@bivelio/savings-layer 0.4.7 → 0.4.9
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/gateway/bin.js +94 -80
- package/dist/index.d.ts +1 -1
- package/dist/index.js +1 -1
- package/docs/terminales/claude-code.md +7 -1
- package/package.json +2 -1
- package/skills/savings-mode/SKILL.md +27 -0
package/dist/gateway/bin.js
CHANGED
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
#!/usr/bin/env node
|
|
2
|
-
import {spawn}from'child_process';import {existsSync,readFileSync}from'fs';import {homedir}from'os';import {join,delimiter}from'path';import {createServer}from'http';import {randomUUID,createHash}from'crypto';import*as R from'zlib';var
|
|
2
|
+
import {spawn}from'child_process';import {existsSync,mkdirSync,writeFileSync,readFileSync}from'fs';import {homedir}from'os';import {join,dirname,delimiter}from'path';import {fileURLToPath}from'url';import {createServer}from'http';import {randomUUID,createHash}from'crypto';import*as R from'zlib';var we=new Set;function T(e,t){we.has(e)||(we.add(e),console.warn(t));}var A=class{options;now;flushIntervalMs;maxBatch;maxBuffer;buffer=[];seatId;ci;timer;inflight=null;fallosSeguidos=0;constructor(t){this.options=t,this.now=t.now??(()=>Date.now()),this.flushIntervalMs=
|
|
3
3
|
t.flushIntervalMs??15e3,this.maxBatch=t.maxBatch??50,this.maxBuffer=t.maxBuffer??1e3,this.ci=t.ci??false;}setSeat(t,n){t&&(this.seatId=t),this.ci=n;}record(t){this.buffer.push({report:t,at:this.now()}),this.
|
|
4
4
|
buffer.length>this.maxBuffer&&this.buffer.splice(0,this.buffer.length-this.maxBuffer),this.buffer.length>=this.maxBatch&&this.flush();}start(){this.timer||(this.timer=setInterval(()=>{this.flush();},this.
|
|
5
5
|
flushIntervalMs),this.timer.unref?.());}stop(){this.timer&&(clearInterval(this.timer),this.timer=void 0),this.flush();}async stopAndDrain(t){this.timer&&(clearInterval(this.timer),this.timer=void 0);let n=t??
|
|
@@ -17,16 +17,16 @@ length>this.maxBuffer&&(this.buffer.length=this.maxBuffer);}toWire(t){let n=t.re
|
|
|
17
17
|
model:n.model,at:t.at,avoidedLlm:n.calls.avoidedLlm,actualLlm:n.calls.actualLlm,baselineInputTokens:n.tokens.baselineInput,actualInputTokens:n.tokens.actualInput,eliminatedInputTokens:n.tokens.eliminatedInput,
|
|
18
18
|
compressedInputTokens:n.tokens.compressedInput,baselineOutputTokens:n.tokens.baselineOutput,actualOutputTokens:n.tokens.actualOutput,providerCachedInputTokens:n.tokens.providerCachedInput??0,cacheResult:n.
|
|
19
19
|
cache.result,baselineCost:n.cost.baselineEstimate,actualCost:n.cost.actual,netSavings:n.cost.netSavings,...n.cost.basis?{basis:n.cost.basis}:{},...n.attribution?.sessionId?{sessionId:n.attribution.sessionId}:
|
|
20
|
-
{},...n.attribution?.agentId?{agentId:n.attribution.agentId}:{}}}};function
|
|
21
|
-
return new RegExp(`-${t}{8,10}-${t}{5}$`).test(e)}function
|
|
22
|
-
e.NETLIFY)}async function
|
|
23
|
-
if(e.KUBERNETES_SERVICE_HOST||
|
|
24
|
-
e.arch,e.execPath,n,t.USER??t.USERNAME??t.LOGNAME,t.HOME??t.USERPROFILE].filter(r=>typeof r=="string"&&r.length>0);return o.length?o.join("|"):void 0}var
|
|
25
|
-
ONS","GITLAB_CI","CIRCLECI","TRAVIS","JENKINS_URL","BUILDKITE","TF_BUILD","TEAMCITY_VERSION","BITBUCKET_BUILD_NUMBER","CODEBUILD_BUILD_ID","HEROKU_TEST_RUN_ID","DRONE"];function
|
|
26
|
-
return false;for(let t of
|
|
20
|
+
{},...n.attribution?.agentId?{agentId:n.attribution.agentId}:{}}}};function U(){return globalThis.process}function Ze(e){if(/^[0-9a-f]{12}$/.test(e)||/^\d{1,3}(\.\d{1,3}){3}$/.test(e)||/^ip-\d{1,3}(-\d{1,3}){3}$/i.test(e))return true;let t="[bcdfghjklmnpqrstvwxz2456789]";
|
|
21
|
+
return new RegExp(`-${t}{8,10}-${t}{5}$`).test(e)}function et(e){return !!(e.AWS_LAMBDA_FUNCTION_NAME??e.LAMBDA_TASK_ROOT??e.VERCEL??e.VERCEL_ENV??e.K_SERVICE??e.FUNCTION_TARGET??e.FUNCTIONS_WORKER_RUNTIME??
|
|
22
|
+
e.NETLIFY)}async function tt(){if(U()?.execPath)try{let t=await import(["node","os"].join(":")),n=typeof t.hostname=="function"?t.hostname():"";return n.length>0?n:void 0}catch{return}}async function nt(e,t=tt){
|
|
23
|
+
if(e.KUBERNETES_SERVICE_HOST||et(e))return;let n=e.HOSTNAME??e.COMPUTERNAME??await t();if(!(!n||Ze(n)))return n}async function ot(){let e=U();if(!e||!e.env)return;let t=e.env,n=await nt(t),o=[e.platform,
|
|
24
|
+
e.arch,e.execPath,n,t.USER??t.USERNAME??t.LOGNAME,t.HOME??t.USERPROFILE].filter(r=>typeof r=="string"&&r.length>0);return o.length?o.join("|"):void 0}var rt=["CI","CONTINUOUS_INTEGRATION","GITHUB_ACTI\
|
|
25
|
+
ONS","GITLAB_CI","CIRCLECI","TRAVIS","JENKINS_URL","BUILDKITE","TF_BUILD","TEAMCITY_VERSION","BITBUCKET_BUILD_NUMBER","CODEBUILD_BUILD_ID","HEROKU_TEST_RUN_ID","DRONE"];function q(){let e=U()?.env;if(!e)
|
|
26
|
+
return false;for(let t of rt){let n=e[t];if(n&&n!=="false"&&n!=="0")return true}return false}async function st(e,t){let n=globalThis.crypto,o=new TextEncoder;if(n?.subtle)try{let i=await n.subtle.importKey("raw",
|
|
27
27
|
o.encode(e),{name:"HMAC",hash:"SHA-256"},!1,["sign"]),a=await n.subtle.sign("HMAC",i,o.encode(t));return [...new Uint8Array(a)].map(l=>l.toString(16).padStart(2,"0")).join("")}catch{}let r=2166136261,s=`${e}\
|
|
28
|
-
|${t}`;for(let i=0;i<s.length;i++)r^=s.charCodeAt(i),r=Math.imul(r,16777619);return (r>>>0).toString(16).padStart(8,"0")}async function
|
|
29
|
-
if(!o)return;let r=await
|
|
28
|
+
|${t}`;for(let i=0;i<s.length;i++)r^=s.charCodeAt(i),r=Math.imul(r,16777619);return (r>>>0).toString(16).padStart(8,"0")}async function K(e){let n=U()?.env?.BVSALA_SEAT_ID;if(n&&it.test(n))return n;let o=await ot();
