@jheavenknows/bluerouter 1.0.57 → 1.0.59

package/app/.next-cli-build/server/app/api/ollama-nodes/[id]/install/route.js

@@ -1,11 +1,12 @@
-(()=>{var a={};a.id=9902,a.ids=[9902],a.modules={261:a=>{"use strict";a.exports=require("next/dist/shared/lib/router/utils/app-paths")},3295:a=>{"use strict";a.exports=require("next/dist/server/app-render/after-task-async-storage.external.js")},6774:(a,b,c)=>{"use strict";c.d(b,{U1:()=>v,Vs:()=>w,YT:()=>z,iJ:()=>y,op:()=>A,ri:()=>x});var d=c(79748),e=c.n(d),f=c(33873),g=c.n(f),h=c(55511),i=c.n(h),j=c(21820),k=c.n(j),l=c(55013);let m=process.env.DATA_DIR||process.env.BLUEROUTER_DATA_DIR||g().join(k().homedir(),".bluerouter"),n=g().join(m,"ollama-nodes.json");async function o(){await e().mkdir(m,{recursive:!0});try{await e().access(n)}catch{await e().writeFile(n,JSON.stringify({nodes:[]},null,2))}}async function p(){await o();try{let a=await e().readFile(n,"utf8"),b=JSON.parse(a||"{}");return{nodes:Array.isArray(b.nodes)?b.nodes:[]}}catch{return{nodes:[]}}}async function q(a){await e().mkdir(m,{recursive:!0}),await e().writeFile(n,JSON.stringify({nodes:a.nodes||[]},null,2))}function r(){return new Date().toISOString()}function s(a,b={}){if(!a)return b;if("object"==typeof a&&!Array.isArray(a))return a;try{let c=JSON.parse(String(a));return c&&"object"==typeof c&&!Array.isArray(c)?c:b}catch{return b}}function t(a){return a?{...a,sshPassword:void 0,sudoPassword:void 0}:null}function u(a={},b={}){var c;let d=String(a.host??b.host??"").trim(),e=Number(a.agentPort??b.agentPort??19999)||19999,f=Array.isArray(c=a.ollamaPorts??b.ollamaPorts??[11434])?[...new Set(c.map(a=>Number(a)).filter(a=>Number.isFinite(a)&&a>0))]:[...new Set(String(c||"11434").split(/[\s,;]+/).map(a=>Number(a.trim())).filter(a=>Number.isFinite(a)&&a>0))],g=s(a.modelMemoryRequirements??b.modelMemoryRequirements,b.modelMemoryRequirements||{}),h=(0,l.nw)(g,{partial:!0,preserveEnabled:!0}),j=b.id||a.id||i().randomUUID(),k=String(a.agentToken??b.agentToken??i().randomBytes(24).toString("hex")).trim(),m=function(a,b,c={}){let d=new Map((Array.isArray(c.instances)?c.instances:[]).map(a=>[Number(a.port),a])),e=new Map((Array.isArray(a)?a:[]).map(a=>[Number(a.port),a]));return b.map(a=>(function(a={},b={}){var c;let d=Number(a.port??b.port??11434)||11434;return{id:String(a.id??b.id??`ollama-${d}`).trim()||`ollama-${d}`,name:String(a.name??b.name??`Ollama :${d}`).trim()||`Ollama :${d}`,port:d,enabled:a.enabled??b.enabled??!0,gpuMode:["auto","cpu","specific","all"].includes(a.gpuMode??b.gpuMode)?a.gpuMode??b.gpuMode:"auto",gpuIndices:Array.isArray(c=a.gpuIndices??b.gpuIndices??[])?[...new Set(c.map(a=>Number(a)).filter(a=>Number.isInteger(a)&&a>=0))]:[...new Set(String(c||"").split(/[\s,;]+/).map(a=>Number(a)).filter(a=>Number.isInteger(a)&&a>=0))],allowMultiGpuSplit:a.allowMultiGpuSplit??b.allowMultiGpuSplit??!1,tags:Array.isArray(a.tags??b.tags)?(a.tags??b.tags).map(a=>String(a).trim()).filter(Boolean):[],modelMemoryRequirements:(0,l.nw)(s(a.modelMemoryRequirements??b.modelMemoryRequirements,b.modelMemoryRequirements||{}),{partial:!0,preserveEnabled:!0})}})(e.get(Number(a))||{},d.get(Number(a))||{port:a}))}(a.instances??b.instances,f.length?f:[11434],b);return{id:j,name:String(a.name??b.name??d??"Ollama Node").trim()||d||"Ollama Node",host:d,sshPort:Number(a.sshPort??b.sshPort??22)||22,sshUser:String(a.sshUser??b.sshUser??"root").trim()||"root",useSudo:a.useSudo??b.useSudo??!1,agentPort:e,agentToken:k,ollamaPorts:f.length?f:[11434],instances:m,refreshIntervalMs:Math.max(5e3,Number(a.refreshIntervalMs??b.refreshIntervalMs??15e3)||15e3),enabled:a.enabled??b.enabled??!0,ufwAllowFrom:String(a.ufwAllowFrom??b.ufwAllowFrom??"").trim(),useGlobalModelMemory:a.useGlobalModelMemory??b.useGlobalModelMemory??!0,modelMemoryRequirements:h,notes:String(a.notes??b.notes??"").trim(),lastStatus:b.lastStatus||null,lastEvaluation:b.lastEvaluation||null,lastStatusAt:b.lastStatusAt||null,lastError:b.lastError||null,lastInstallAt:b.lastInstallAt||null,lastInstallOk:b.lastInstallOk||!1,lastInstallLog:b.lastInstallLog||"",createdAt:b.createdAt||r(),updatedAt:r()}}async function v(){return(await p()).nodes.map(a=>t(u(a,a)))}async function w(a){let b=(await p()).nodes.find(b=>b.id===a)||null;return b?u(b,b):null}async function x(a){let b=await p(),c=u(a);return b.nodes.push(c),await q(b),t(c)}async function y(a,b){let c=await p(),d=c.nodes.findIndex(b=>b.id===a);if(d<0)return null;let e=u(b,c.nodes[d]);return c.nodes[d]=e,await q(c),t(e)}async function z(a,b){let c=await p(),d=c.nodes.findIndex(b=>b.id===a);return d<0?null:(c.nodes[d]={...c.nodes[d],...b,updatedAt:r()},await q(c),t(u(c.nodes[d],c.nodes[d])))}async function A(a){let b=await p(),c=b.nodes.length;return b.nodes=b.nodes.filter(b=>b.id!==a),await q(b),b.nodes.length!==c}},10846:a=>{"use strict";a.exports=require("next/dist/compiled/next-server/app-page.runtime.prod.js")},21820:a=>{"use strict";a.exports=require("os")},29294:a=>{"use strict";a.exports=require("next/dist/server/app-render/work-async-storage.external.js")},33873:a=>{"use strict";a.exports=require("path")},44870:a=>{"use strict";a.exports=require("next/dist/compiled/next-server/app-route.runtime.prod.js")},55013:(a,b,c)=>{"use strict";c.d(b,{LM:()=>v,LQ:()=>u,MW:()=>s,eY:()=>t,nw:()=>o});var d=c(79748),e=c.n(d),f=c(33873),g=c.n(f),h=c(21820),i=c.n(h);let j=process.env.DATA_DIR||process.env.BLUEROUTER_DATA_DIR||g().join(i().homedir(),".bluerouter"),k=g().join(j,"ollama-model-memory.json"),l={"qwen3:1.7b":{minRamFreeMb:2048,minVramFreeMb:0,loadedRamFreeMb:512,loadedVramFreeMb:0,safetyMarginMb:512,notes:"small text model"},"qwen3:4b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:0,safetyMarginMb:512,notes:"small/medium text model"},"qwen3:8b":{minRamFreeMb:6144,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:512,safetyMarginMb:1024,notes:"8B text model"},"qwen3:14b":{minRamFreeMb:8192,minVramFreeMb:1e4,loadedRamFreeMb:2048,loadedVramFreeMb:1024,safetyMarginMb:1024,notes:"14B class model"},"qwen3:32b":{minRamFreeMb:4096,minVramFreeMb:16e3,loadedRamFreeMb:2048,loadedVramFreeMb:2048,safetyMarginMb:2048,notes:"32B class model, GPU preferred"},"qwen3-coder:30b":{minRamFreeMb:8192,minVramFreeMb:24e3,loadedRamFreeMb:2048,loadedVramFreeMb:3072,safetyMarginMb:2048,notes:"30B coder"},"qwen3-coder:latest":{minRamFreeMb:8192,minVramFreeMb:24e3,loadedRamFreeMb:2048,loadedVramFreeMb:3072,safetyMarginMb:2048,notes:"qwen3 