Read-only first
The first paid path starts from PR-scoped evidence and names missing proof instead of silently widening the task.
Codaegis turns one AI-assisted GitHub pull request into one governed packet so a human and their AI agents can read a bounded review posture before any merge decision. Codaegis is decision support only. It does not merge, deploy, approve, certify, enforce, warrant, replace human judgment, or include public API access in Starter.
Visible uncertainty, narrow evidence, human-owned decisions.
The first paid path starts from PR-scoped evidence and names missing proof instead of silently widening the task.
The governed packet can support a decision, but Codaegis does not merge, deploy, certify, or approve code.
The first paid path is GitHub PR review, while public API, SDK, adapter, and non-PR work-family surfaces remain separately admitted future or higher-tier paths.
Formal trust and compliance materials remain held; Codaegis only names safeguards that are true and evidenced today.
When evidence is missing, the governed packet shows the gap and the next review action so buyers can inspect the limits.
If setup, repo access, analysis path, or packet transport is unavailable, Codaegis explains the stop and recovery step instead of filling the gap with a fake answer.
On a frozen 29-pull-request evaluation corpus — 20 higher-risk, 9 clean — the shipped review engine produced 0 dangerous misses and 0 false stops.
Dangerous misses
False stops
Verdict accuracy
Confidence calibration
This is a held evaluation corpus scored by frozen replay, not production traffic and not a guarantee. The provider is OpenAI; the model version and routing are held as internal engine configuration. Full methodology is published in the Codaegis product benchmark publication.
The operational questions a serious buyer asks before connecting a repository, answered only with what is true and evidenced today.
Data handling
Only what an admitted pull-request review needs: the pull request title and body, changed-file references, review, commit, and discussion summaries, and CI or status signals for that pull request. The posture is read-only first, scoped to the one admitted pull request, not your whole repository.
Data handling
Codaegis uses OpenAI as the third-party AI provider that processes pull-request and code content to produce the governed packet. The specific model version, routing, and prompt configuration are internal engine configuration, not a data-handling boundary, so they are not disclosed.
Data handling
No. Codaegis does not use customer packet, repository, or code content for broad model training without explicit opt-in or a later reviewed policy.
Data handling
Packet history is bounded by tier: Free keeps 3 days and Starter keeps 30 days. After a downgrade, data outside the shorter visible window may be held but unavailable for up to three days so the change can be reversed. You can export each governed packet as JSON or Markdown, and you remain responsible for storing packets you need beyond retention.
Data handling
Yes. You can revoke GitHub access at any time and request export or deletion where applicable through support@codaegis.com. Payments run through Stripe-hosted flows, and Codaegis does not store full card details.
Security
Responsible disclosure is live. A machine-readable contact is served at /.well-known/security.txt (RFC 9116) alongside a published security policy; report privately to support@codaegis.com and allow a reasonable remediation window before public disclosure.
Security
The packet is decision support only. It does not merge, deploy, approve, certify, or enforce, and write access to your repository is not the default posture. A human owns the decision the packet supports.
Security
Only safeguards that are true and evidenced today: read-only review posture, bounded packet retention, no merge or deploy authority, and Stripe-handled payment data. Formal compliance features and trust artifacts remain held; Codaegis claims no certification or audit approval.
These limits are not fine print. They are how the first paid module stays honest enough for a serious buyer to evaluate.
This boundary keeps public explanation honest while the product continues under held public-surface evidence.
This boundary keeps public explanation honest while the product continues under held public-surface evidence.
This boundary keeps public explanation honest while the product continues under held public-surface evidence.
This boundary keeps public explanation honest while the product continues under held public-surface evidence.
This boundary keeps public explanation honest while the product continues under held public-surface evidence.
This boundary keeps public explanation honest while the product continues under held public-surface evidence.
This boundary keeps public explanation honest while the product continues under held public-surface evidence.
This boundary keeps public explanation honest while the product continues under held public-surface evidence.
This boundary keeps public explanation honest while the product continues under held public-surface evidence.