|
|
29
|
+
if(!o)return;let r=await st(e,o);if(r)return `m-${r.slice(0,32)}`}var it=/^[A-Za-z0-9._:/-]{1,128}$/;var te={"gpt-4.1":{inputPer1k:.002,outputPer1k:.008,cachedInputPer1k:5e-4},"gpt-4.1-mini":{inputPer1k:4e-4,outputPer1k:.0016,cachedInputPer1k:1e-4},"gpt-4.1-nano":{inputPer1k:1e-4,outputPer1k:4e-4,cachedInputPer1k:25e-6},
|
|
30
30
|
"gpt-4o":{inputPer1k:.0025,outputPer1k:.01,cachedInputPer1k:.00125},"gpt-4o-mini":{inputPer1k:15e-5,outputPer1k:6e-4,cachedInputPer1k:75e-6},"gpt-4-turbo":{inputPer1k:.01,outputPer1k:.03},"gpt-4":{inputPer1k:.03,
|
|
31
31
|
outputPer1k:.06},"gpt-3.5-turbo":{inputPer1k:5e-4,outputPer1k:.0015},"gpt-5":{inputPer1k:.00125,outputPer1k:.01,cachedInputPer1k:125e-6},"gpt-5-codex":{inputPer1k:.00125,outputPer1k:.01,cachedInputPer1k:125e-6},
|
|
32
32
|
"gpt-5-mini":{inputPer1k:25e-5,outputPer1k:.002,cachedInputPer1k:25e-6},"gpt-5-nano":{inputPer1k:5e-5,outputPer1k:4e-4,cachedInputPer1k:5e-6},"gpt-5.1-codex":{inputPer1k:.00125,outputPer1k:.01,cachedInputPer1k:125e-6},
|
|
@@ -42,74 +42,74 @@ able-5":{inputPer1k:.01,outputPer1k:.05,cachedInputPer1k:.001},"claude-sonnet-4"
|
|
|
42
42
|
"claude-3-5-sonnet":{inputPer1k:.003,outputPer1k:.015,cachedInputPer1k:3e-4},"claude-3-5-haiku":{inputPer1k:8e-4,outputPer1k:.004,cachedInputPer1k:8e-5},"claude-3-opus":{inputPer1k:.015,outputPer1k:.075,
|
|
43
43
|
cachedInputPer1k:.0015},"claude-3-haiku":{inputPer1k:25e-5,outputPer1k:.00125,cachedInputPer1k:3e-5},"gemini-3.7-flash":{inputPer1k:75e-5,outputPer1k:.00375,cachedInputPer1k:75e-6},"gemini-3.5-flash":{
|
|
44
44
|
inputPer1k:.0015,outputPer1k:.009,cachedInputPer1k:15e-5},"gemini-3.5-flash-lite":{inputPer1k:3e-4,outputPer1k:.0025,cachedInputPer1k:3e-5},"gemini-3.1-pro-preview":{inputPer1k:.002,outputPer1k:.012,cachedInputPer1k:2e-4},
|
|
45
|
-
"gemini-2.5-pro":{inputPer1k:.00125,outputPer1k:.01,cachedInputPer1k:125e-6},"gemini-2.5-flash":{inputPer1k:3e-4,outputPer1k:.0025,cachedInputPer1k:3e-5}};function
|
|
45
|
+
"gemini-2.5-pro":{inputPer1k:.00125,outputPer1k:.01,cachedInputPer1k:125e-6},"gemini-2.5-flash":{inputPer1k:3e-4,outputPer1k:.0025,cachedInputPer1k:3e-5}};function Te(e,t,n){if(!n)return;let o=e[n];if(o)
|
|
46
46
|
return o;let r=n.lastIndexOf("/"),s=r>=0?n.slice(r+1):n,i=e[s];if(i)return i;for(let a of t){if(!s.startsWith(a))continue;let l=s.charAt(a.length);if(l!==""&&l!=="-"&&l!=="."&&l!=="_")continue;let p=s.
|
|
47
|
-
slice(a.length+1);if(!(/^\d/.test(p)&&!/^\d{4}/.test(p)))return e[a]}}function
|
|
47
|
+
slice(a.length+1);if(!(/^\d/.test(p)&&!/^\d{4}/.test(p)))return e[a]}}function ne(e){return Object.keys(e).sort((t,n)=>n.length-t.length)}var at=ne(te);function $(e){return Te(te,at,e)}var j=class{resto="";uso=null;eventos={};observar(t){this.resto+=typeof t=="string"?t:t.toString("utf8");let n;for(;(n=this.resto.indexOf(`
|
|
48
48
|
|
|
49
49
|
`))>=0;){let o=this.resto.slice(0,n);this.resto=this.resto.slice(n+2);let r=/^event: (.+)$/m.exec(o)?.[1];if(r&&(this.eventos[r]=(this.eventos[r]??0)+1),r!=="message_start"&&r!=="message_delta")continue;
|
|
50
50
|
let s=/^data: (.+)$/m.exec(o)?.[1];if(s)try{let i=JSON.parse(s),a=i.usage??i.message?.usage;if(a){let l=a.cache_creation||this.uso?.cache_creation?{...this.uso?.cache_creation,...a.cache_creation}:void 0;
|
|
51
|
-
this.uso={...this.uso,...a,...l?{cache_creation:l}:{}};}}catch{}}}usage(){return this.uso}usageFinal(){return (this.eventos.message_delta??0)>0?this.uso:null}},
|
|
51
|
+
this.uso={...this.uso,...a,...l?{cache_creation:l}:{}};}}catch{}}}usage(){return this.uso}usageFinal(){return (this.eventos.message_delta??0)>0?this.uso:null}},H=class{resto="";uso=null;eventos={};observar(t){
|
|
52
52
|
this.resto+=typeof t=="string"?t:t.toString("utf8");let n;for(;(n=this.resto.indexOf(`
|
|
53
53
|
|
|
54
54
|
`))>=0;){let o=this.resto.slice(0,n);this.resto=this.resto.slice(n+2);let r=/^event: (.+)$/m.exec(o)?.[1];r&&(this.eventos[r]=(this.eventos[r]??0)+1);let s=/^data: (.+)$/m.exec(o)?.[1];if(s)try{let i=JSON.
|
|
55
|
-
parse(s);if((r??i.type)!=="response.completed")continue;let a=i.response?.usage;a&&(this.uso=
|
|
55
|
+
parse(s);if((r??i.type)!=="response.completed")continue;let a=i.response?.usage;a&&(this.uso=ct(a));}catch{}}}usage(){return this.uso}usageFinal(){return this.uso}};function ct(e){let t=Math.max(0,e.input_tokens??
|
|
56
56
|
0),n=Math.min(Math.max(0,e.input_tokens_details?.cached_tokens??0),t),o=Math.min(Math.max(0,e.input_tokens_details?.cache_write_tokens??0),t-n);return {input_tokens:t-n-o,output_tokens:Math.max(0,e.output_tokens??
|
|
57
|
-
0),cache_read_input_tokens:n,cache_creation_input_tokens:o}}var
|
|
57
|
+
0),cache_read_input_tokens:n,cache_creation_input_tokens:o}}var Ce=1.25,lt=2;function Re(e,t){let n=$(e);if(!n)return null;let o=t.input_tokens??0,r=t.output_tokens??0,s=t.cache_read_input_tokens??0,i=t.
|
|
58
58
|
cache_creation,a=t.cache_creation_input_tokens??(i?.ephemeral_5m_input_tokens??0)+(i?.ephemeral_1h_input_tokens??0),l=Math.min(i?.ephemeral_1h_input_tokens??0,a),p=n.cachedInputPer1k??n.inputPer1k;return {
|
|
59
|
-
costeUsd:o/1e3*n.inputPer1k+s/1e3*p+(a-l)/1e3*n.inputPer1k*
|
|
60
|
-
try{let n=e,o={maxOutputLength:
|
|
61
|
-
break;case "zstd":{let s=R.zstdDecompressSync;if(typeof s!="function")return "";n=s(e,o);break}default:break}let r=JSON.parse(n.toString("utf8")).model;return typeof r=="string"?r:""}catch{return ""}}var
|
|
62
|
-
1024*1024;function
|
|
63
|
-
test(t)?