coder latest, treat as 30B unless overridden"},"qwen2.5-coder:14b":{minRamFreeMb:8192,minVramFreeMb:1e4,loadedRamFreeMb:2048,loadedVramFreeMb:1024,safetyMarginMb:1024,notes:"14B coder"},"qwen2.5vl:7b":{minRamFreeMb:8192,minVramFreeMb:8e3,loadedRamFreeMb:2048,loadedVramFreeMb:1024,safetyMarginMb:1024,notes:"7B vision-language"},"qwen2.5vl:latest":{minRamFreeMb:8192,minVramFreeMb:1e4,loadedRamFreeMb:2048,loadedVramFreeMb:1024,safetyMarginMb:1024,notes:"vision-language latest"},"bge-m3:latest":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:512,loadedVramFreeMb:0,safetyMarginMb:512,notes:"embedding model"},"deepseek-coder:6.7b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:512,safetyMarginMb:1024,notes:"6.7B coder"},"gemma4:e4b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:0,safetyMarginMb:1024,notes:"small Gemma variant"},"gemma:7b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:0,safetyMarginMb:1024,notes:"7B text model"},"glm-4.7-flash:latest":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:512,safetyMarginMb:1024,notes:"flash/local text model"},"gpt-oss:20b":{minRamFreeMb:8192,minVramFreeMb:12e3,loadedRamFreeMb:2048,loadedVramFreeMb:2048,safetyMarginMb:2048,notes:"20B class model"},"llama3.1:8b":{minRamFreeMb:6144,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:512,safetyMarginMb:1024,notes:"8B text model"},"llama3.2-vision:latest":{minRamFreeMb:8192,minVramFreeMb:1e4,loadedRamFreeMb:2048,loadedVramFreeMb:1024,safetyMarginMb:1024,notes:"vision model"},"qwen3-*-translate-f16:*":{minRamFreeMb:8192,minVramFreeMb:16e3,loadedRamFreeMb:2048,loadedVramFreeMb:2048,safetyMarginMb:2048,notes:"custom Qwen3 f16 translate wildcard"},"qwen3-*-translate*:latest":{minRamFreeMb:6144,minVramFreeMb:8e3,loadedRamFreeMb:1024,loadedVramFreeMb:1024,safetyMarginMb:1024,notes:"custom Qwen3 translate wildcard"},"bge-m3:*":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:512,loadedVramFreeMb:0,safetyMarginMb:512,notes:"embedding model wildcard"},"gemma*:e4b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:0,safetyMarginMb:1024,notes:"Gemma E4B wildcard"},"gemma*:7b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:0,safetyMarginMb:1024,notes:"Gemma 7B wildcard"},"*:1.7b":{minRamFreeMb:2048,minVramFreeMb:0,loadedRamFreeMb:512,loadedVramFreeMb:0,safetyMarginMb:512,notes:"generic 1.7B fallback"},"*:4b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:0,safetyMarginMb:512,notes:"generic 4B fallback"},"*:6.7b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:512,safetyMarginMb:1024,notes:"generic 6-7B fallback"},"*:7b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:512,safetyMarginMb:1024,notes:"generic 7B fallback"},"*:8b":{minRamFreeMb:6144,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:512,safetyMarginMb:1024,notes:"generic 8B fallback"},"*:14b":{minRamFreeMb:8192,minVramFreeMb:1e4,loadedRamFreeMb:2048,loadedVramFreeMb:1024,safetyMarginMb:1024,notes:"generic 14B fallback"},"*:20b":{minRamFreeMb:8192,minVramFreeMb:12e3,loadedRamFreeMb:2048,loadedVramFreeMb:2048,safetyMarginMb:2048,notes:"generic 20B fallback"},"*:30b":{minRamFreeMb:8192,minVramFreeMb:24e3,loadedRamFreeMb:2048,loadedVramFreeMb:3072,safetyMarginMb:2048,notes:"generic 30B fallback"},"*:32b":{minRamFreeMb:4096,minVramFreeMb:16e3,loadedRamFreeMb:2048,loadedVramFreeMb:2048,safetyMarginMb:2048,notes:"generic 32B fallback"}},m=["minRamFreeMb","minVramFreeMb","loadedRamFreeMb","loadedVramFreeMb","safetyMarginMb"];function n(a,b=0){let c=Number(a);return Number.isFinite(c)&&c>=0?c:b}function o(a={},b={}){let c={};if(!a||"object"!=typeof a||Array.isArray(a))return c;for(let[d,e]of Object.entries(a)){let a=String(d||"").trim();if(!a||!e||"object"!=typeof e||Array.isArray(e))continue;let f=function(a,{partial:b=!1,preserveEnabled:c=!1}={}){let d={};for(let c of m)if(b){var e;null!=(e=a?.[c])&&""!==String(e).trim()&&(d[c]=n(a[c],0))}else d[c]=n(a?.[c],0);let f=String(a?.notes||"").trim();return f&&(d.notes=f),c&&a&&Object.prototype.hasOwnProperty.call(a,"enabled")&&(d.enabled=!1!==a.enabled),d}(e,b),g=m.some(a=>Object.prototype.hasOwnProperty.call(f,a)),h=b.preserveEnabled&&Object.prototype.hasOwnProperty.call(f,"enabled");(g||f.notes||h)&&(c[a]=f)}return c}async function p(){await e().mkdir(j,{recursive:!0});try{await e().access(k)}catch{await e().writeFile(k,JSON.stringify({requirements:l,updatedAt:new Date().toISOString()},null,2))}}async function q(){await p();try{let a=await e().readFile(k,"utf8"),b=JSON.parse(a||"{}");return{requirements:function(a={}){return o({...l,...a})}(b.requirements||{}),updatedAt:b.updatedAt||null}}catch{return{requirements:o(l),updatedAt:null}}}async function r(a){await e().mkdir(j,{recursive:!0});let b={requirements:o(a),updatedAt:new Date().toISOString()};return await e().writeFile(k,JSON.stringify(b,null,2)),b}async function s(){return(await q()).requirements}async function t(){return q()}async function u(a){return r(a)}async function v(){return r(l)}},55511:a=>{"use strict";a.exports=require("crypto")},63033:a=>{"use strict";a.exports=require("next/dist/server/app-render/work-unit-async-storage.external.js")},68757:(a,b,c)=>{"use strict";c.r(b),c.d(b,{handler:()=>J,patchFetch:()=>I,routeModule:()=>E,serverHooks:()=>H,workAsyncStorage:()=>F,workUnitAsyncStorage:()=>G});var d={};c.r(d),c.d(d,{POST:()=>D,dynamic:()=>C,runtime:()=>B});var e=c(19225),f=c(84006),g=c(8317),h=c(99373),i=c(34775),j=c(24235),k=c(261),l=c(54365),m=c(90771),n=c(73461),o=c(67798),p=c(92280),q=c(62018),r=c(45696),s=c(47929),t=c(86439),u=c(37527),v=c(23211),w=c(6774),x=c(79646);function y(a){return`'${String(a??"").replace(/'/g,"'\"'\"'")}'`}function z(a,b,{input:c="",timeoutMs:d=12e4}={}){return new Promise(e=>{let f=(0,x.spawn)(a,b,{stdio:["pipe","pipe","pipe"]}),g="",h="",i=!1,j=setTimeout(()=>{i||f.kill("SIGKILL")},d);f.stdout.on("data",a=>{g+=a.toString()}),f.stderr.on("data",a=>{h+=a.toString()}),f.on("close",a=>{i=!0,clearTimeout(j),e({ok:0===a,code:a,stdout:g,stderr:h})}),c&&f.stdin.write(c),f.stdin.end()})}async function A(a,b={}){let c=a.sshUser||"root",d=Number(a.sshPort||22),e=`${c}@${a.host}`,f=function(a,b={}){let c="/opt/bluerouter-ollama-agent",d=(a.ollamaPorts||[11434]).join(","),e=String.raw`#!/usr/bin/env python3
