Replay a recorded run of the same task done twice — once with SourceIndex, once without — and watch both agents work, side by side, on real code.
Loading the live comparison…
The four runs replayed above, on Claude Code with Opus 4.6: the same task with and without SourceIndex.
| Example · task | Cost | Cost saving | Minutes | Time saving |
|---|---|---|---|---|
| beets Add Discogs extra tag filters | $1.69→$0.47 | 72% | 5.2→1.3 | 75% |
| Faker Add Algerian company providers | $1.12→$0.71 | 37% | 3.2→1.8 | 46% |
| Jinja Support namespaces in tuple assignment | $2.23→$0.89 | 60% | 7.3→4.9 | 32% |
| Requests Correct UTF-8 Content-Length handling | $1.24→$0.42 | 66% | 5.3→1.7 | 68% |
These are the representative runs per example, selected for the demo.