Hanvietech 推出 AI 代码安全智能体 Jikimi，AI 审查由 Claude 驱动

English · 한국어

Jikimi is an AI code security agent that connects the whole process in one workflow: finding security issues in code, reviewing them with AI and people together, deciding according to your organization's policy, and verifying again after the fix. Built by Hanvietech.

Coming soon. This product is preparing for launch. This repository holds public documentation: a product overview and how to use it. The product source code is not public. Features and structure described here may change before release.

See the demo · Result screens built with sample data. No real customer data is shown.

How it works

Security scanning: Finds potential security issues in your code and the open source it uses.
AI + human review: AI organizes and explains the results, and a reviewer confirms them.
Policy decision: Decides whether to pass, hold, or fix, based on your organization's security policy.
Fix & re-scan: Re-scans after the fix to confirm the issue is resolved. Unresolved issues flow back into scanning.

Features

Integrated security checks: Examines source code, open source components, secrets, and licenses in a single scan, and organizes the results on one standard.
Results at a glance: Shows finding counts by project and category, counting the same issue once. New, recurring, and resolved findings are tracked across scans.
Why it matters and how to fix it: Every finding explains why it is risky and how to fix it.
AI review: AI classifies findings as true positive, false positive, or needs review, with its evidence and a suggested fix. AI review runs on Claude by Anthropic.
People make the final call: AI judgments are suggestions. Only findings a reviewer confirms are treated as false positives and removed from the counts.

How to use it

Security reviewers

Account setup: When Hanvietech opens your organization's account, set a password from the link you receive and sign in.
Agent distribution: Create an agent download link in the dashboard and send it to developers.
Results: Review results by project on the dashboard, and open a finding to see its location, why it is risky, and how to fix it.
AI judgments: Review the findings AI judged likely to be false positives, with their evidence, and confirm them at once.

Developers

Unpack the bundle from your security reviewer and run the following from the project root. Connection details are included in the bundle, so there is nothing to enter.

./jikimi install --yes   # prepare for scanning
./jikimi scan .          # run the scan

When the scan finishes, the results appear on the security reviewer's dashboard. macOS is supported first. Windows support is planned.

Early access

To receive a pre-release demo and details, choose Jikimi early access in the contact form on hanvietech.com, or email ceo@hanvietech.com.

Sister product: Doumi — AI web vulnerability testing agent