+(()=>{var a={};a.id=9902,a.ids=[9902],a.modules={261:a=>{"use strict";a.exports=require("next/dist/shared/lib/router/utils/app-paths")},3295:a=>{"use strict";a.exports=require("next/dist/server/app-render/after-task-async-storage.external.js")},6774:(a,b,c)=>{"use strict";c.d(b,{U1:()=>v,Vs:()=>w,YT:()=>z,iJ:()=>y,op:()=>A,ri:()=>x});var d=c(79748),e=c.n(d),f=c(33873),g=c.n(f),h=c(55511),i=c.n(h),j=c(21820),k=c.n(j),l=c(55013);let m=process.env.DATA_DIR||process.env.BLUEROUTER_DATA_DIR||g().join(k().homedir(),".bluerouter"),n=g().join(m,"ollama-nodes.json");async function o(){await e().mkdir(m,{recursive:!0});try{await e().access(n)}catch{await e().writeFile(n,JSON.stringify({nodes:[]},null,2))}}async function p(){await o();try{let a=await e().readFile(n,"utf8"),b=JSON.parse(a||"{}");return{nodes:Array.isArray(b.nodes)?b.nodes:[]}}catch{return{nodes:[]}}}async function q(a){await e().mkdir(m,{recursive:!0}),await e().writeFile(n,JSON.stringify({nodes:a.nodes||[]},null,2))}function r(){return new Date().toISOString()}function s(a,b={}){if(!a)return b;if("object"==typeof a&&!Array.isArray(a))return a;try{let c=JSON.parse(String(a));return c&&"object"==typeof c&&!Array.isArray(c)?c:b}catch{return b}}function t(a){return a?{...a,sshPassword:void 0,sudoPassword:void 0}:null}function u(a={},b={}){var c;let d=String(a.host??b.host??"").trim(),e=Number(a.agentPort??b.agentPort??19999)||19999,f=Array.isArray(c=a.ollamaPorts??b.ollamaPorts??[11434])?[...new Set(c.map(a=>Number(a)).filter(a=>Number.isFinite(a)&&a>0))]:[...new Set(String(c||"11434").split(/[\s,;]+/).map(a=>Number(a.trim())).filter(a=>Number.isFinite(a)&&a>0))],g=s(a.modelMemoryRequirements??b.modelMemoryRequirements,b.modelMemoryRequirements||{}),h=(0,l.nw)(g,{partial:!0,preserveEnabled:!0}),j=b.id||a.id||i().randomUUID(),k=String(a.agentToken??b.agentToken??i().randomBytes(24).toString("hex")).trim(),m=function(a,b,c={}){let d=new Map((Array.isArray(c.instances)?c.instances:[]).map(a=>[Number(a.port),a])),e=new Map((Array.isArray(a)?a:[]).map(a=>[Number(a.port),a]));return b.map(a=>(function(a={},b={}){var c;let d=Number(a.port??b.port??11434)||11434;return{id:String(a.id??b.id??`ollama-${d}`).trim()||`ollama-${d}`,name:String(a.name??b.name??`Ollama :${d}`).trim()||`Ollama :${d}`,port:d,enabled:a.enabled??b.enabled??!0,gpuMode:["auto","cpu","specific","all"].includes(a.gpuMode??b.gpuMode)?a.gpuMode??b.gpuMode:"auto",gpuIndices:Array.isArray(c=a.gpuIndices??b.gpuIndices??[])?[...new Set(c.map(a=>Number(a)).filter(a=>Number.isInteger(a)&&a>=0))]:[...new Set(String(c||"").split(/[\s,;]+/).map(a=>Number(a)).filter(a=>Number.isInteger(a)&&a>=0))],allowMultiGpuSplit:a.allowMultiGpuSplit??b.allowMultiGpuSplit??!1,tags:Array.isArray(a.tags??b.tags)?(a.tags??b.tags).map(a=>String(a).trim()).filter(Boolean):[],modelMemoryRequirements:(0,l.nw)(s(a.modelMemoryRequirements??b.modelMemoryRequirements,b.modelMemoryRequirements||{}),{partial:!0,preserveEnabled:!0})}})(e.get(Number(a))||{},d.get(Number(a))||{port:a}))}(a.instances??b.instances,f.length?f:[11434],b);return{id:j,name:String(a.name??b.name??d??"Ollama Node").trim()||d||"Ollama Node",host:d,sshPort:Number(a.sshPort??b.sshPort??22)||22,sshUser:String(a.sshUser??b.sshUser??"root").trim()||"root",useSudo:a.useSudo??b.useSudo??!1,agentPort:e,agentToken:k,ollamaPorts:f.length?f:[11434],instances:m,refreshIntervalMs:Math.max(5e3,Number(a.refreshIntervalMs??b.refreshIntervalMs??15e3)||15e3),enabled:a.enabled??b.enabled??!0,ufwAllowFrom:String(a.ufwAllowFrom??b.ufwAllowFrom??"").trim(),useGlobalModelMemory:a.useGlobalModelMemory??b.useGlobalModelMemory??!0,modelMemoryRequirements:h,notes:String(a.notes??b.notes??"").trim(),lastStatus:b.lastStatus||null,lastEvaluation:b.lastEvaluation||null,lastStatusAt:b.lastStatusAt||null,lastError:b.lastError||null,lastInstallAt:b.lastInstallAt||null,lastInstallOk:b.lastInstallOk||!1,lastInstallLog:b.lastInstallLog||"",createdAt:b.createdAt||r(),updatedAt:r()}}async function v(){return(await p()).nodes.map(a=>t(u(a,a)))}async function w(a){let b=(await p()).nodes.find(b=>b.id===a)||null;return b?u(b,b):null}async function x(a){let b=await p(),c=u(a);return b.nodes.push(c),await q(b),t(c)}async function y(a,b){let c=await p(),d=c.nodes.findIndex(b=>b.id===a);if(d<0)return null;let e=u(b,c.nodes[d]);return c.nodes[d]=e,await q(c),t(e)}async function z(a,b){let c=await p(),d=c.nodes.findIndex(b=>b.id===a);return d<0?null:(c.nodes[d]={...c.nodes[d],...b,updatedAt:r()},await q(c),t(u(c.nodes[d],c.nodes[d])))}async function A(a){let b=await p(),c=b.nodes.length;return b.nodes=b.nodes.filter(b=>b.id!==a),await q(b),b.nodes.length!==c}},10846:a=>{"use strict";a.exports=require("next/dist/compiled/next-server/app-page.runtime.prod.js")},21820:a=>{"use strict";a.exports=require("os")},29294:a=>{"use strict";a.exports=require("next/dist/server/app-render/work-async-storage.external.js")},33873:a=>{"use strict";a.exports=require("path")},44870:a=>{"use strict";a.exports=require("next/dist/compiled/next-server/app-route.runtime.prod.js")},55013:(a,b,c)=>{"use strict";c.d(b,{LM:()=>v,LQ:()=>u,MW:()=>s,dV:()=>l,eY:()=>t,iP:()=>m,nw:()=>o});var d=c(79748),e=c.n(d),f=c(33873),g=c.n(f),h=c(21820),i=c.n(h);let j=process.env.DATA_DIR||process.env.BLUEROUTER_DATA_DIR||g().join(i().homedir(),".bluerouter"),k=g().join(j,"ollama-model-memory.json"),l={"qwen3:1.7b":{minRamFreeMb:2048,minVramFreeMb:0,loadedRamFreeMb:512,loadedVramFreeMb:0,safetyMarginMb:512,notes:"small text model"},"qwen3:4b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:0,safetyMarginMb:512,notes:"small/medium text model"},"qwen3:8b":{minRamFreeMb:6144,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:512,safetyMarginMb:1024,notes:"8B text model"},"qwen3:14b":{minRamFreeMb:8192,minVramFreeMb:1e4,loadedRamFreeMb:2048,loadedVramFreeMb:1024,safetyMarginMb:1024,notes:"14B class model"},"qwen3:32b":{minRamFreeMb:4096,minVramFreeMb:16e3,loadedRamFreeMb:2048,loadedVramFreeMb:2048,safetyMarginMb:2048,notes:"32B class model, GPU preferred"},"qwen3-coder:30b":{minRamFreeMb:8192,minVramFreeMb:24e3,loadedRamFreeMb:2048,loadedVramFreeMb:3072,safetyMarginMb:2048,notes:"30B coder"},"qwen3-coder:latest":{minRamFreeMb:8192,minVramFreeMb:24e3,loadedRamFreeMb:2048,loadedVramFreeMb:3072,safetyMarginMb:2048,notes:"qwen3 coder latest, treat as 30B unless overridden"},"qwen2.5-coder:14b":{minRamFreeMb:8192,minVramFreeMb:1e4,loadedRamFreeMb:2048,loadedVramFreeMb:1024,safetyMarginMb:1024,notes:"14B