|
|
64
|
-
return
|
|
59
|
+
costeUsd:o/1e3*n.inputPer1k+s/1e3*p+(a-l)/1e3*n.inputPer1k*dt(e)+l/1e3*n.inputPer1k*lt+r/1e3*n.outputPer1k,precio:n,tokens:{entrada:o,salida:r,cacheLeida:s,cacheEscrita:a,cacheEscrita1h:l}}}function Ee(e,t){
|
|
60
|
+
try{let n=e,o={maxOutputLength:ut};switch((t??"").trim().toLowerCase()){case "gzip":case "x-gzip":n=R.gunzipSync(e,o);break;case "deflate":n=R.inflateSync(e,o);break;case "br":n=R.brotliDecompressSync(e,o);
|
|
61
|
+
break;case "zstd":{let s=R.zstdDecompressSync;if(typeof s!="function")return "";n=s(e,o);break}default:break}let r=JSON.parse(n.toString("utf8")).model;return typeof r=="string"?r:""}catch{return ""}}var ut=5*
|
|
62
|
+
1024*1024;function Ie(){return typeof R.zstdDecompressSync=="function"}function dt(e){let t=e.toLowerCase();return t.startsWith("gpt-")||t.startsWith("o1")||t.startsWith("o3")||t.startsWith("o4")?/^gpt-5\.6/.
|
|
63
|
+
test(t)?Ce:1:t.startsWith("gemini")?1:Ce}function oe(e){try{return JSON.parse(typeof e=="string"?e:e.toString("utf8")).usage??null}catch{return null}}function Pe(e,t,n,o){return Ae(e,t,n,false,o)}function _e(e,t,n,o){
|
|
64
|
+
return Ae(e,t,n,true,o)}function Ae(e,t,n,o,r){let s=n.tokens.entrada+n.tokens.cacheLeida+n.tokens.cacheEscrita;return {requestId:e,model:t,...r&&(r.sessionId||r.agentId)?{attribution:r}:{},calls:o?{avoidedLlm:1,
|
|
65
65
|
actualLlm:0}:{avoidedLlm:0,actualLlm:1},tokens:{baselineInput:s,actualInput:o?0:s,eliminatedInput:0,compressedInput:0,baselineOutput:n.tokens.salida,actualOutput:o?0:n.tokens.salida,providerCachedInput:o?
|
|
66
66
|
0:n.tokens.cacheLeida},cache:{result:o?"singleflight":"miss"},cost:o?{baselineEstimate:n.costeUsd,actual:0,netSavings:1,basis:"estimated"}:{baselineEstimate:n.costeUsd,actual:n.costeUsd,netSavings:0,basis:"\
|
|
67
|
-
measured"}}}var
|
|
67
|
+
measured"}}}var re=5*1024*1024,mt=["authorization","x-api-key","anthropic-version","anthropic-beta"];function Oe(e,t,n,o=""){return createHash("sha256").update(e).update(`
|
|
68
68
|
`).update(t).update(`
|
|
69
69
|
`).update(o).update(`
|
|
70
|
-
`).update(n).digest("hex")}function
|
|
71
|
-
`)}var
|
|
72
|
-
i=false;return {capturar:a=>{if(!i){if(s+=a.length,s>
|
|
70
|
+
`).update(n).digest("hex")}function De(e){return mt.map(t=>`${t}=${e(t)??""}`).join(`
|
|
71
|
+
`)}var J=class{enVuelo=new Map;primarias=0;gemelas=0;registrar(t){if(this.enVuelo.has(t))return null;this.primarias+=1;let n,o=new Promise(a=>n=a);this.enVuelo.set(t,{promesa:o,resolver:n});let r=[],s=0,
|
|
72
|
+
i=false;return {capturar:a=>{if(!i){if(s+=a.length,s>re){i=true,r.length=0;return}r.push(a);}},terminar:(a,l)=>{let p=!i&&a>=200&&a<300;this.enVuelo.get(t)?.resolver(p?{status:a,headers:l,chunks:r}:null),this.
|
|
73
73
|
enVuelo.delete(t);},abortar:()=>{this.enVuelo.get(t)?.resolver(null),this.enVuelo.delete(t);}}}async esperar(t){let n=this.enVuelo.get(t);if(!n)return null;let o=await n.promesa;return o&&(this.gemelas+=
|
|
74
|
-
1),o}tieneEnVuelo(t){return this.enVuelo.has(t)}};var
|
|
75
|
-
c)),r;if(e.serviceAccountKey){r=new
|
|
74
|
+
1),o}tieneEnVuelo(t){return this.enVuelo.has(t)}};var gt=["content-encoding","content-length","transfer-encoding"];async function Le(e={}){let t=(e.upstream??"https://api.anthropic.com").replace(/\/+$/,""),n=e.fetchImpl??fetch,o=e.log??(c=>console.log(
|
|
75
|
+
c)),r;if(e.serviceAccountKey){r=new A({url:e.meteringUrl??"https://savings.bivelio.com/api/metering/usage",serviceAccountKey:e.serviceAccountKey,...e.fetchImpl?{fetchImpl:e.fetchImpl}:{},onAuthError:u=>T(
|
|
76
76
|
"gateway-ingest-auth",`[bvsala gateway] la ingesta de uso rechaza la credencial (${u}). El tr\xE1fico sigue fluyendo (avisar, nunca romper), pero este uso no est\xE1 llegando al panel. Revisa BIVELIO_SERVIC\
|
|
77
|
-
E_KEY y el estado de la suscripci\xF3n en savings.bivelio.com.`)});let c=
|
|
77
|
+
E_KEY y el estado de la suscripci\xF3n en savings.bivelio.com.`)});let c=q();e.seatId?r.setSeat(e.seatId,c):K(e.serviceAccountKey).then(u=>r?.setSeat(u,c)).catch(()=>{}),r.start();}else T("gateway-sin-\
|
|
78
78
|
clave","[bvsala gateway] sin BIVELIO_SERVICE_KEY: el gateway reenv\xEDa el tr\xE1fico pero NO mide. Consigue tu clave en savings.bivelio.com/dashboard/install para ver el gasto en tu panel.");let s=e.
|
|
79
|
-
coalescer?new
|
|
80
|
-
o leer el modelo del cuerpo de ${m} (content-encoding: ${d??"ninguno"}); su uso se reenv\xEDa pero no se mide.`);return}let k=
|
|
81
|
-
"\xB7 COALESCIDA (llamada evitada)":""} \xB7 ${
|
|
82
|
-
""} cache\u2193${k.tokens.cacheLeida}, out ${k.tokens.salida})`)):T(`gateway-sin-precio:${
|
|
83
|
-
l=createServer(async(c,u)=>{let f=randomUUID(),
|
|
84
|
-
isArray(L)?L[0]:L,
|
|
85
|
-
id"]),
|
|
86
|
-
null&&c.method==="POST"&&x.length>0,
|
|
87
|
-
|
|
88
|
-
null);a(f,x,
|
|
89
|
-
let
|
|
79
|
+
coalescer?new J:null,i=c=>c==="/v1/messages"?new j:c.endsWith("/responses")?new H:null,a=(c,u,f,b,m,d,y)=>{if(!r||!f)return;let S=Ee(u,d);if(!S){T(`gateway-sin-modelo:${m}`,`[bvsala gateway] no se pud\
|
|
80
|
+
o leer el modelo del cuerpo de ${m} (content-encoding: ${d??"ninguno"}); su uso se reenv\xEDa pero no se mide.`);return}let k=Re(S,f);k?(r.record(b?_e(c,S,k,y):Pe(c,S,k,y)),o(`[gateway] POST ${m} ${b?
