01
安装到代码库
选择 Codna 可映射的代码库，并配置私有模式出口规则。
How it works
From first issue to merged fix.
02
自动分类问题
Codna 映射代码库并评论可疑文件、根因及建议的后续步骤。
03
开启修复 PR
通过标签、命令或失败检查触发经验证的修复 Pull Request。
04
保持掌控
在合并前要求审查、分支保护、测试门控和审批。
评论即可运行 Codna。
直接在 GitHub 中触发修复，同时在 Issue 中保留完整操作记录。
@codna fix this @codna explain blast radius @codna open-pr
每个 PR 都自带说明。
Codna 附带根因分析、测试计划、置信度评分和回归风险估算，让审查者能更快推进。
根因session race
置信度91%
风险Low
Frequently asked
Codna's deterministic engine maps the relevant code in ~60ms for zero LLM tokens, then an AI agent works from a ~600-token evidence bundle to write the fix. Your own tests verify it before the pull request opens.
No. Codna only opens an automated fix pull request when the fix passes your test suite. If verification fails, nothing is merged and nothing is opened.
Because the AI agent works from a ~600-token evidence bundle — 162x less context than reading the whole repo — a verified fix runs about $0.037 in measured testing, where a typical agentic edit that reads the whole repo costs roughly $1.
No. You self-host the engine, bring your own API key, and egress is fail-closed. Your code is never used for training.
Codna supports 250+ languages — it mapped 130 repos in 9.2 seconds for zero tokens, so coverage is broad by design.
In head-to-head testing against Cursor across 87 tasks, Codna used 5× fewer tokens and ran 1.7× faster, with every fix verified by a passing test.