Code review that fixes AI-written code, on your machine.

The agents ship. Your engineers carry the risk. Truemend finds the mistakes coding agents repeat, shows each fix as a diff you can read, and hands bigger changes to your team.

Read the sample report
  • Runs on your machine, no model
  • Nothing changes until you apply a fix
  • Free, no account needed
src/db/query.ts · line 34illustrative
mediumempty-catch-block

The catch block is empty, so a failed query is dropped and never reaches the logger.

try {
return await db.query(sql, params);
-} catch (err) {}
+} catch (err) {
+ console.error(err);
+}
Applied for youdiff attached · re-parsed · review before merge

Rather start with a person? I run audits alongside your team.

Truemend does the reading. Your team makes the calls.

RulesLead
health ≥ 70 / 100run 1
AnalyzeTruemend
src/db/query.ts
1,184local
FindTruemend
Dead code after returnsrc/utils/format.ts:77
You decideYour team
Applied for you0
Bounded refactor0
Your call0
Reviewed and kept
Applied for youMechanical fixes, applied as a diff you can review or revert.
Bounded refactorA larger change, prepared as a diff for you to approve. Opt-in, using the model you choose.
Your callA design decision. Truemend explains the problem and estimates the cost to fix it.

Illustrative sample run. Review diffs and run your tests before merging. How to read a finding

What a run gives you.

What it catches →
terminalhalyard-api
$ truemend .
47 files · 512 functions
health 66 / 100
214 findings
● 101 applied for you $0
● 72 bounded refactor ≈ $2,400
● 41 your call ≈ $2,450
report → truemend-report.html
Run it

One command on your machine. Nothing on disk changes until you apply a fix.

pull request · src/ingest/run.pyreview
86try:
87 rows = load_batch(path)
88except:
89 rows = []
truemendmediumbare_exceptline 88

A bare except hides the real failure mode. Narrow it to the real exception type and log the context.

Bounded refactor · reviewed diff available
Review it

Each finding sits on its line in the pull request, with its tier and the change on offer.

audit report · halyard-api2026-07-06
Health66 / 10047 files · 512 functions · 214 findings
By severity
Critical9High38Medium102Low65
Cost to clear, by tier≈ $4,850
Applied for you101$0
Bounded refactor72≈ $2,400
Your call41≈ $2,450
Plan it

The report prices the backlog by tier, so the work can be planned before the sprint starts.

Open the full sample report →

Illustrative output, built from the halyard-api sample report.

Measured in public, losses included.

All results, with dates and scope →

Truemend10.9%

Semgrep17.1%

Share of findings judged wrong on Netflix Dispatch. Lower is better.

On NetBox, Bandit, Pylint and Ruff combined beat it. That result is published too.

A product, and a person who works through it with you.

I lead every engagement myself, and bring in engineers I’ve worked with when a job needs more hands. The goal is a team that keeps its own code straight without outside help.

Why Truemend exists →
ConsultingA codebase audit. I run Truemend on your code and agree with your engineers what gets fixed first.Engagements →TrainingHands-on training on your own codebase, then an audit of what the team ships next.AI engineering velocity →

Run it on a repository you already have.

Read the sample report
  • Runs on your machine, no model
  • Nothing changes until you apply a fix
  • Free, no account needed
$ pip install truemend$ truemend .