coder"},"qwen2.5vl:7b":{minRamFreeMb:8192,minVramFreeMb:8e3,loadedRamFreeMb:2048,loadedVramFreeMb:1024,safetyMarginMb:1024,notes:"7B vision-language"},"qwen2.5vl:latest":{minRamFreeMb:8192,minVramFreeMb:1e4,loadedRamFreeMb:2048,loadedVramFreeMb:1024,safetyMarginMb:1024,notes:"vision-language latest"},"bge-m3:latest":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:512,loadedVramFreeMb:0,safetyMarginMb:512,notes:"embedding model"},"deepseek-coder:6.7b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:512,safetyMarginMb:1024,notes:"6.7B coder"},"gemma4:e4b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:0,safetyMarginMb:1024,notes:"small Gemma variant"},"gemma:7b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:0,safetyMarginMb:1024,notes:"7B text model"},"glm-4.7-flash:latest":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:512,safetyMarginMb:1024,notes:"flash/local text model"},"gpt-oss:20b":{minRamFreeMb:8192,minVramFreeMb:12e3,loadedRamFreeMb:2048,loadedVramFreeMb:2048,safetyMarginMb:2048,notes:"20B class model"},"llama3.1:8b":{minRamFreeMb:6144,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:512,safetyMarginMb:1024,notes:"8B text model"},"llama3.2-vision:latest":{minRamFreeMb:8192,minVramFreeMb:1e4,loadedRamFreeMb:2048,loadedVramFreeMb:1024,safetyMarginMb:1024,notes:"vision model"},"qwen3-*-translate-f16:*":{minRamFreeMb:8192,minVramFreeMb:16e3,loadedRamFreeMb:2048,loadedVramFreeMb:2048,safetyMarginMb:2048,notes:"custom Qwen3 f16 translate wildcard"},"qwen3-*-translate*:latest":{minRamFreeMb:6144,minVramFreeMb:8e3,loadedRamFreeMb:1024,loadedVramFreeMb:1024,safetyMarginMb:1024,notes:"custom Qwen3 translate wildcard"},"bge-m3:*":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:512,loadedVramFreeMb:0,safetyMarginMb:512,notes:"embedding model wildcard"},"gemma*:e4b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:0,safetyMarginMb:1024,notes:"Gemma E4B wildcard"},"gemma*:7b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:0,safetyMarginMb:1024,notes:"Gemma 7B wildcard"},"*:1.7b":{minRamFreeMb:2048,minVramFreeMb:0,loadedRamFreeMb:512,loadedVramFreeMb:0,safetyMarginMb:512,notes:"generic 1.7B fallback"},"*:4b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:0,safetyMarginMb:512,notes:"generic 4B fallback"},"*:6.7b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:512,safetyMarginMb:1024,notes:"generic 6-7B fallback"},"*:7b":{minRamFreeMb:4096,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:512,safetyMarginMb:1024,notes:"generic 7B fallback"},"*:8b":{minRamFreeMb:6144,minVramFreeMb:0,loadedRamFreeMb:1024,loadedVramFreeMb:512,safetyMarginMb:1024,notes:"generic 8B fallback"},"*:14b":{minRamFreeMb:8192,minVramFreeMb:1e4,loadedRamFreeMb:2048,loadedVramFreeMb:1024,safetyMarginMb:1024,notes:"generic 14B fallback"},"*:20b":{minRamFreeMb:8192,minVramFreeMb:12e3,loadedRamFreeMb:2048,loadedVramFreeMb:2048,safetyMarginMb:2048,notes:"generic 20B fallback"},"*:30b":{minRamFreeMb:8192,minVramFreeMb:24e3,loadedRamFreeMb:2048,loadedVramFreeMb:3072,safetyMarginMb:2048,notes:"generic 30B fallback"},"*:32b":{minRamFreeMb:4096,minVramFreeMb:16e3,loadedRamFreeMb:2048,loadedVramFreeMb:2048,safetyMarginMb:2048,notes:"generic 32B fallback"}},m=["minRamFreeMb","minVramFreeMb","loadedRamFreeMb","loadedVramFreeMb","safetyMarginMb"];function n(a,b=0){let c=Number(a);return Number.isFinite(c)&&c>=0?c:b}function o(a={},b={}){let c={};if(!a||"object"!=typeof a||Array.isArray(a))return c;for(let[d,e]of Object.entries(a)){let a=String(d||"").trim();if(!a||!e||"object"!=typeof e||Array.isArray(e))continue;let f=function(a,{partial:b=!1,preserveEnabled:c=!1}={}){let d={};for(let c of m)if(b){var e;null!=(e=a?.[c])&&""!==String(e).trim()&&(d[c]=n(a[c],0))}else d[c]=n(a?.[c],0);let f=String(a?.notes||"").trim();return f&&(d.notes=f),c&&a&&Object.prototype.hasOwnProperty.call(a,"enabled")&&(d.enabled=!1!==a.enabled),d}(e,b),g=m.some(a=>Object.prototype.hasOwnProperty.call(f,a)),h=b.preserveEnabled&&Object.prototype.hasOwnProperty.call(f,"enabled");(g||f.notes||h)&&(c[a]=f)}return c}async function p(){await e().mkdir(j,{recursive:!0});try{await e().access(k)}catch{await e().writeFile(k,JSON.stringify({requirements:l,updatedAt:new Date().toISOString()},null,2))}}async function q(){await p();try{let a=await e().readFile(k,"utf8"),b=JSON.parse(a||"{}");return{requirements:function(a={}){return o({...l,...a})}(b.requirements||{}),updatedAt:b.updatedAt||null}}catch{return{requirements:o(l),updatedAt:null}}}async function r(a){await e().mkdir(j,{recursive:!0});let b={requirements:o(a),updatedAt:new Date().toISOString()};return await e().writeFile(k,JSON.stringify(b,null,2)),b}async function s(){return(await q()).requirements}async function t(){return q()}async function u(a){return r(a)}async function v(){return r(l)}},55511:a=>{"use strict";a.exports=require("crypto")},63033:a=>{"use strict";a.exports=require("next/dist/server/app-render/work-unit-async-storage.external.js")},68757:(a,b,c)=>{"use strict";c.r(b),c.d(b,{handler:()=>J,patchFetch:()=>I,routeModule:()=>E,serverHooks:()=>H,workAsyncStorage:()=>F,workUnitAsyncStorage:()=>G});var d={};c.r(d),c.d(d,{POST:()=>D,dynamic:()=>C,runtime:()=>B});var e=c(19225),f=c(84006),g=c(8317),h=c(99373),i=c(34775),j=c(24235),k=c(261),l=c(54365),m=c(90771),n=c(73461),o=c(67798),p=c(92280),q=c(62018),r=c(45696),s=c(47929),t=c(86439),u=c(37527),v=c(23211),w=c(6774),x=c(79646);function y(a){return`'${String(a??"").replace(/'/g,"'\"'\"'")}'`}function z(a,b,{input:c="",timeoutMs:d=12e4}={}){return new Promise(e=>{let f=(0,x.spawn)(a,b,{stdio:["pipe","pipe","pipe"]}),g="",h="",i=!1,j=setTimeout(()=>{i||f.kill("SIGKILL")},d);f.stdout.on("data",a=>{g+=a.toString()}),f.stderr.on("data",a=>{h+=a.toString()}),f.on("close",a=>{i=!0,clearTimeout(j),e({ok:0===a,code:a,stdout:g,stderr:h})}),c&&f.stdin.write(c),f.stdin.end()})}async function A(a,b={}){let c=a.sshUser||"root",d=Number(a.sshPort||22),e=`${c}@${a.host}`,f=function(a,b={}){let c="/opt/bluerouter-ollama-agent",d=(a.ollamaPorts||[11434]).join(","),e=String.raw`#!/usr/bin/env python3
 import json
 import os
 import re
+import socket
 import subprocess
 import urllib.request
-from http.server import BaseHTTPRequestHandler, HTTPServer
 from datetime import datetime, timezone
+from http.server import BaseHTTPRequestHandler, HTTPServer
 