|
|
81
|
+
"\xB7 COALESCIDA (llamada evitada)":""} \xB7 ${S} \xB7 $${k.costeUsd.toFixed(6)} (in ${k.tokens.entrada}, cache\u2191${k.tokens.cacheEscrita}${k.tokens.cacheEscrita1h?` (1h ${k.tokens.cacheEscrita1h})`:
|
|
82
|
+
""} cache\u2193${k.tokens.cacheLeida}, out ${k.tokens.salida})`)):T(`gateway-sin-precio:${S}`,`[bvsala gateway] el modelo "${S}" no est\xE1 en el cat\xE1logo de precios; su uso se reenv\xEDa pero no se mide.`);},
|
|
83
|
+
l=createServer(async(c,u)=>{let f=randomUUID(),b=new AbortController;u.on("close",()=>{u.writableEnded||b.abort();});let m=null,d=new URL(c.url??"/",t),y=null,S=Buffer.alloc(0),k=false,L=c.headers["content-encoding"],V=Array.
|
|
84
|
+
isArray(L)?L[0]:L,be=C=>{let x=Array.isArray(C)?C[0]:C;return x!==void 0&&/^[A-Za-z0-9._:/-]{1,128}$/.test(x)?x:void 0},Y=be(c.headers["x-claude-code-session-id"]),Q=be(c.headers["x-claude-code-agent-\
|
|
85
|
+
id"]),F=Y!==void 0||Q!==void 0?{...Y!==void 0?{sessionId:Y}:{},...Q!==void 0?{agentId:Q}:{}}:void 0;try{let C=[];for await(let h of c)C.push(h);let x=Buffer.concat(C);S=x,d=new URL(c.url??"/",t);let ve=s!==
|
|
86
|
+
null&&c.method==="POST"&&x.length>0,Qe=De(h=>{let v=c.headers[h];return Array.isArray(v)?v.join(", "):v}),X=ve?Oe("POST",d.pathname+d.search,x,Qe):"";if(ve)if(s.tieneEnVuelo(X)){let h=await s.esperar(
|
|
87
|
+
X);if(h){u.writeHead(h.status,h.headers);let v=i(d.pathname)??new j;for(let Se of h.chunks)u.write(Se),v.observar(Se);u.end();let Xe=v.usage()??(d.pathname==="/v1/messages"?oe(Buffer.concat(h.chunks)):
|
|
88
|
+
null);a(f,x,Xe,!0,d.pathname,V,F);return}}else m=s.registrar(X);let ye={};for(let[h,v]of Object.entries(c.headers))h==="host"||h==="content-length"||v===void 0||(ye[h]=Array.isArray(v)?v.join(", "):v);
|
|
89
|
+
let w;try{w=await n(d,{method:c.method,headers:ye,body:["GET","HEAD"].includes(c.method??"")?void 0:x,redirect:"manual",signal:b.signal});}catch(h){m?.abortar(),u.headersSent||(u.writeHead(502,{"conten\
|
|
90
90
|
t-type":"application/json"}),u.end(JSON.stringify({error:"bvsala gateway: upstream unreachable",detail:String(h)}))),o(`[gateway] ${c.method} ${d.pathname} \u2192 upstream inalcanzable: ${String(h)}`);
|
|
91
|
-
return}let
|
|
92
|
-
pathname)!==null;y=
|
|
93
|
-
for await(let h of
|
|
94
|
-
|
|
95
|
-
|
|
96
|
-
|
|
97
|
-
|
|
98
|
-
soportado; el cliente cae a HTTP)`),u.once("error",()=>{}),u.once("finish",()=>u.destroy()),u.end(`HTTP/1.1 426 Upgrade Required\r
|
|
91
|
+
return}let Z=Object.fromEntries(w.headers.entries());for(let h of gt)delete Z[h];u.writeHead(w.status,Z);let ke=w.headers.get("content-type")??"",ee=ke.includes("text/event-stream")||ke===""&&w.ok&&i(
|
|
92
|
+
d.pathname)!==null;y=ee?i(d.pathname):null;let M=!ee&&w.ok&&r!==void 0&&c.method==="POST"&&d.pathname==="/v1/messages"&&(w.headers.get("content-type")??"").includes("application/json")?[]:null,xe=0;if(w.
|
|
93
|
+
body)for await(let h of w.body){let v=Buffer.from(h);u.write(v),y?.observar(v),M&&(xe+=v.length,xe>re?M=null:M.push(v)),m?.capturar(v);}u.end(),m?.terminar(w.status,Z),y&&w.ok?(k=!0,a(f,x,y.usage(),!1,
|
|
94
|
+
d.pathname,V,F)):M?a(f,x,oe(Buffer.concat(M)),!1,d.pathname,V,F):ee||o(`[gateway] ${c.method} ${d.pathname} \u2192 ${w.status}`);}catch(C){if(m?.abortar(),b.signal.aborted){let x=!k&&y?y.usageFinal():null;
|
|
95
|
+
x?(o(`[gateway] ${c.method} ${d.pathname} \u2192 el cliente cerr\xF3 tras el usage final; se mide igual`),a(f,S,x,false,d.pathname,V,F)):o(`[gateway] ${c.method} ${d.pathname} \u2192 el cliente abort\xF3; upstre\
|
|
96
|
+
am cancelado`);}else if(o(`[gateway] ${c.method} ${d.pathname} \u2192 error sirviendo: ${String(C)}`),!u.headersSent){u.writeHead(502,{"content-type":"application/json"}),u.end(JSON.stringify({error:"b\
|
|
97
|
+
vsala gateway: stream interrupted",detail:String(C)}));return}u.writableEnded||u.end();}});return l.on("upgrade",(c,u)=>{o(`[gateway] ${c.method} ${new URL(c.url??"/","http://x").pathname} \u2192 426 (WebSo\
|
|
98
|
+
cket no soportado; el cliente cae a HTTP)`),u.once("error",()=>{}),u.once("finish",()=>u.destroy()),u.end(`HTTP/1.1 426 Upgrade Required\r
|
|
99
99
|
connection: close\r
|
|
100
100
|
content-length: 0\r
|
|
101
101
|
\r
|
|
102
|
-
`);}),await new Promise((c,u)=>{l.once("error",u),l.listen(e.puerto??8402,"127.0.0.1",()=>c());}),{puerto:l.address().port,cerrar:async()=>{await r?.stopAndDrain(),await new Promise(c=>l.close(()=>c()));}}}var
|
|
102
|
+
`);}),await new Promise((c,u)=>{l.once("error",u),l.listen(e.puerto??8402,"127.0.0.1",()=>c());}),{puerto:l.address().port,cerrar:async()=>{await r?.stopAndDrain(),await new Promise(c=>l.close(()=>c()));}}}var se=["claude","codex","goose","qwen"];function $e(e){return se.includes(e??"")}function je(e,t={}){let n=t.path??process.env.PATH??"",o=t.
|
|
103
103
|
existe??existsSync,r=null;for(let s of n.split(";").filter(i=>i.length>0)){for(let i of [".exe",".com"]){let a=join(s,e+i);if(o(a))return {tipo:"exe",ruta:a}}if(r===null)for(let i of [".cmd",".bat"]){let a=join(s,e+
|
|
104
|
-
i);if(o(a)){r=a;break}}}return r!==null?{tipo:"solo-shim",ruta:r}:{tipo:"no-encontrado"}}function
|
|
104
|
+
i);if(o(a)){r=a;break}}}return r!==null?{tipo:"solo-shim",ruta:r}:{tipo:"no-encontrado"}}function Ne(e={}){let n=(e.env??process.env).BVSALA_CODEX_AUTH?.toLowerCase();if(n==="api-key"||n==="apikey")return "\
|
|
105
105
|
api-key";if(n==="chatgpt")return "chatgpt";let o=null;try{o=e.leerAuth?e.leerAuth():null;}catch{o=null;}if(o===null)return "api-key";try{let r=JSON.parse(o);return typeof r.auth_mode=="string"?r.auth_mode.
