Enterprise Model Gateway

/features/model

Supports on-prem private model deployment and external vendor LLM API calls. Context policy and transfer audit run throughout.

季度复盘报告· Q2 2026

本地模型MiniMax-M3

把 Q2 的项目周报聚合成季度复盘,按目标 / 关键产出 / 风险分类

已从 12 份周报中提取关键事件,正在调用知识库做交叉核验…

tool: knowledge.search320ms

{ query: "Q2 OKR 复盘", top_k: 8, filters: ["dept=产品", "type=周报"] }

命中 8 条 · 置信度 0.86

已生成草稿,包含 3 大目标 / 12 项产出 / 5 项风险

输入任务,Enter 发送 · Shift+Enter 换行

Two model channels

On-prem private deployment (intranet / mainstream GPU clusters) and external vendor LLM APIs (Qwen / DeepSeek / Doubao / OpenAI-compatible) run in parallel.

Multi-model integration

Unified access to Qwen / DeepSeek / Doubao / third-party OpenAI-compatible services. Auto-routing by data sensitivity and scenario.

Context policy

Conversation summary, attachment / knowledge snippets opt-in by policy. maxUserMessageChars truncation with confirmation; external context-limit error classification.

Security boundary

Client does not receive raw keys / endpoints. Backups / diagnostics / status are not exposed. Model switching is managed by Runtime.