 PORT = int(os.environ.get("AGENT_PORT", "19999"))
 OLLAMA_HOST = os.environ.get("OLLAMA_HOST", "127.0.0.1")
@@ -13,13 +14,23 @@ OLLAMA_PORTS = [p.strip() for p in os.environ.get("OLLAMA_PORTS", "11434").split
 AGENT_TOKEN = os.environ.get("AGENT_TOKEN", "")
 
 
-def run_cmd(cmd, timeout=3):
+def run_cmd(cmd, timeout=4):
     try:
         return subprocess.check_output(cmd, stderr=subprocess.STDOUT, timeout=timeout, text=True).strip()
     except Exception:
         return None
 
 
+def run_json(cmd, timeout=6):
+    out = run_cmd(cmd, timeout=timeout)
+    if not out:
+        return None
+    try:
+        return json.loads(out)
+    except Exception:
+        return None
+
+
 def http_json(url, timeout=3):
     try:
         with urllib.request.urlopen(url, timeout=timeout) as r:
@@ -28,32 +39,83 @@ def http_json(url, timeout=3):
         return None
 
 
-def get_gpu_uuid_map():
-    out = run_cmd([
-        "nvidia-smi",
-        "--query-gpu=index,uuid,name",
-        "--format=csv,noheader,nounits",
-    ])
-    uuid_to_gpu = {}
-    if not out:
-        return uuid_to_gpu
-    for line in out.splitlines():
-        parts = [x.strip() for x in line.split(",")]
-        if len(parts) < 3:
+def read_file(path):
+    try:
+        with open(path, "rb") as f:
+            return f.read()
+    except Exception:
+        return b""
+
+
+def read_text(path):
+    try:
+        with open(path, "r", encoding="utf-8", errors="ignore") as f:
+            return f.read()
+    except Exception:
+        return ""
+
+
+def read_cmdline(pid):
+    raw = read_file(f"/proc/{pid}/cmdline")
+    if not raw:
+        return ""
+    return raw.replace(b"\x00", b" ").decode("utf-8", errors="ignore").strip()
+
+
+def read_environ(pid):
+    raw = read_file(f"/proc/{pid}/environ")
+    env = {}
+    if not raw:
+        return env
+    for item in raw.split(b"\x00"):
+        if not item or b"=" not in item:
             continue
-        try:
-            idx, uuid, name = parts[:3]
-            uuid_to_gpu[uuid] = {"index": int(float(idx)), "uuid": uuid, "name": name}
-        except Exception:
+        k, v = item.split(b"=", 1)
+        env[k.decode("utf-8", errors="ignore")] = v.decode("utf-8", errors="ignore")
+    return env
+
+
+def parse_systemd_environment(text):
+    # systemctl show -p Environment usually returns: Environment=A=1 B=two C=/x/y
+    env = {}
+    if not text:
+        return env
+    if text.startswith("Environment="):
+        text = text[len("Environment="):]
+    for part in re.findall(r'(?:[^\s"\']+|"[^"]*"|\'[^\']*\')+', text):
+        item = part.strip().strip('"').strip("'")
+        if "=" not in item:
             continue
-    return uuid_to_gpu
+        k, v = item.split("=", 1)
+        if k:
+            env[k] = v
+    return env
+
+
+def get_service_from_cgroup(pid):
+    text = read_text(f"/proc/{pid}/cgroup")
+    m = re.search(r'/system\.slice/([^/]+\.service)', text)
+    if m:
+        return m.group(1)
+    # user services may appear under user.slice/user-1000.slice/.../app.slice/name.service
+    m = re.search(r'/([^/]+\.service)(?:$|/)', text)
+    if m:
+        return m.group(1)
+    return None
+
+
+def get_systemd_env(service):
+    if not service:
+        return {}
+    out = run_cmd(["systemctl", "show", service, "-p", "Environment"], timeout=3)
+    return parse_systemd_environment(out or "")
 
 
 def get_gpu_status():
     out = run_cmd([
         "nvidia-smi",
         "--query-gpu=index,uuid,name,memory.total,memory.used,memory.free,utilization.gpu,temperature.gpu",
-        "--format=csv,noheader,nounits",
+        "--format=csv,noheader,nounits"
     ])
     if not out:
         return []
@@ -74,105 +136,13 @@ def get_gpu_status():
                 "vramFreeMb": free,
                 "vramUsedPercent": round((used / total) * 100, 2) if total else None,
                 "gpuUsagePercent": int(float(util)),
-                "temperatureC": int(float(temp)),
+                "temperatureC": int(float(temp))
             })
         except Exception:
             pass
     return gpus
 
 
-def get_gpu_processes():
-    """Return process-level GPU usage from nvidia-smi compute apps.
-
-    This is the key signal used to infer port -> PID -> GPU index. It is still
-    process-level, not per-model. If one Ollama process serves multiple models,
-    the mapping says which GPUs that port/process uses, not which model uses each GPU.
-    """
-    uuid_map = get_gpu_uuid_map()
-    out = run_cmd([
-        "nvidia-smi",
-        "--query-compute-apps=gpu_uuid,pid,process_name,used_memory",
-        "--format=csv,noheader,nounits",
-    ])
-    processes = []
-    if not out:
-        return processes
-    for line in out.splitlines():
-        parts = [x.strip() for x in line.split(",")]
-        if len(parts) < 4:
-            continue
-        uuid, pid, process_name, used_memory = parts[:4]
-        try:
-            pid_i = int(float(pid))
-            used_mb = int(float(str(used_memory).replace("MiB", "").strip()))
-            gpu = uuid_map.get(uuid, {})
-            processes.append({
-                "gpuUuid": uuid,
-                "gpuIndex": gpu.get("index"),
-                "gpuName": gpu.get("name"),
-                "pid": pid_i,
-                "processName": process_name,
-                "usedMemoryMb": used_mb,
-            })
-        except Exception:
-            continue
-    return processes
-
-
-def get_process_name(pid):
-    if not pid:
-        return None
-    out = run_cmd(["ps", "-p", str(pid), "-o", "comm="], timeout=2)
-    return out.strip() if out else None
-
-
-def get_listening_pid(port):
-    """Best-effort Linux TCP listener PID detection for a local Ollama port."""
-    port_s = str(port)
-
-    # ss is installed on most modern Linux distributions.
-    out = run_cmd(["ss", "-ltnp", f"sport = :{port_s}"], timeout=3)
-    if out:
-        # users:(("ollama",pid=1234,fd=3))
-        m = re.search(r"pid=(\d+)", out)
-        if m:
-            try:
-                return int(m.group(1))
-            except Exception:
-                pass
-
-    # Fallback to lsof when available.
-    out = run_cmd(["lsof", "-nP", f"-iTCP:{port_s}", "-sTCP:LISTEN", "-t"], timeout=3)
-    if out:
-        for line in out.splitlines():
-            line = line.strip()
-            if line.isdigit():
-                return int(line)
-
-    # Last-resort netstat parse.
-    out = run_cmd(["netstat", "-ltnp"], timeout=3)
-    if out:
-        for line in out.splitlines():
-            if f":{port_s} " not in line:
-                continue
-            m = re.search(r"\s(\d+)/", line)
-            if m:
-                return int(m.group(1))
-    return None
-
-
-def gpu_binding_for_pid(pid, gpu_processes):
-    if not pid:
-        return {"gpuIndices": [], "gpuVramUsedMb": 0, "gpuProcesses": []}
-    matched = [p for p in gpu_processes if int(p.get("pid") or -1) == int(pid)]
-    indices = sorted({p.get("gpuIndex") for p in matched if p.get("gpuIndex") is not None})
-    return {
-        "gpuIndices": indices,
-        "gpuVramUsedMb": sum(int(p.get("usedMemoryMb") or 0) for p in matched),
-        "gpuProcesses": matched,
-    }
-
-
 def get_ram_status():
     out = run_cmd(["free", "-m"])
     if not out:
@@ -187,7 +157,7 @@ def get_ram_status():
                 "usedMb": used,
                 "freeMb": free,
                 "availableMb": avail,
-                "usedPercent": round((used / total) * 100, 2) if total else None,
+                "usedPercent": round((used / total) * 100, 2) if total else None
             }
     return None
 
