraw: 687 files, 5,001 symbols and 9,639 resolved calls, indexed in under half a second. Try it live →

map

Modules found from real dependencies, the most central files (PageRank), key symbols and entry points. Zoom into any directory for an outline with line numbers.

search

Symbol names, BM25 and a local code embedding model over whole functions, methods and classes. Ask in plain words or by name. Returns file:line ranges and the matching lines.

context

One call for a symbol's code, its callers with call sites, callees, subtypes, members and the tests that reach it.

trace

The shortest call path between two symbols, each hop cited file:line, or the call tree above or below one.

impact

Blast radius before you edit, or of your current git diff. Callers by depth, importing files, modules touched, tests to run, risk level.

Fast and local

Parallel Rust indexer with a per-file cache and an in-memory graph that refreshes as you edit. Nothing leaves your machine.

One issue-text search finds every fix file in the top 10 for 83.0% of SWE-bench Verified issues (semble: 71.0%). File retrieval, not issue resolution.

SWE-bench Verified file localization500 issues, one issue-text search each. Acc@k counts issues where every gold fix file is among the top k files. repomap: Acc@1 48.8%, Acc@5 74.8%, Acc@10 83.0%. semble: Acc@1 34.0%, Acc@5 61.6%, Acc@10 71.0%. BM25: Acc@1 20.8%, Acc@5 45.4%, Acc@10 55.8%. Higher is better. This measures file retrieval, not issue resolution.Find the files behind an issueSWE-bench Verified · 500 issues · one searchrepomapsembleBM250255075100Acc@1repomap Acc@1: 48.8%repomap48.8%semble Acc@1: 34.0%semble34.0%BM25 Acc@1: 20.8%BM2520.8%Acc@5repomap Acc@5: 74.8%repomap74.8%semble Acc@5: 61.6%semble61.6%BM25 Acc@5: 45.4%BM2545.4%Acc@10repomap Acc@10: 83.0%repomap83.0%semble Acc@10: 71.0%semble71.0%BM25 Acc@10: 55.8%BM2555.8%Issues with every fix file in the top k (%)
Chart values and evaluation protocol
SWE-bench Verified: every gold fix file in the top k files
MethodAcc@1Acc@5Acc@10
repomap48.8%74.8%83.0%
semble34.0%61.6%71.0%
BM2520.8%45.4%55.8%

Defaults selected on 300 disjoint SWE-bench test instances, excluding every Verified ID, then frozen before one evaluation on all 500 Verified instances. One search per issue, at its base commit; gold files are used only for scoring. All three methods scored 500/500, with zero errors.

Split definitions, pinned revisions and full results · Committed chart data

Trade-off: on semble's public code-search set, NDCG@10 is 0.845 versus semble's 0.851; semble wins there. This is not a claim of better search on every task. Measured search quality and limitations.