Skip to content

NetSerial AI V0.9.0 正式版 / Production Release

Choose a tag to compare

@chenwei666 chenwei666 released this 25 Aug 03:18
2efa1e3

NetSerial AI V0.9.0 Production Release / 正式版

V0.9.0 turns NetSerial AI into a complete in-app network operations assistant while preserving the rule that AI never sends device commands automatically.

Highlights / 重点功能

  • Full multi-turn AI chat across all 18 configured provider entries, including Zhipu GLM and custom OpenAI-compatible gateways.
  • Encrypted local conversation history, device memory, terminal context, cancellation, retry, history management, copy, and redacted sharing.
  • OpenAI-compatible, Anthropic Messages, and HTTPS Ollama chat adapters with active-provider-first failover.
  • Allow-listed fenced command extraction, deterministic R0–R4 reclassification, and hard blocking of R4 loading from AI.
  • One-tap read-only incident evidence plans for H3C, Huawei, Cisco, Ruijie, and generic devices.
  • Guided runbooks with prompt matching, timeouts, bounded retry, stop-on-error, and explicit stop conditions.
  • Configuration drift severity, expanded security posture checks, and evidence-before-change-closure validation.

Safety / 安全边界

AI output remains advisory. Every command requires human review and continues through the existing terminal target, maintenance-window, and risk gates. Chat history is encrypted with Android Keystore on Android 6.0+, while Android 5.x never falls back to plaintext history. Terminal context and exported conversations are redacted locally.

Verification / 验证

  • 59 test classes / 181 tests: all passed
  • Debug and Release Lint: passed, 0 errors
  • Debug and production-signed Release APK: built successfully
  • ZIP alignment: passed
  • V1/V2/V3 signatures: verified
  • APK SHA-256: b4c4040cfc2f1895404dd0c9b6f8024f3824f4a1baf161621e132e504ee29697

Real Android devices, USB adapters, target switch versions, and owned AI accounts still require authorized field acceptance before production use.