|
|
106
|
-
toLowerCase().replace(/[^a-z]/g,"")==="apikey"?"api-key":"chatgpt":r.tokens!==void 0&&r.tokens!==null?"chatgpt":"api-key"}catch{return "api-key"}}var
|
|
107
|
-
/v1"},chatgpt:{upstream:"https://chatgpt.com",prefijo:"/backend-api/codex"}};function
|
|
106
|
+
toLowerCase().replace(/[^a-z]/g,"")==="apikey"?"api-key":"chatgpt":r.tokens!==void 0&&r.tokens!==null?"chatgpt":"api-key"}catch{return "api-key"}}var vt={"api-key":{upstream:"https://api.openai.com",prefijo:"\
|
|
107
|
+
/v1"},chatgpt:{upstream:"https://chatgpt.com",prefijo:"/backend-api/codex"}};function ie(e,t,n,o={}){let r=`http://127.0.0.1:${t}`;switch(e){case "claude":return {bin:"claude",args:n,env:{ANTHROPIC_BASE_URL:r}};case "\
|
|
108
108
|
goose":return {bin:"goose",args:n,env:{ANTHROPIC_HOST:r}};case "qwen":return {bin:"qwen",args:n,env:{ANTHROPIC_BASE_URL:r},nota:'Qwen Code solo enruta por aqu\xED con `security.auth.selectedType: "anthropic\
|
|
109
|
-
"` en ~/.qwen/settings.json (ver `bvsala doctor`).'};case "codex":{let s=o.modoCodex??"api-key",i=
|
|
109
|
+
"` en ~/.qwen/settings.json (ver `bvsala doctor`).'};case "codex":{let s=o.modoCodex??"api-key",i=vt[s];return {bin:"codex",args:["-c",`openai_base_url="${r}${i.prefijo}"`,...n],env:{},upstream:i.upstream,
|
|
110
110
|
nota:s==="chatgpt"?"Auth ChatGPT detectada: el gateway enruta a chatgpt.com/backend-api/codex. El tr\xE1fico de suscripci\xF3n se mide (equivalente a precio de lista) pero no se monetiza: no hay factura por\
|
|
111
|
-
tokens.":"Auth por API key: el gateway enruta a api.openai.com/v1."}}}}var
|
|
112
|
-
s `/v1/v1/messages`, sobra un /v1; si ves `POST /messages` a secas, falta uno.",
|
|
111
|
+
tokens.":"Auth por API key: el gateway enruta a api.openai.com/v1."}}}}var St="comprobar: en el log del gateway debe aparecer `[gateway] POST /v1/messages \u2026`; si ve\
|
|
112
|
+
s `/v1/v1/messages`, sobra un /v1; si ves `POST /messages` a secas, falta uno.",wt=[{id:"claude",nombre:"Claude Code",binario:"claude",configs:[],detectable:true,receta:({base:e})=>["usa `bvsala claude`\
|
|
113
113
|
\u2014 arranca el gateway en un puerto ef\xEDmero, lanza `claude` apunt\xE1ndole y lo apaga al salir.",`a mano: ANTHROPIC_BASE_URL=${e} claude (sin /v1: el cliente a\xF1ade /v1/messages)`]},{id:"co\
|
|
114
114
|
dex",nombre:"Codex CLI",binario:"codex",configs:[join(".codex","config.toml"),join(".codex","auth.json")],detectable:true,receta:({base:e,rutas:t})=>["usa `bvsala codex` \u2014 el prefijo del base URL depende del \
|
|
115
115
|
modo de auth y lo resuelve el envoltorio:",` API key \u2192 openai_base_url=${e}/v1`,` suscripci\xF3n ChatGPT \u2192 openai_base_url=${e}/backend-api/codex (tiene que terminar exactament\
|
|
@@ -125,21 +125,26 @@ Droid (Factory)",binario:"droid",configs:[join(".factory","settings.json")],dete
|
|
|
125
125
|
idar en mano contra el log del gateway].`,"{",' "customModels": ['," {",' "model": "<id de modelo Anthropic>",',' "displayName": "<nombre visible>",',' "provider": "anthropic",',` \
|
|
126
126
|
"baseUrl": "${e}",`,' "apiKey": "<tu API key de Anthropic>"'," }"," ]","}"]},{id:"cline",nombre:"Cline",configs:[],detectable:false,receta:({baseV1:e})=>["mecanismo: UI \u2192 proveedor Anthropic\
|
|
127
127
|
\u2192 marca \xABUse custom base URL\xBB. CON /v1 (Cline fija https://api.anthropic.com/v1 y solo normaliza la URL oficial exacta).",`Base URL: ${e}`]},{id:"kilo",nombre:"Kilo Code",configs:[],detectable:false,
|
|
128
|
-
receta:({baseV1:e})=>["mecanismo: UI \u2192 proveedor custom, Provider API \xABAnthropic Messages\xBB. Probar CON /v1 primero (linaje Cline) [inferido: validar en mano].",`Base URL: ${e}`]}];function
|
|
129
|
-
let n=e.split(delimiter).filter(r=>r.length>0),o=process.platform==="win32"?["",".exe",".cmd",".bat"]:[""];return r=>n.some(s=>o.some(i=>t(join(s,r+i))))}async function
|
|
130
|
-
e>65535)throw new RangeError(`[bvsala doctor] puerto inv\xE1lido: ${String(e)}`);let n=t.home??homedir(),o=t.existe??existsSync,r=t.enPath??
|
|
128
|
+
receta:({baseV1:e})=>["mecanismo: UI \u2192 proveedor custom, Provider API \xABAnthropic Messages\xBB. Probar CON /v1 primero (linaje Cline) [inferido: validar en mano].",`Base URL: ${e}`]}];function Tt(e,t){
|
|
129
|
+
let n=e.split(delimiter).filter(r=>r.length>0),o=process.platform==="win32"?["",".exe",".cmd",".bat"]:[""];return r=>n.some(s=>o.some(i=>t(join(s,r+i))))}async function ze(e,t={}){if(!Number.isInteger(e)||e<1||
|
|
130
|
+
e>65535)throw new RangeError(`[bvsala doctor] puerto inv\xE1lido: ${String(e)}`);let n=t.home??homedir(),o=t.existe??existsSync,r=t.enPath??Tt(t.path??process.env.PATH??"",o),s=`http://127.0.0.1:${e}`,i=`${s}/v1`,
|
|
131
131
|
a=[],l=[`bvsala doctor \u2014 gateway local en ${s}`,"Solo imprime: no escribe ninguna configuraci\xF3n. Detecci\xF3n por existencia (binarios en PATH y","archivos de configuraci\xF3n de usuario); nunca \
|
|
132
|
-
se lee el contenido de ning\xFAn archivo.",""];for(let p of
|
|
133
|
-
|
|
134
|
-
push(`\xBB ${p.id}: ${m}`),l.push(` ${p.nombre}`);for(let d of p.receta({base:s,baseV1:i,rutas:c}))l.push(` ${d}`);l.push(` ${
|
|
132
|
+
se lee el contenido de ning\xFAn archivo.",""];for(let p of wt){let c=p.configs.map(d=>join(n,d)),u=p.binario!==void 0&&r(p.binario),f=c.filter(d=>o(d)),b=u||f.length>0;b&&a.push(p.id);let m;p.detectable?
|
|
133
|
+
b?m=`detectado \u2014 ${[...u?[`binario \`${p.binario}\` en PATH`]:[],...f.map(y=>`config presente: ${y}`)].join("; ")}`:m="no detectado":m="sin detecci\xF3n autom\xE1tica (extensi\xF3n de editor)",l.
