CamelHive
Production-readiness audit
The harness for AI-written code
AI wrote it. Know if the code is ready to sell and ship to production.
CamelHive checks your repository against a public, versioned set of rules for code quality and security. Every finding points to where it was seen, and anything the engine could not decide is reported as undetermined, never as passed.
Request early access, paid plans from $20/month, or a custom quote for an engineer-reviewed, signed attestation.
Connect your repo. Audit before you ship.
Install the CamelHive GitHub App on the repositories you choose and start an audit from your dashboard whenever you ship. Each project shows when its default branch has moved past the last audited commit. Failed rules come with how to fix them — the change stays yours.
Checked against 141 code rules.
Rules that don't apply to your stack are shown as not applicable. 26 of them are deterministic checks that run without a model; the rest go through a model pass.
Deterministic checks report what their detector found. In the model pass, critical and high failures are only accepted after an adversarial review tries to refute them; medium and low failures that skip that review are marked as such.
Every score is shown with its coverage: how many rules were actually decided.
No findings means no check found anything within the stated scope and date; it is not a guarantee that there are no problems.
The free Scan runs the deterministic checks only.
Infrastructure (15 rules) and owner questionnaire (28 rules): Not included in this audit.
We run this on ourselves
104 of 139 applicable code rules decided in our own audit · 13 failures found
Measured Oct 16, 2026
Pricing
Pick how much of the audit you want automated.
Coverage below is counted live from the rule catalog, not typed in by hand — if the catalog grows, the numbers do too.
Scan
Free
Structural checks only — no model in the loop, no cost.
26 of 141 code rules evaluated
Structural scans, no model
Audit
$20/month
One model-driven pass. Critical and high failures are only accepted after an adversarial review tries to refute them.
Up to 141 of 141 code rules evaluated
1 agentic run/month
Audit Pro
$100/month
Same engine, more runs — track drift across a sprint, not just a single snapshot.
Up to 141 of 141 code rules evaluated
4 agentic runs/month
Reviewed Attestation
Custom
An engineer reviews what the automated audit could not decide and signs the result.
What the automated audit left undecided, reviewed by an engineer
Quoted per engagement
"Up to" — an agentic run can abstain when there isn't enough evidence. An abstention is reported, never shown as a pass. Target spend, not a hard stop — an in-flight batch can't be interrupted mid-run, so actual spend can land a small, bounded amount over target.
"I could just ask ChatGPT."
Fair question. Here's what a chat window doesn't give you.
A versioned catalog
Rule ids that never get renumbered — a run today is comparable to one three months from now. A chat isn't reproducible even with itself.
Evidence, or it doesn't count
Every finding points to where it was seen in the snapshot. What we can't verify is reported as undetermined instead.
An adversarial second pass
In the model pass, a critical or high failure is only accepted after a second pass tries to refute it.
Declared coverage
We say exactly how many rules were evaluated, and which ones we couldn't.
A pinned, immutable snapshot
The report points to an exact commit SHA — re-verifiable, and diffable against your next run.
Want an engineer to sign off?
With Reviewed Attestation, an engineer reviews what the automated audit could not decide and signs the result. Quoted separately, on request. Scan, Audit and Audit Pro have no human review.
The rules are public.
Every rule, its severity and how it is checked, versioned with a changelog. Read them before you trust a score.
Read the standard →Entry Point
Not sure which plan fits? Let's talk.
Tell us about your MVP. We'll point you to the right plan — including whether a Reviewed Attestation is worth it for where you're at.
Already know what you want? Pick a plan above — this form works for any of them.