Six dimensions, four products. Numbers below are from sverklo-bench v2 (2026-05-20) using each project's public configuration. The "win" column is per-dimension. Sverklo doesn't win every column on purpose — the rows where competitors win are part of the point.
| Dimension | Sverklo | Serena | GitNexus | Greptile |
|---|---|---|---|---|
| License | MIT | MIT | BSL (restricted commercial) | Closed source |
| Where the code runs | Local-first, never leaves the machine | Local-first | Local + optional cloud | Cloud-only (uploads source) |
| Audit-gap detection (sverklo-bench v2) | 89% | 62% | 71% | 78% |
| Languages indexed | 4 (TS, JS, Python, Go) | 12 (incl. Rust, Java, C++) | 8 | 12+ via cloud parser |
| Install steps (zero-to-MCP) | 1 (npm i -g sverklo && sverklo init) |
3 (clone, build, configure) | 4 (clone, build, license-key, configure) | 2 (signup, install IDE plugin) |
| Cost | Free, MIT | Free, MIT | Free under BSL terms | $30/seat/month |
Each comparison links to a longer page with full bench-row reproduction commands, the specific use cases where the product wins, and the trade-offs.
If you need a language sverklo doesn't index (Rust, Java, C++, Ruby, Swift), pick Serena. If you want a fully managed service with no local install, pick Greptile. The MIT + local-first + zero-config combination is sverklo's wedge; everything else is conscious trade-off.
Open a PR on github.com/sverklo/sverklo with a reproduction. If the number on this page is wrong, we'd rather know. Try sverklo.