|
|
134
|
+
push(`\xBB ${p.id}: ${m}`),l.push(` ${p.nombre}`);for(let d of p.receta({base:s,baseV1:i,rutas:c}))l.push(` ${d}`);l.push(` ${St}`),l.push("");}return l.push("Regla BYOK: las recetas del anillo Anth\
|
|
135
135
|
ropic (Goose, Qwen Code, opencode, Crush, Droid, Cline, Kilo) son","para API key. Nunca enrutes por el gateway el OAuth de una suscripci\xF3n Claude Pro/Max desde un harness","de terceros \u2014 Anthropic \
|
|
136
136
|
lo proh\xEDbe. Claude Code y Codex con suscripci\xF3n se enrutan y se miden, pero no","se monetizan: no hay factura por tokens.","Detalle por terminal: docs/terminales/<terminal>.md"),{texto:l.join(`
|
|
137
|
-
`),detectados:a}}var
|
|
137
|
+
`),detectados:a}}var Be="savings-mode";function Ct(e,t){for(let n of e){let o=join(n,"skills",Be,"SKILL.md");if(t.existe(o))return o}return null}function Ve(e){let t=Ct(e.origenes,e.fs);if(!t)
|
|
138
|
+
return {ok:false,destino:"",mensaje:`[bvsala skill] no encuentro el SKILL.md empaquetado (ra\xEDces probadas: ${e.origenes.join(", ")})`};let n=e.proyecto?join(e.cwd,".claude","skills"):join(e.home,".claude","s\
|
|
139
|
+
kills"),o=join(n,Be),r=join(o,"SKILL.md"),s=e.fs.leer(t);if(e.fs.existe(r)&&!e.forzar){let i=e.fs.leer(r)===s;return {ok:i,destino:r,mensaje:i?`[bvsala skill] ya instalado y al d\xEDa: ${r}`:`[bvsala skill] \
|
|
140
|
+
ya existe ${r} con contenido DISTINTO \u2014 no lo piso; usa --forzar para sobrescribir`}}return e.fs.crearDir(o),e.fs.escribir(r,s),{ok:true,destino:r,mensaje:`[bvsala skill] instalado: ${r}
|
|
141
|
+
[bvsala skill] arranca (o reinicia) Claude Code y p\xEDdele \xABsavings mode\xBB \u2014 o deja que se active solo al hablar de l\xEDmites.
|
|
142
|
+
[bvsala skill] para medir su efecto de verdad: trabaja unos d\xEDas con \xE9l y otros sin \xE9l bajo \`bvsala claude\`, y compara en el panel.`}}var ae=class{id;exact=false;constructor(t="heuristic"){this.id=t;}count(t){if(t.length===0)return 0;let n=t.match(/[A-Za-z]+|\d+|\s+|[^\sA-Za-z\d]/g)??[],o=0;for(let r of n){let s=r[0];if(/\s/.test(s)){let i=(r.
|
|
138
143
|
match(/\n/g)??[]).length;o+=i+Math.floor(Math.max(0,r.length-1)/2);}else /[A-Za-z]/.test(s)?o+=Math.max(1,Math.ceil(r.length/4)):/\d/.test(s)?o+=Math.max(1,Math.ceil(r.length/2)):o+=1;}return Math.max(o,
|
|
139
|
-
Math.ceil(t.length/4))}},
|
|
144
|
+
Math.ceil(t.length/4))}},ce=new ae;var le=class{estimators=new Map;modelToTokenizer=new Map;fallbackId;constructor(t=ce){this.fallbackId=t.id,this.estimators.set(t.id,t);}register(t){return this.estimators.set(t.id,t),this}mapModel(t,n){
|
|
140
145
|
return this.modelToTokenizer.set(t,n),this}resolve(t){let n=t.tokenizer?this.estimators.get(t.tokenizer):void 0;if(n)return n;if(t.model){let o=this.modelToTokenizer.get(t.model);if(o){let r=this.estimators.
|
|
141
|
-
get(o);if(r)return r}}return this.estimators.get(this.fallbackId)}count(t,n){return this.resolve(n).count(t)}};new
|
|
142
|
-
|
|
146
|
+
get(o);if(r)return r}}return this.estimators.get(this.fallbackId)}count(t,n){return this.resolve(n).count(t)}};new le;new TextEncoder;new TextDecoder;new TextDecoder;var Vt={"gpt-4o-mini":{deltaTokens:-16,salidaDelBrazoA:72,contrato:"triage (6 fields, long enums)",pares:8},"gpt-4o":{deltaTokens:3,salidaDelBrazoA:75,contrato:"triage (6 fields, long enums)",pares:8}};
|
|
147
|
+
ne(Vt);var me="0.4.9";var[,,g,...P]=process.argv;function Wt(){console.log(`bvsala ${me} \u2014 BiVelio Savings Layer
|
|
143
148
|
|
|
144
149
|
Comandos:
|
|
145
150
|
gateway [--puerto N] Proxy local medidor (por defecto 8402) para terminales que
|
|
@@ -150,6 +155,11 @@ Comandos:
|
|
|
150
155
|
goose [args\u2026] Lanza Goose (ANTHROPIC_HOST) a trav\xE9s del gateway.
|
|
151
156
|
qwen [args\u2026] Lanza Qwen Code (ANTHROPIC_BASE_URL) a trav\xE9s del gateway.
|
|
152
157
|
doctor [--puerto N] Detecta terminales instalados e imprime su configuraci\xF3n exacta.
|
|
158
|
+
skill [--proyecto] [--forzar]
|
|
159
|
+
Instala el skill \xABsavings-mode\xBB de Claude Code (~/.claude/skills;
|
|
160
|
+
con --proyecto, en .claude/skills del directorio actual): salida
|
|
161
|
+
tersa + herramientas quir\xFArgicas para estirar los topes de
|
|
162
|
+
suscripci\xF3n. Su efecto se mide con el gateway, no se promete.
|
|
153
163
|
|
|
154
164
|
Entorno:
|
|
155
165
|
BIVELIO_SERVICE_KEY Clave de servicio (bvk_\u2026): habilita metering + asiento.
|
|
@@ -157,22 +167,26 @@ Entorno:
|
|
|
157
167
|
BVSALA_GATEWAY_UPSTREAM Upstream alternativo (por defecto el del terminal).
|
|
158
168
|
BVSALA_GATEWAY_COALESCE =1 enciende la coalescencia de duplicados (F2, experimental).
|
|
159
169
|
BVSALA_CODEX_AUTH api-key | chatgpt \u2014 fuerza el modo de auth de Codex.
|
|
170
|
+
BVSALA_GATEWAY_LOG =1 ense\xF1a el log por petici\xF3n tambi\xE9n dentro de los envoltorios
|
|
171
|
+
(por defecto va silenciado para no pisar el TUI; mira el panel).
|
|
160
172
|
CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY=1 (Claude Code, solo API key) tus modelos en /model.
|
|
161
173
|
|
|
162
174
|
El gateway nunca corta el tr\xE1fico: sin clave o con la ingesta ca\xEDda, reenv\xEDa y avisa.
