Repository navigation
v0.3.4
Before you upgrade
- Upgrade with
mimry upgrade. On Windows, close the agents that use MIMRY first, or runmimry upgrade --stop-servers. - The first command after upgrading rebuilds the index, so Python
from pkg import moduleimports and tsconfig path aliases link files. find,preflightand context packs rank long task texts such as issue bodies better: on SWE-bench Verified the right file is in the top 10 for 74.6% of issues, up from 66.2%.
See the changelog for every change.
What's Changed
New features
- Show indexed files per language in mimry status by @jbacalso24 in #231
- Rank a file the query names by path or module by @jbacalso24 in #240
- Keep BM25 in step with long queries by @jbacalso24 in #241
Fixes
- Read intent from the task headline, not the whole body by @jbacalso24 in #239
- Resolve callers through imported modules and tsconfig path aliases by @jbacalso24 in #243
- Stop redacting 'pass ' prose in feedback by @jbacalso24 in #244
- Record changed files the suggestions missed in feedback by @jbacalso24 in #245
Documentation
- Say in the README that the index is deterministic by @jbacalso24 in #214
- Introduce MIMRY as the local context engine for AI coding agents by @jbacalso24 in #225
- Release 0.3.4 by @jbacalso24 in #248
Other changes
- Add a SWE-bench Verified file localization benchmark by @jbacalso24 in #226
- Report recall@k and missed files in the token benchmark by @jbacalso24 in #229
- Add the first SWE-bench Verified localization results by @jbacalso24 in #230
- Add a SWE-bench dev set for tuning localization by @jbacalso24 in #233
- Add a diagnose mode to the localization benchmark by @jbacalso24 in #234
- Retry a cut-short dataset API response by @jbacalso24 in #235
- Add SWE-bench Verified results after the ranking fixes by @jbacalso24 in #242
- Run the token benchmark in a GitHub workflow by @jbacalso24 in #246
- Raise the CI job limit to 30 minutes by @jbacalso24 in #228
Full Changelog: v0.3.3...v0.3.4