openlit: 2.1.0
OpenLIT 2.1.0 introduces MiniMax provider support, Mem0/Claude/Zep memory connectors, agent loop detection, trace governance compliance packs (NIST AI RMF, EU AI Act, OWASP ASI), and enhanced security hardening.
原始内容为英文；当前页面提供中文导航与来源说明，具体事实请以原文为准。
为什么值得关注
OpenLIT 2.1.0 brings major enhancements to model coverage, agent observability, memory integrations, and governance. Key updates include: - Providers & Memory: Added built-in MiniMax provider (M3, M2.7) with metadata, pricing, cache rates, and streaming. Added Mem0, Claude, and Zep memory connectors with scoping, feedback, writes, and Ask Otter integration. - Observability & Agent Health: Introduced stuck-agent loop detection flagging repeated tool calls with token/cost impact across ClickHouse, Tempo, and Jaeger. Added generation-health span classification (truncated, filtered, empty, swapped). - Governance & Security: Trace governance panel and API with findings, coding-agent security checks, Passport export endpoint, and policy packs for NIST AI RMF, EU AI Act, and OWASP ASI. Hardened authentication, tenant isolation, cron access, input sanitization, and API key redaction. - Operational Fixes: Bundled platform-compatible Prisma engines in the client image to support air-gapped deployments, fixed onboarding redirection issues, and optimized telemetry volume.
可执行摘要
OpenLIT version 2.1.0 adds MiniMax (M3, M2.7) integration, new memory connectors (Mem0, Claude, Zep), stuck-agent loop detection across ClickHouse/Tempo/Jaeger, generation-health trace filters, trace governance policies/passports, and tenant/auth hardening.
- Agent 实用度
- 82/100
- 可信度
- 96%
- 机器格式
- JSON + Markdown
开发者应核对什么
- Upgrade OpenLIT deployment images to 2.1.0 to leverage air-gapped Prisma engine bundling and security fixes.
- Review API authentication configurations, including new organization/project/environment headers and redacted key listings.
- Explore stuck-agent loop detection filters to identify wasteful or recurring tool-call cycles in traces.