|
|
163
|
-
Tus prompts jam\xE1s salen de tu m\xE1quina \u2014 solo cuentas y coste. M\xE1s: savings.bivelio.com`);}var
|
|
164
|
-
|
|
165
|
-
|
|
166
|
-
|
|
167
|
-
|
|
168
|
-
|
|
169
|
-
|
|
170
|
-
|
|
171
|
-
|
|
172
|
-
|
|
173
|
-
{
|
|
174
|
-
|
|
175
|
-
|
|
176
|
-
|
|
177
|
-
|
|
178
|
-
|
|
175
|
+
Tus prompts jam\xE1s salen de tu m\xE1quina \u2014 solo cuentas y coste. M\xE1s: savings.bivelio.com`);}var B=$e(g);g!=="gateway"&&g!=="doctor"&&g!=="skill"&&!B&&(Wt(),process.exit(g===void 0||g==="hel\
|
|
176
|
+
p"||g==="--help"?0:1));if(g==="skill"){let e=dirname(fileURLToPath(import.meta.url)),t=Ve({origenes:[join(e,"..",".."),join(e,"..")],home:homedir(),cwd:process.cwd(),proyecto:P.includes("--proyecto"),forzar:P.includes("--forz\
|
|
177
|
+
ar"),fs:{existe:existsSync,leer:n=>readFileSync(n,"utf8"),escribir:(n,o)=>writeFileSync(n,o),crearDir:n=>mkdirSync(n,{recursive:true})}});console.log(t.mensaje),process.exit(t.ok?0:1);}var he=P.indexOf("--puerto"),D=B?0:he>=0?Number(P[he+1]):
|
|
178
|
+
8402;(!Number.isInteger(D)||(g==="doctor"?D<1:D<0)||D>65535)&&(console.error(`[bvsala ${g}] puerto inv\xE1lido: ${P[he+1]}`),process.exit(1));if(g==="doctor"){P.includes("--write")&&(console.error("[b\
|
|
179
|
+
vsala doctor] --write todav\xEDa no existe: el doctor solo imprime; copia el bloque a mano."),process.exit(1));let e=await ze(D);console.log(e.texto),process.exit(0);}var I=B?ie(g,0,P,{modoCodex:g==="c\
|
|
180
|
+
odex"?Ne({leerAuth:()=>readFileSync(join(homedir(),".codex","auth.json"),"utf8")}):void 0}):null,ge=process.env.BIVELIO_SERVICE_KEY,Gt=B&&process.env.BVSALA_GATEWAY_LOG!=="1",_=await Le({puerto:D,...Gt?{log:()=>{}}:{},
|
|
181
|
+
...ge?{serviceAccountKey:ge}:{},...process.env.BVSALA_SEAT_ID?{seatId:process.env.BVSALA_SEAT_ID}:{},...process.env.BVSALA_GATEWAY_UPSTREAM?{upstream:process.env.BVSALA_GATEWAY_UPSTREAM}:I?.upstream?{
|
|
182
|
+
upstream:I.upstream}:{},...process.env.BVSALA_GATEWAY_COALESCE==="1"?{coalescer:true}:{},...process.env.BVSALA_METERING_URL?{meteringUrl:process.env.BVSALA_METERING_URL}:{}}),Yt=process.env.BVSALA_GATEWAY_UPSTREAM??
|
|
183
|
+
I?.upstream??"https://api.anthropic.com";console.log(`[bvsala gateway] v${me} escuchando en http://127.0.0.1:${_.puerto} \u2192 ${Yt}`);console.log(`[bvsala gateway] medici\xF3n: ${ge?"ACTIVA (clave d\
|
|
184
|
+
e servicio presente)":"APAGADA (falta BIVELIO_SERVICE_KEY)"}`);process.env.BVSALA_GATEWAY_UPSTREAM&&I?.upstream&&process.env.BVSALA_GATEWAY_UPSTREAM!==I.upstream&&console.error(`[bvsala ${g}] AVISO: B\
|
|
185
|
+
VSALA_GATEWAY_UPSTREAM=${process.env.BVSALA_GATEWAY_UPSTREAM} pisa el upstream de este terminal (${I.upstream}). Si no es lo que quieres, quita la variable.`);g==="codex"&&!Ie()&&console.error(`[bvsal\
|
|
186
|
+
a codex] AVISO: este Node (${process.version}) no descomprime zstd y Codex manda el cuerpo en zstd: el tr\xE1fico pasa pero NO se mide. Usa Node \u2265 22.15.`);if(I&&B){let e=ie(g,_.puerto,P,{...I.upstream?
|
|
187
|
+
{modoCodex:I.upstream.includes("chatgpt.com")?"chatgpt":"api-key"}:{}});e.nota&&console.log(`[bvsala ${g}] ${e.nota}`);let t=e.bin;if(process.platform==="win32"){let o=je(e.bin);o.tipo==="exe"?t=o.ruta:
|
|
188
|
+
o.tipo==="solo-shim"&&(console.error(`[bvsala ${g}] \`${e.bin}\` est\xE1 instalado como shim de npm (${o.ruta}) y Windows no puede lanzarlo de forma segura con argumentos (CVE-2024-27980). Inst\xE1lalo c\
|
|
189
|
+
on su instalador nativo (deja un ${e.bin}.exe en el PATH) o usa la receta manual de docs/terminales/.`),await _.cerrar(),process.exit(127));}let n=spawn(t,e.args,{stdio:"inherit",env:{...process.env,...e.
|
|
190
|
+
env}});n.on("error",o=>{console.error(`[bvsala ${g}] no se pudo lanzar \`${e.bin}\`: ${String(o)}. \xBFEst\xE1 instalado y en el PATH?`),_.cerrar().then(()=>process.exit(127));}),n.on("exit",(o,r)=>{_.
|
|
191
|
+
cerrar().then(()=>process.exit(o??(r?1:0)));});}else console.log(`[bvsala gateway] apunta tu terminal (o usa \`bvsala ${se.join("|")}\`): ANTHROPIC_BASE_URL=http://127.0.0.1:${_.puerto} claude \xB7 bv\
|
|
192
|
+
sala doctor para el resto`);var Ye=async()=>{await _.cerrar(),process.exit(0);};process.on("SIGINT",()=>{Ye();});process.on("SIGTERM",()=>{Ye();});
|
package/dist/index.d.ts
CHANGED
|
@@ -3250,6 +3250,6 @@ declare function createSavingsLayer(config: Omit<SavingsLayerConfig, "tokenizer"
|
|
|
3250
3250
|
* constant to package.json so the two can never drift.
|
|
3251
3251
|
*/
|
|
3252
3252
|
|
|
3253
|
-
declare const BV_SALA_VERSION = "0.4.