@@ -200,10 +170,39 @@ def get_cpu_status():
         "load1": load[0],
         "load5": load[1],
         "load15": load[2],
-        "load1PercentOfCores": round((load[0] / cores) * 100, 2),
+        "load1PercentOfCores": round((load[0] / cores) * 100, 2)
     }
 
 
+def get_gpu_processes(gpus):
+    uuid_to_index = {g.get("uuid"): g.get("index") for g in gpus if g.get("uuid")}
+    out = run_cmd([
+        "nvidia-smi",
+        "--query-compute-apps=gpu_uuid,pid,process_name,used_memory",
+        "--format=csv,noheader,nounits"
+    ])
+    processes = []
+    if out:
+        for line in out.splitlines():
+            parts = [x.strip() for x in line.split(",")]
+            if len(parts) < 4:
+                continue
+            gpu_uuid, pid, process_name, used_memory = parts[:4]
+            try:
+                pid_n = int(float(pid))
+                used_n = int(float(str(used_memory).replace("MiB", "").strip()))
+            except Exception:
+                continue
+            processes.append({
+                "gpuUuid": gpu_uuid,
+                "gpuIndex": uuid_to_index.get(gpu_uuid),
+                "pid": pid_n,
+                "processName": process_name,
+                "usedMemoryMb": used_n
+            })
+    return processes
+
+
 def compact_model(item):
     if not isinstance(item, dict):
         return None
@@ -215,11 +214,401 @@ def compact_model(item):
         "size_vram": item.get("size_vram"),
         "expires_at": item.get("expires_at"),
         "context_length": item.get("context_length"),
-        "details": item.get("details") or {},
+        "details": item.get("details") or {}
+    }
+
+
+def find_listener_pid_by_ss(port):
+    candidates = []
+    commands = [
+        ["ss", "-H", "-ltnp", "sport", "=", f":{port}"],
+        ["ss", "-H", "-ltnp"],
+    ]
+    for cmd in commands:
+        out = run_cmd(cmd, timeout=3)
+        if not out:
+            continue
+        for line in out.splitlines():
+            if f":{port}" not in line:
+                continue
+            for pid in re.findall(r'pid=([0-9]+)', line):
+                candidates.append({"pid": int(pid), "source": "ss", "line": line})
+    return candidates
+
+
+def find_listener_pid_by_lsof(port):
+    candidates = []
+    out = run_cmd(["lsof", "-nP", f"-iTCP:{port}", "-sTCP:LISTEN"], timeout=4)
+    if not out:
+        return candidates
+    for line in out.splitlines()[1:]:
+        parts = line.split()
+        if len(parts) >= 2 and parts[1].isdigit():
+            candidates.append({"pid": int(parts[1]), "source": "lsof", "line": line})
+    return candidates
+
+
+def find_listener_pid_by_fuser(port):
+    candidates = []
+    out = run_cmd(["fuser", "-n", "tcp", str(port)], timeout=4)
+    if not out:
+        return candidates
+    for pid in re.findall(r'\b[0-9]+\b', out):
+        candidates.append({"pid": int(pid), "source": "fuser", "line": out})
+    return candidates
+
+
+def tcp_port_hex(port):
+    return f"{int(port):04X}"
+
+
+def listener_inodes_for_port(port):
+    inodes = set()
+    want = tcp_port_hex(port)
+    for path in ["/proc/net/tcp", "/proc/net/tcp6"]:
+        text = read_text(path)
+        for line in text.splitlines()[1:]:
+            cols = line.split()
+            if len(cols) < 10:
+                continue
+            local = cols[1]
+            state = cols[3]
+            inode = cols[9]
+            if state != "0A":
+                continue
+            if local.split(":")[-1].upper() == want:
+                inodes.add(inode)
+    return inodes
+
+
+def find_listener_pid_by_proc(port):
+    candidates = []
+    inodes = listener_inodes_for_port(port)
+    if not inodes:
+        return candidates
+    proc_root = "/proc"
+    for pid in os.listdir(proc_root):
+        if not pid.isdigit():
+            continue
+        fd_dir = os.path.join(proc_root, pid, "fd")
+        try:
+            fds = os.listdir(fd_dir)
+        except Exception:
+            continue
+        for fd in fds:
+            try:
+                target = os.readlink(os.path.join(fd_dir, fd))
+            except Exception:
+                continue
+            m = re.match(r'socket:\[([0-9]+)\]', target)
+            if m and m.group(1) in inodes:
+                candidates.append({"pid": int(pid), "source": "proc-net", "line": f"inode={m.group(1)} fd={fd}"})
+                break
+    return candidates
+
+
+def find_listener_pid(port):
+    all_candidates = []
+    for fn in [find_listener_pid_by_ss, find_listener_pid_by_lsof, find_listener_pid_by_fuser, find_listener_pid_by_proc]:
+        try:
+            all_candidates.extend(fn(port))
+        except Exception:
+            pass
+    seen = set()
+    unique = []
+    for c in all_candidates:
+        pid = c.get("pid")
+        if not pid or pid in seen:
+            continue
+        seen.add(pid)
+        unique.append(c)
+    if not unique:
+        return None, None, []
+    chosen = unique[0]
+    return chosen.get("pid"), chosen.get("source"), unique
+
+
+def process_tree_pids(root_pid):
+    try:
+        root_pid = int(root_pid)
+    except Exception:
+        return []
+    out = run_cmd(["ps", "-eo", "pid=,ppid=,cmd="], timeout=4)
+    children = {}
+    if out:
+        for line in out.splitlines():
+            parts = line.strip().split(None, 2)
+            if len(parts) < 2:
+                continue
+            try:
+                pid = int(parts[0]); ppid = int(parts[1])
+            except Exception:
+                continue
+            children.setdefault(ppid, []).append(pid)
+    result = []
+    stack = [root_pid]
+    seen = set()
+    while stack:
+        pid = stack.pop()
+        if pid in seen:
+            continue
+        seen.add(pid)
+        result.append(pid)
+        stack.extend(children.get(pid, []))
+    return result
+
+
+def docker_ps_rows():
+    out = run_cmd(["docker", "ps", "--format", "{{.ID}}\t{{.Names}}\t{{.Ports}}"], timeout=5)
+    rows = []
+    if not out:
+        return rows
+    for line in out.splitlines():
+        parts = line.split("\t")
+        if len(parts) >= 3:
+            rows.append({"id": parts[0], "name": parts[1], "ports": parts[2]})
+    return rows
+
+
+def docker_inspect(cid):
+    data = run_json(["docker", "inspect", cid], timeout=6)
+    if isinstance(data, list) and data:
+        return data[0]
+    return None
+
+
+def find_docker_container_for_port(port, docker_proxy_info=None):
+    # Prefer Docker's authoritative NetworkSettings.Ports mapping.
+    for row in docker_ps_rows():
+        info = docker_inspect(row["id"])
+        if not info:
+            continue
+        ports = (((info.get("NetworkSettings") or {}).get("Ports")) or {})
+        for container_port, host_bindings in ports.items():
+            if not str(container_port).startswith(str(port)) and not str(container_port).startswith(str((docker_proxy_info or {}).get("containerPort", ""))):
+                pass
+            for binding in host_bindings or []:
+                if str(binding.get("HostPort")) == str(port):
+                    return row, info
+        # Fallback: match container IP parsed from docker-proxy command.
+        container_ip = (docker_proxy_info or {}).get("containerIp")
+        if container_ip:
+            networks = ((info.get("NetworkSettings") or {}).get("Networks")) or {}
+            for net in networks.values():
+                if net.get("IPAddress") == container_ip:
+                    return row, info
+        # Fallback: match docker ps formatted Ports text.
+        if f":{port}->" in row.get("ports", "") or f"0.0.0.0:{port}" in row.get("ports", "") or f"[::]:{port}" in row.get("ports", ""):
+            return row, info
+    return None, None
+
+
+def parse_docker_proxy_cmd(cmdline):
+    if "docker-proxy" not in (cmdline or ""):
+        return None
+    tokens = cmdline.split()
+    out = {"kind": "docker-proxy"}
+    for i, token in enumerate(tokens):
+        if token in ["-host-port", "--host-port"] and i + 1 < len(tokens):
+            out["hostPort"] = tokens[i + 1]
+        elif token in ["-container-port", "--container-port"] and i + 1 < len(tokens):
+            out["containerPort"] = tokens[i + 1]
+        elif token in ["-container-ip", "--container-ip"] and i + 1 < len(tokens):
+            out["containerIp"] = tokens[i + 1]
+    return out
+
+
+def docker_env_and_meta(port, pid, cmdline):
+    proxy = parse_docker_proxy_cmd(cmdline)
+    if not proxy:
+        return None
+    row, info = find_docker_container_for_port(port, proxy)
+    if not info:
+        return {
+            "runtimeType": "docker-proxy",
+            "dockerProxy": proxy,
+            "containerFound": False,
+            "env": {},
+            "source": "docker-proxy-no-container"
+        }
+    env_list = ((info.get("Config") or {}).get("Env")) or []
+    env = {}
+    for item in env_list:
+        if "=" in item:
+            k, v = item.split("=", 1)
+            env[k] = v
+    host_config = info.get("HostConfig") or {}
+    state = info.get("State") or {}
+    labels = (info.get("Config") or {}).get("Labels") or {}
+    networks = ((info.get("NetworkSettings") or {}).get("Networks")) or {}
+    ips = []
+    for net_name, net in networks.items():
+        if net.get("IPAddress"):
+            ips.append({"network": net_name, "ip": net.get("IPAddress")})
+    return {
+        "runtimeType": "docker",
+        "dockerProxy": proxy,
+        "containerFound": True,
+        "containerId": (info.get("Id") or row.get("id"))[:12],
+        "containerName": (info.get("Name") or row.get("name") or "").lstrip("/"),
+        "containerImage": (info.get("Config") or {}).get("Image"),
+        "containerPid": state.get("Pid"),
+        "containerIps": ips,
+        "dockerPorts": row.get("ports"),
+        "dockerNetworkMode": host_config.get("NetworkMode"),
+        "dockerDeviceRequests": host_config.get("DeviceRequests") or [],
+        "dockerLabels": labels,
+        "env": env,
+        "source": "docker-inspect"
+    }
+
+
+def parse_visible_devices(value, gpus, source):
+    if value is None:
+        return None
+    raw = str(value).strip()
+    lower = raw.lower()
+    if raw == "":
+        return None
+    if lower in ["none", "void", "no", "false", "-1"]:
+        return {"indices": [], "mode": "disabled", "source": source, "raw": raw}
+    if lower == "all":
+        return {"indices": [g["index"] for g in gpus], "mode": "all-visible", "source": source, "raw": raw}
+    uuid_to_index = {str(g.get("uuid")): g.get("index") for g in gpus if g.get("uuid")}
+    indices = []
+    unknown = []
+    for item in re.split(r'[,;\s]+', raw):
+        if not item:
+            continue
+        if item.isdigit():
+            indices.append(int(item))
+        elif item in uuid_to_index:
+            indices.append(uuid_to_index[item])
+        elif item.startswith("GPU-") and item in uuid_to_index:
+            indices.append(uuid_to_index[item])
+        elif item.startswith("MIG-"):
+            unknown.append(item)
+        else:
+            unknown.append(item)
+    return {
+        "indices": sorted(list(set(indices))),
+        "mode": "specific" if indices else "unknown",
+        "source": source,
+        "raw": raw,
+        "unknownDevices": unknown
     }
 
