47 lines
2.0 KiB
JSON
47 lines
2.0 KiB
JSON
{
|
||
"module": "omlx_server",
|
||
"version": "1.0",
|
||
"purpose": "OMLX 推理服务 — 运行在小果（Mac 192.168.1.122:18003），提供本地 LLM 推理能力。通过 OMLX 框架加载本地模型，为 Agent 提供低延迟推理。",
|
||
"ui_location": "F Tab → AI网关 → OMLX推理服务(小果)",
|
||
|
||
"human_help": {
|
||
"title": "OMLX 本地推理服务",
|
||
"description": [
|
||
"小果（Mac Mini）上运行的本地推理服务，使用 OMLX 框架加载 GGUF 格式的本地模型。",
|
||
"提供 OpenAI 兼容接口，延迟低、无需网络、不消耗云端配额。",
|
||
"适合轻量推理、快速响应场景；复杂推理仍走云 Provider。"
|
||
],
|
||
"usage": [
|
||
"状态在 F Tab 的 AI网关 层查看（小果的远程服务标记为 remote）",
|
||
"Dashboard 运行在 246 Linux，无法直接检测 Mac 端口 → 通过 ejabberd 在线状态推断"
|
||
],
|
||
"troubleshooting": [
|
||
"如果 OMLX 服务不可达：检查小果的 omlx_server 进程是否运行",
|
||
"如果 Mac 离线：检查 EasyTier VPN 是否连通",
|
||
"日志在小果本地：~/omlx/server.log"
|
||
],
|
||
"related": "依赖 EasyTier VPN 连通 246 ↔ Mac；依赖本地 GGUF 模型文件"
|
||
},
|
||
|
||
"ai_spec": {
|
||
"apis": [
|
||
{"method": "GET", "path": "/v1/health", "returns": "{ok, model_loaded}", "desc": "健康检查 + 模型加载状态"},
|
||
{"method": "POST", "path": "/v1/chat/completions", "body": "OpenAI chat completion 格式", "returns": "OpenAI 格式响应", "desc": "本地 LLM 推理"}
|
||
],
|
||
"dependencies": [
|
||
"EasyTier VPN — 跨平台网络",
|
||
"本地 GGUF 模型文件",
|
||
"OMLX 框架"
|
||
],
|
||
"constraints": [
|
||
"仅能在 Mac（192.168.1.122）上运行",
|
||
"首次加载模型较慢（冷启动）",
|
||
"Dashboard 只能通过 XMPP 在线状态间接监控"
|
||
],
|
||
"related_files": [
|
||
"gateway/scripts/specs/omlx_server.json — 本 spec",
|
||
"config/agents.yaml — agent-003 定义"
|
||
]
|
||
}
|
||
}
|