|
|
3253
|
+
declare const BV_SALA_VERSION = "0.4.9";
|
|
3254
3254
|
|
|
3255
3255
|
export { type AnthropicOptions, AnthropicProvider, BV_SALA_VERSION, type BivelioHandoff, CATALOGO, COMPACT_ROWS_INSTRUCTION, type CacheOutcome, type CacheTier, type CachedAnswer, type CanonicalRequest, type ChatMessage, type CompactRowsDecodeOptions, type CompactRowsOptions, type CompactVerification, type CompressionEconomics, type CompressionPolicyInput, type CompressionRequest, type CompressionResult, type Compressor, type ConstraintBackend, type ConstraintBackendId, type ConstraintPlan, type ConversationState, type CostBasis, type CostModel, DEFAULT_COST_MODEL, DEFAULT_MAX_STALENESS_SECONDS, type DeferredExecutionReport, type DeterministicResolver, EMBEDDED_PUBLIC_KEY, type EmbeddingFn, type EvidenceDoc, type ExactCache, ExactTokenizer, type ExecutionClass, ExecutionRouter, type FormatChoice, FormatSelector, type FormatSelectorOptions, type HandoffFact, HeuristicEstimator, InMemoryExactCache, InMemorySemanticCache, type JsonSchema, LicenseError, type LicenseGate, type LicenseKeyset, type LicenseStatus, type LicenseTokenPayload, type LocalClassifier, MODELOS_POR_CODIFICACION, MODELS_BY_ENCODING, type MemoryDelta, type MessageRole, MockProvider, type MockProviderOptions, type ModelCascadeMode, type ModelCascadeReport, type ModelCascadeSettings, type OpenAICompatibleOptions, OpenAICompatibleProvider, type OutputContract, PRECIOS_AL_DIA, type Policy, type ProtectedSpan, type Provider, type ProviderRequest, type ProviderResponse, type ProviderUsage, QUALITY_STATUSES, type QualityStatus, type RequestContext, type Reranker, type ResolvedDeps, type ResolvedGate, type ResolvedLicense, ResolverRegistry, type ResolverResult, type RetrievalOptions, type RetrievalOutcome, type RetrievalPlan, type Retriever, type RevocationListBody, type RevocationListVerification, type RevocationPollEvent, RevocationPoller, type RevocationPollerOptions, RevocationSnapshot, type RouteDecision, type SavingsLayer, type SavingsLayerConfig, SavingsLedger, type SavingsReport, type SavingsRequest, type SavingsResult, type SavingsToggles, SchemaRegistry, type SemanticCacheOptions, type SemanticHit, type SemanticRecord, type SemanticVerifier, type SerializationFormat, type Serializer, Singleflight, type TokenEstimator, TokenizerRegistry, type ToolCallRequest, type ToolDescriptor, type ToolRisk, type ToolSelection, ToolSelector, type ToolSelectorOptions, TraceCollector, type TraceEvent, UMBRAL_LOCAL, UsageEmitter, type UsageEmitterOptions, type ValidationError, type ValidationResult, adaptiveK, arithmeticResolver, buildExactKey, buildInflightKey, canonicalize, compactRowsSerializer, compresionFiel, computeNextDelay, cosine, createCompactRowsSerializer, createDefaultResolvers, createHandoff, createMockProvider, createSavingsLayer, csvSerializer, decodeCompactRows, dedupe, defaultTokenizerRegistry, deriveSeatId, deriveStatus, detectCi, embebedorLocal, emptyState, encodeCompactRows, evalArithmetic, evaluatePolicy, executeRetrievalPlan, gateFromKey, heuristicEstimator, isKeyset, isStrictCompatible, jsonCompactSerializer, keyTokensMatch, liveGateFromKey, mergeDelta, modelFamilyOf, multisetEqual, negationCount, noopCompressor, normalizeText, numericTokens, precioDeLista, projectContext, providerNativeBackend, randomId, registerExactTokenizer, registrarTokenizadorExacto, renderState, resolveConstraintBackend, resolveDependencies, resolveLicense, sha256, shouldCompress, stableStringify, uniformColumns, validateAgainstSchema, vectorLocal, verifyCompactSignature, verifyLicenseToken, verifyRevocationList, withProtectedSpans, words };
|
package/dist/index.js
CHANGED
|
@@ -782,7 +782,7 @@ ng/usage`:void 0);if(a){i=new dt({url:a,serviceAccountKey:t.serviceAccountKey,..
|
|
|
782
782
|
{},...t.metering?.maxBatch!=null?{maxBatch:t.metering.maxBatch}:{},...t.metering?.maxBuffer!=null?{maxBuffer:t.metering.maxBuffer}:{},...t.meteringFetch?{fetchImpl:t.meteringFetch}:t.licenseFetch?{fetchImpl:t.
|
|
783
783
|
licenseFetch}:{},...t.now?{now:t.now}:{},...t.metering?.ci!=null?{ci:t.metering.ci}:{},...t.silenceWarnings!=null?{silenceWarnings:t.silenceWarnings}:{}});let c=t.metering?.ci??cr();if(t.metering?.seatId)
|
|
784
784
|
i.setSeat(t.metering.seatId,c);else {let l=i;lr(t.serviceAccountKey).then(u=>l.setSeat(u,c)).catch(()=>{});}i.start();}}return {generate:async a=>{let c=await s.generate(a);if(i)try{i.record(c.report);}catch{}
|
|
785
|
-
return c},close:()=>{r?.stop(),o?.stop(),i?.stop();}}}var af="0.4.
|
|
785
|
+
return c},close:()=>{r?.stop(),o?.stop(),i?.stop();}}}var af="0.4.9";/*! Bundled license information:
|
|
786
786
|
|
|
787
787
|
@noble/curves/utils.js:
|
|
788
788
|
@noble/curves/abstract/modular.js:
|
|
@@ -42,7 +42,13 @@ Claude Code añade `/v1/messages` por su cuenta. Sin barra final.
|
|
|
42
42
|
|
|
43
43
|
## Comprobar que funciona
|
|
44
44
|
|
|
45
|
-
|
|
45
|
+
**Con `bvsala claude` (el envoltorio), la comprobación es tu dashboard**: el
|
|
46
|
+
log por petición va silenciado por defecto desde la 0.4.8 — imprimirlo pisaba
|
|
47
|
+
el TUI de Claude Code (las palabras se montaban sobre los diálogos). Si lo
|
|
48
|
+
quieres ver igualmente, `BVSALA_GATEWAY_LOG=1 bvsala claude`.
|
|
49
|
+
|
|
50
|
+
Con `bvsala gateway` a secas (dos terminales), el log sí es la comprobación,
|
|
51
|
+
en este orden:
|
|
46
52
|
|
|
47
53
|
```
|
|
48
54
|
[gateway] HEAD /api/hello → 200
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@bivelio/savings-layer",
|
|
3
|
-
"version": "0.4.
|
|
3
|
+
"version": "0.4.9",
|
|
4
4
|
"description": "BV-SALA — provider-neutral LLM execution optimizer. A decision layer in front of any LLM provider that cuts the cost of every request while preserving required quality, and reports the saving per call with token-level accounting you can check against your bill. Proprietary and confidential — property of BiVelio Inc.",
|
|
5
5
|
"keywords": [
|
|
6
6
|
"llm",
|
|
@@ -42,6 +42,7 @@
|
|
|
42
42
|
"files": [
|
|
43
43
|
"dist",
|
|
44
44
|
"docs/terminales",
|
|
45
|
+
"skills",
|
|
45
46
|
"README.md",
|
|
46
47
|
"LICENSE"
|
|
47
48
|
],
|
|
@@ -0,0 +1,27 @@
|
|
|
1
|
+
---
|
|
2
|
+
name: savings-mode
|
|
3
|
+
description: Stretch Claude subscription usage limits (5-hour and weekly caps) by cutting token burn — terse output, surgical tool calls, lean context — without sacrificing correctness. Use when the user mentions usage limits, running out of quota, or asks to save tokens.
|
|
4
|
+
---
|
|
5
|
+
|
|
6
|
+
# Savings Mode — by BiVelio
|
|
7
|
+
|
|
8
|
+
Measured on real Claude Code traffic (provider meters, not estimates): model output is ~12% of the money-weighted budget despite being under 1% of tokens, and tool results are the single biggest context burner — each one is cache-written at a premium and re-read on every later turn. This mode cuts exactly those two masses and touches nothing else.
|
|
9
|
+
|
|
10
|
+
## Output
|
|
11
|
+
- Lead with the answer. No preamble, no recap of what the user just saw.
|
|
12
|
+
- Never restate file contents, diffs, or command output — reference `path:line` instead.
|
|
13
|
+
- One line for routine confirmations; no closing summaries after routine edits.
|
|
14
|
+
- Prose only where it adds information — no decorative headers or tables for simple answers.
|
|
15
|
+
|
|
16
|
+
## Tool calls (the big one)
|
|
17
|
+
- Read the slice you need (offset/limit), not whole files; never re-read a file that has not changed.
|
|
18
|
+
- Constrain searches: filter by glob, ask for counts or file lists first, cap result volume. One targeted search after thinking beats several speculative ones.
|
|
19
|
+
- Filter command output at the source (grep, tail, -q) so only the signal enters the context; never let a verbose log dump through.
|
|
20
|
+
- Batch independent tool calls in a single block.
|
|
21
|
+
|
|
22
|
+
## Context
|
|
23
|
+
- Write large artifacts to files instead of pasting them into the conversation.
|
|
24
|
+
- At a natural boundary where the earlier bulk no longer matters, suggest the user run /compact — never assume it for them.
|
|
25
|
+
|
|
26
|
+
## Never trade away
|
|
27
|
+
Correctness, tests, and safety checks are never skipped to save tokens. Code you write stays clear — terseness applies to conversation, not to code. When terse conflicts with clear, clear wins.
|