 
-def get_ollama_instance(port, gpu_processes):
+def parse_device_requests(device_requests, gpus):
+    if not device_requests:
+        return None
+    ids = []
+    all_gpu = False
+    for req in device_requests:
+        count = req.get("Count")
+        caps = req.get("Capabilities") or []
+        driver = req.get("Driver")
+        req_ids = req.get("DeviceIDs") or []
+        if driver == "nvidia" or any("gpu" in str(x).lower() for cap in caps for x in cap):
+            if count == -1 and not req_ids:
+                all_gpu = True
+            ids.extend(req_ids)
+    if all_gpu:
+        return {"indices": [g["index"] for g in gpus], "mode": "all-visible", "source": "docker-device-requests:all", "raw": "all"}
+    if ids:
+        return parse_visible_devices(",".join(ids), gpus, "docker-device-requests")
+    return None
+
+
+def configured_gpu_from_env(runtime_type, env, gpus, docker_meta=None):
+    # Prefer NVIDIA_VISIBLE_DEVICES in Docker, CUDA_VISIBLE_DEVICES for direct/systemd.
+    candidates = []
+    if runtime_type == "docker":
+        candidates = [
+            ("NVIDIA_VISIBLE_DEVICES", env.get("NVIDIA_VISIBLE_DEVICES")),
+            ("CUDA_VISIBLE_DEVICES", env.get("CUDA_VISIBLE_DEVICES")),
+        ]
+        req = parse_device_requests((docker_meta or {}).get("dockerDeviceRequests"), gpus)
+        if req:
+            return req
+    else:
+        candidates = [
+            ("CUDA_VISIBLE_DEVICES", env.get("CUDA_VISIBLE_DEVICES")),
+            ("NVIDIA_VISIBLE_DEVICES", env.get("NVIDIA_VISIBLE_DEVICES")),
+            ("HIP_VISIBLE_DEVICES", env.get("HIP_VISIBLE_DEVICES")),
+            ("ROCR_VISIBLE_DEVICES", env.get("ROCR_VISIBLE_DEVICES")),
+        ]
+    for key, value in candidates:
+        parsed = parse_visible_devices(value, gpus, f"{runtime_type}-env:{key}")
+        if parsed:
+            return parsed
+    if runtime_type in ["systemd", "direct"]:
+        return {"indices": [g["index"] for g in gpus], "mode": "all-visible", "source": "all-visible-no-env", "raw": None}
+    if runtime_type == "docker":
+        # For Docker, lack of NVIDIA_VISIBLE_DEVICES is less conclusive.
+        return {"indices": [], "mode": "unknown", "source": "docker-no-visible-devices-env", "raw": None}
+    return {"indices": [], "mode": "unknown", "source": "no-gpu-config", "raw": None}
+
+
+def process_metadata(port, pid, gpus):
+    if not pid:
+        return {"runtimeType": "unknown", "env": {}, "configured": configured_gpu_from_env("unknown", {}, gpus)}
+    cmdline = read_cmdline(pid)
+    service = get_service_from_cgroup(pid)
+    runtime_type = "systemd" if service else "direct"
+    docker_meta = docker_env_and_meta(port, pid, cmdline)
+    if docker_meta:
+        runtime_type = docker_meta.get("runtimeType") or "docker"
+        env = docker_meta.get("env") or {}
+        configured = configured_gpu_from_env("docker", env, gpus, docker_meta)
+        return {
+            "runtimeType": runtime_type,
+            "processCmdline": cmdline,
+            "serviceName": service,
+            "env": env,
+            "configured": configured,
+            **{k: v for k, v in docker_meta.items() if k not in ["env"]}
+        }
+    env = {}
+    env.update(get_systemd_env(service))
+    env.update(read_environ(pid))
+    configured = configured_gpu_from_env("systemd" if service else "direct", env, gpus)
+    return {
+        "runtimeType": runtime_type,
+        "processCmdline": cmdline,
+        "serviceName": service,
+        "env": {k: v for k, v in env.items() if re.search(r'OLLAMA|CUDA|NVIDIA|HIP|ROCR|GPU', k, re.I)},
+        "configured": configured,
+    }
+
+
+def gpu_processes_for_pids(gpu_processes, pids):
+    pid_set = set(int(x) for x in pids if x)
+    return [p for p in gpu_processes if int(p.get("pid") or -1) in pid_set]
+
+
+def infer_gpu_from_ps(ps_models, gpus, configured):
+    size_bytes = sum(int(m.get("size_vram") or 0) for m in ps_models or [])
+    size_mb = int(size_bytes / (1024 * 1024)) if size_bytes else 0
+    if size_mb <= 0:
+        return [], 0, None
+    configured_indices = configured.get("indices") or [] if configured else []
+    if configured_indices:
+        return configured_indices, size_mb, "ollama-ps-configured-gpu-inferred"
+    if len(gpus) == 1:
+        return [gpus[0]["index"]], size_mb, "ollama-ps-single-gpu-inferred"
+    # Conservative multi-GPU inference: only infer if one GPU's used memory is close enough.
+    candidates = []
+    for gpu in gpus:
+        used = int(gpu.get("vramUsedMb") or 0)
+        if used >= max(512, int(size_mb * 0.45)):
+            candidates.append(gpu)
+    if len(candidates) == 1:
+        return [candidates[0]["index"]], size_mb, "ollama-ps-single-used-gpu-inferred"
+    return [], 0, None
+
+
+def get_ollama_instance(port, gpus, gpu_processes):
     base = f"http://{OLLAMA_HOST}:{port}"
     tags = http_json(f"{base}/api/tags")
     ps = http_json(f"{base}/api/ps")
@@ -242,36 +631,88 @@ def get_ollama_instance(port, gpu_processes):
                 ps_models.append(cm)
                 loaded_models.append(cm["name"])
 
