1
Anchor the trace to the graph
Codna maps the whole repo into a dependency and blast-radius graph in about 60ms for zero LLM tokens, then pins the stack trace to the suspect files and call paths behind it.
Use case
From Sentry alert to fix PR with Codna
A production error in Sentry is a precise signal, not a vague bug report. Codna is an AI coding agent that turns the stack trace into a localized production bug fix — mapped deterministically, fixed from a tiny evidence bundle, verified by your tests, and opened as a PR.
The problem
Turning a stack trace into a fix is the slow part
A Sentry alert tells you where the code threw, not why it broke. The frame at the top of the stack trace is usually a symptom — the real cause is a change several files away, behind imports, call paths, and a release you half-remember. Engineers lose hours pasting traces into a chat, re-reading unfamiliar modules, and guessing which fix won't break three other call sites. To fix Sentry errors safely you first have to understand the blast radius, and that understanding is exactly what gets skipped under incident pressure.
How Codna fixes it
From Sentry alert to verified PR
Whole repo100,000+
Evidence bundle~600
2
Fix from a ~600-token bundle
Codna hands the agent a ~600-token evidence bundle scoped to the real cause — 162x less context than reading the repo — so the production bug fix addresses the root, not the line that happened to throw.
3
Open a test-verified PR
The patch runs against your own test suite before it lands, then the native GitHub App opens a pull request with the root cause, blast radius, and passing check attached.
codna fix . --issue "Sentry CHECKOUT-1421: TypeError in session handler"
What you get
What you get from each alert
Stack trace mapped, zero tokens
A deterministic dependency and blast-radius graph anchors the trace to its real cause in ~60ms — no RAG, no embeddings, no tokens spent reading the repo.
Every fix proven by your tests
Codna verifies the patch against your project's own test suite before it lands, so a Sentry fix only opens a PR when the checks are already green.
About $0.04 per verified fix
Because the agent works from a ~600-token evidence bundle instead of the whole codebase, turning an alert into a fix costs pennies, not dollars.
The proof
Fewer tokens. Faster. Verified.
Frequently asked
Point Codna at the Sentry issue. It maps the repo deterministically, anchors the stack trace to the suspect files, fixes the cause from a ~600-token evidence bundle, and opens a PR once your tests pass.
Codna resolves the trace against its dependency graph for you. It walks the call paths, scopes the agent to the few files that matter, and leaves the unrelated 100k+ tokens out of the work.
The error and trace are enough. Codna resolves the rest from the deterministic graph, so you don't have to hand-collect surrounding files or context.
Codna maps the blast radius before it edits, then runs your own test suite to verify the patch. A fix only lands when every affected path stays green.
Yes. Each pull request comes with the root cause, confidence, blast radius, and the passing test attached — so review is fast and the fix is auditable.
Yes. Codna self-hosts with bring-your-own-key, fail-closed egress, and no training on your code, so production traces and source never leave your control.
Related
