Secuarden is a GitHub app that flags risky code in a pull request, and a local recorder of what Claude Code actually did.
Two product lines, described at two different stages
Code Intelligence is the GitHub app, marked available now. Change Assurance is the record of how an AI-assisted change was made: the request, the agent activity, the policy decision, and the approval. The about page still calls that second line Coding Agent Governance, says it is being developed, and asks for two companies that already use coding agents as design partners. The homepage marks the session ledger, the CCR score, and the Policy Enforcement Engine as available now, and says policy-gate enforcement joined the platform in August 2026. The CLI page says the Policy Enforcement Engine will evaluate agent actions, and that the local gate command does not replace enterprise approval.
The scan that looks at the code
Code Intelligence installs from the GitHub Marketplace. You choose repositories, it comments on risks in the pull request, and it suggests a fix for a person to test. It does not apply the fix. The page says a clean scan is one input to a review, and that compliance still depends on the requirements, the evidence, and the approval. The free plan's scan allowance is not printed on the site. They point you at the Marketplace listing for the current number. Languages listed there are JavaScript, TypeScript, Python, and PHP.
The worked example is a SQL string built from a name in app/search.py. They label it illustrative, not a customer scan, and they show the suggested change as a bound parameter. The context map beside it is described as a simplified picture, not a claim about a particular data-flow engine. The question of whether the code was written by an AI is a different product. Lineage sits under Change Assurance.
The homepage has a separate box. Paste a public GitHub URL, no account, three free scans, results in about 20 seconds. It reads commit patterns, pull-request metadata, and AI attribution. It estimates how many commits in the last 30 days look AI-assisted, whether those commits touched auth, payments, or config, and whether an attributed pull request has no human review signal. The same page says this is not a vulnerability scan. Connecting GitHub for a private repository is described as read-only, with no code stored, and with results cached for 24 hours. A CCR of 74 is already drawn in that box, above the field where you paste a repository.
The recorder hooks Claude Code, and lists the others as planned
The open-source CLI is secuardenai/secuarden-cli, Apache-2.0, written in Go. The README's support table lists Claude Code as supported. Cursor, GitHub Copilot, Windsurf, and OpenAI Codex are marked planned. The homepage's sample ledger names Copilot and Cursor anyway. Those lines are a demo. The README says its own terminal demo replays sanitized Claude Code fixtures and is not a live agent session.
- Without an API key, events stay in ~/.secuarden/secuarden.db. The hooks named in the README are PreToolUse, PostToolUse, SessionStart, and SessionEnd.
- secuarden serve opens a read-only view on 127.0.0.1 port 8743 and makes no external request.
- An API key turns on sync of already-redacted session envelopes. The sample status points that sync at https://app.secuarden.ai. The repository has to be connected in the dashboard before change-set history persists across sessions.
- secuarden verify checks a per-session SHA-256 chain. Records from before that chain stay readable and are labeled legacy. Verify, the bill-of-materials export, and the gate fail closed on them.
- The gate scores paths tied to identity, payments, secrets, cryptography, infrastructure, and schemas. Tests and docs score lower. The README calls this a governance signal, not a SAST scan or a license scan.
- The marketing CLI page labels the product v0.2.0. The README's sample status prints Secuarden Capture Agent v0.1.0.
The README offers two installs: a shell script at install.secuarden.ai, or the Homebrew tap secuarden/tap/secuarden. A sample status in that file uses the address [email protected]. That is sample output, not a contact address.
What the record claims, and what it refuses
Change Assurance says it connects the developer's request, agent activity, policy decisions, approvals, and the files in the pull request. A Context BOM is that per-session record. They contrast it with an AI-BOM, which lists AI systems in the environment rather than what entered the code. Export fields named on the CLI page include the agent and model, the developer, the repository and branch, the files changed, file-content hashes, and whether the hash chain is intact.
The change-assurance page says Secuarden does not automatically make an organisation compliant. Which coding agents are supported, whether source code is stored, how long evidence is kept, and how GitHub or CI is connected are confirmed in an enterprise evaluation. The CLI itself is described as keeping source code on the machine. Framework names they map evidence toward are SOC 2 CC8.1, ISO 27001, PCI DSS 4.0, NIST SSDF, NIST AI RMF, and the EU AI Act.
A different public repository, secuardenai/context-confidence-rating, uses the same initials for another job: a Python score of whether a scanner finding looks exploitable in that codebase. The website's CCR is a 0 to 100 score of how complete the captured evidence and controls are. The two are not the same report.
The founder, and the price that is not on the page
The about page names Gaurab B, founder, in Melbourne. It says the product comes from more than 20 years in security, including time at Microsoft and GE, and lists CISSP, CSSLP, and CCSP. The line on that page is: "As agents take on more of the work, teams need a reliable answer to three questions: what changed, why, and under whose control?" The GitHub organization that links to the site lists [email protected]. The footer shows an NVIDIA Inception Program member badge. No dollar price is printed for Change Assurance, or for a paid Marketplace tier.