-    pid = get_listening_pid(port)
-    binding = gpu_binding_for_pid(pid, gpu_processes)
+    pid, pid_source, pid_candidates = find_listener_pid(port)
+    meta = process_metadata(port, pid, gpus)
+    pid_tree = process_tree_pids(pid) if pid else []
+    # Docker containers may expose a container init PID whose children hold GPU handles.
+    container_pid = meta.get("containerPid")
+    if container_pid:
+        pid_tree = sorted(list(set(pid_tree + process_tree_pids(container_pid))))
+    matched_gpu_processes = gpu_processes_for_pids(gpu_processes, pid_tree)
+    runtime_indices = sorted(list(set([p.get("gpuIndex") for p in matched_gpu_processes if p.get("gpuIndex") is not None])))
+    runtime_used = sum(int(p.get("usedMemoryMb") or 0) for p in matched_gpu_processes)
+    runtime_source = "port-pid-nvidia-smi" if runtime_indices else None
+    ps_indices, ps_used, ps_source = infer_gpu_from_ps(ps_models, gpus, meta.get("configured") or {})
+
+    configured = meta.get("configured") or {}
+    effective_indices = []
+    effective_source = None
+    if configured.get("indices"):
+        effective_indices = configured.get("indices")
+        effective_source = configured.get("source")
+    elif runtime_indices:
+        effective_indices = runtime_indices
+        effective_source = runtime_source
+    elif ps_indices:
+        effective_indices = ps_indices
+        effective_source = ps_source
+
+    detected_indices = runtime_indices or ps_indices
+    detected_used = runtime_used if runtime_indices else ps_used
+    detected_source = runtime_source or ps_source or ("port-pid-no-gpu" if pid else None)
 
     return {
         "id": f"ollama-{port}",
         "port": int(port),
         "baseUrl": base,
         "alive": tags is not None,
-        "pid": pid,
-        "processName": get_process_name(pid),
         "models": models,
         "modelDetails": model_details,
         "loadedModels": loaded_models,
         "psModels": ps_models,
         "loadedDigests": sorted(list(set([m.get("digest") for m in ps_models if m.get("digest")]))),
-        "detectedGpuIndices": binding["gpuIndices"],
-        "detectedGpuVramUsedMb": binding["gpuVramUsedMb"],
-        "detectedGpuProcesses": binding["gpuProcesses"],
-        "detectedGpuBindingSource": "port-pid-nvidia-smi" if pid and binding["gpuIndices"] else ("port-pid-no-gpu" if pid else "unknown"),
+        "pid": pid,
+        "pidDetectionSource": pid_source,
+        "pidCandidates": pid_candidates,
+        "pidTree": pid_tree,
+        "runtimeType": meta.get("runtimeType"),
+        "serviceName": meta.get("serviceName"),
+        "processCmdline": meta.get("processCmdline"),
+        "processEnv": meta.get("env"),
+        "containerId": meta.get("containerId"),
+        "containerName": meta.get("containerName"),
+        "containerImage": meta.get("containerImage"),
+        "containerPid": meta.get("containerPid"),
+        "containerIps": meta.get("containerIps"),
+        "dockerNetworkMode": meta.get("dockerNetworkMode"),
+        "dockerPorts": meta.get("dockerPorts"),
+        "dockerDeviceRequests": meta.get("dockerDeviceRequests"),
+        "dockerLabels": meta.get("dockerLabels"),
+        "dockerProxy": meta.get("dockerProxy"),
+        "configuredGpuIndices": configured.get("indices") or [],
+        "configuredGpuMode": configured.get("mode"),
+        "configuredGpuSource": configured.get("source"),
+        "configuredGpuRaw": configured.get("raw"),
+        "configuredGpuUnknownDevices": configured.get("unknownDevices") or [],
+        "runtimeGpuIndices": runtime_indices,
+        "runtimeGpuVramUsedMb": runtime_used,
+        "runtimeGpuProcesses": matched_gpu_processes,
+        "detectedGpuIndices": detected_indices,
+        "detectedGpuVramUsedMb": detected_used,
+        "detectedGpuProcesses": matched_gpu_processes,
+        "detectedGpuBindingSource": detected_source,
+        "effectiveGpuIndices": effective_indices,
+        "effectiveGpuSource": effective_source,
     }
 
 
 def build_status():
     hostname = run_cmd(["hostname"]) or "unknown"
     gpus = get_gpu_status()
-    gpu_processes = get_gpu_processes()
+    gpu_processes = get_gpu_processes(gpus)
     ram = get_ram_status()
     cpu = get_cpu_status()
-    instances = [get_ollama_instance(p, gpu_processes) for p in OLLAMA_PORTS]
-
+    instances = [get_ollama_instance(p, gpus, gpu_processes) for p in OLLAMA_PORTS]
     tags = ["ollama", "lan"]
     if gpus:
         tags += ["gpu", "nvidia"]
@@ -279,11 +720,9 @@ def build_status():
             tags.append(f"gpu:{gpu.get('index')}:{str(gpu.get('name','')).lower().replace(' ','-')}")
     else:
         tags.append("cpu")
-
     for inst in instances:
         for model in inst.get("models", []):
             tags.append(f"model_support:{model}")
-
     best_gpu = max(gpus, key=lambda x: x.get("vramFreeMb", 0)) if gpus else None
     return {
         "ok": True,
@@ -294,10 +733,10 @@ def build_status():
         "score": {
             "gpuFreeVramMb": best_gpu.get("vramFreeMb") if best_gpu else 0,
             "gpuUsagePercent": best_gpu.get("gpuUsagePercent") if best_gpu else None,
-            "ramAvailableMb": ram.get("availableMb") if ram else None,
+            "ramAvailableMb": ram.get("availableMb") if ram else None
         },
         "tags": sorted(set(tags)),
-        "updatedAt": datetime.now(timezone.utc).isoformat(),
+        "updatedAt": datetime.now(timezone.utc).isoformat()
     }
 
 
@@ -346,7 +785,7 @@ chmod +x "$AGENT_DIR/agent.py"
 cat > /etc/systemd/system/bluerouter-ollama-agent.service <<'SERVICE'
 [Unit]
 Description=BlueRouter Ollama Hardware Status Agent
-After=network.target ollama.service
+After=network.target docker.service ollama.service
 
 [Service]
 Type=simple