billion-context 0.1.111-pr.787.631 → 0.1.111

package/dist/index.js

@@ -60869,7 +60869,7 @@ async function decodeRequestBody(contentEncoding, input, maxOutputBytes) {
   return { body, decoded: true };
 }
 
-// src/server/observability.ts
+// src/server.ts
 function bodyDumpEnabled() {
   return process.env.ACP_DUMP_BODY === "1";
 }
@@ -60898,8 +60898,6 @@ function logUnrecognizedPath(log2, url) {
 function isModelDiscoveryPath(path20) {
   return path20.replace(/\/+$/, "").endsWith("/models");
 }
-
-// src/server/context-window.ts
 var BILI_HOP_HEADER = "x-bili-hop";
 function parseLauncherModelWindows(raw) {
   if (!raw) return {};
@@ -60935,8 +60933,6 @@ function anthropicBetaContextWindow(headers) {
   }
   return best;
 }
-
-// src/server/headers.ts
 function safeSessionId(id) {
   return (id ?? "unknown").replace(/[^a-zA-Z0-9._-]/g, "_");
 }
@@ -60978,175 +60974,6 @@ function buildForwardHeaders(headers) {
   out["content-type"] = "application/json";
   return out;
 }
-
-// src/server/side-request.ts
-var SIDE_REQUEST_MAX_TOKENS = 200;
-function isSideRequest(parsed) {
-  if (!parsed || typeof parsed !== "object") return false;
-  const p2 = parsed;
-  if (Array.isArray(p2.tools) && p2.tools.length > 0) return false;
-  const raw = p2.max_tokens ?? p2.max_completion_tokens ?? p2.max_output_tokens;
-  return typeof raw === "number" && raw > 0 && raw <= SIDE_REQUEST_MAX_TOKENS;
-}
-function outputBudgetField(parsed) {
-  if (!parsed || typeof parsed !== "object") return null;
-  const p2 = parsed;
-  for (const field of ["max_tokens", "max_completion_tokens", "max_output_tokens"]) {
-    if (typeof p2[field] === "number" && p2[field] > 0) return field;
-  }
-  return null;
-}
-function restoreOutputBudget(parsed, session, log2) {
-  const field = outputBudgetField(parsed);
-  if (!field) return;
-  const p2 = parsed;
-  const value = p2[field];
-  if (value > SIDE_REQUEST_MAX_TOKENS) {
-    session.metadata.outputBudgetHighWater = value;
-    return;
-  }
-  if (!Array.isArray(p2.tools) || p2.tools.length === 0) return;
-  const highWater = session.metadata.outputBudgetHighWater;
-  if (typeof highWater === "number" && highWater > SIDE_REQUEST_MAX_TOKENS) {
-    p2[field] = highWater;
-    log2("info", `[${session.id}] output budget restored ${value} -> ${highWater} (#546: client shrank it from its raw-history estimate)`);
-  }
-}
-var SIDE_REQUEST_GUARD_TOLERANCE = 1.15;
-function sideRequestGuard(parsed, protocol, modelContextLimit, learnedLimit, imageBilling = "bytes") {
-  let limit = modelContextLimit;
-  if (typeof learnedLimit === "number" && learnedLimit > 0 && learnedLimit < limit) limit = learnedLimit;
-  const field = outputBudgetField(parsed);
-  const maxOut = field ? parsed[field] : 0;
-  if (limit > 0 && shouldReserveOutputHeadroom(protocol)) limit = reserveOutputHeadroom(limit, maxOut);
-  const estimate = estimateRawBodyTokens(parsed) + imageTokensInParsedBody(protocol, parsed, imageBilling);
-  return { blocked: limit > 0 && estimate >= limit * SIDE_REQUEST_GUARD_TOLERANCE, estimate, limit };
-}
-
-// src/server/budget.ts
-var OUTPUT_CLAMP_MARGIN_PCT = 0.05;
-var OUTPUT_CLAMP_MIN_MARGIN = 2048;
-var OUTPUT_CLAMP_FLOOR = 1024;
-var EMERGENCY_NUDGE_ESCALATION_PCT = 0.7;
-function countSystemAndToolsTokens(systemText, tools) {
-  return defaultCountTokens(systemText ?? "") + defaultCountTokens(JSON.stringify(tools ?? []));
-}
-function estimateInputTokens(processedMessages, systemText, tools, lastInputTokens) {
-  const est = estimateCoreMessages(processedMessages) + countSystemAndToolsTokens(systemText, tools);
-  return Math.max(lastInputTokens > 0 ? lastInputTokens : 0, est);
-}
-function estimateWireOverhead(protocol, body) {
-  let parsed;
-  try {
-    parsed = JSON.parse(typeof body === "string" ? body : body.toString("utf8"));
-  } catch {
-    return 0;
-  }
-  const sysRaw = protocol === "responses" ? parsed.instructions : parsed.system;
-  let sysText = "";
-  if (typeof sysRaw === "string") {
-    sysText = sysRaw;
-  } else if (Array.isArray(sysRaw)) {
-    sysText = sysRaw.map((part) => typeof part?.text === "string" ? part.text : "").join("\n");
-  }
-  if (protocol === "openai" && Array.isArray(parsed.messages)) {
-    const hoisted = parsed.messages.filter((m2) => m2.role === "system" || m2.role === "developer").map((m2) => typeof m2.content === "string" ? m2.content : "").join("\n");
-    sysText = sysText ? `${sysText}
-${hoisted}` : hoisted;
-  }
-  return defaultCountTokens(sysText) + defaultCountTokens(JSON.stringify(parsed.tools ?? []));
-}
-function clampOutputBudget(requested, inputEstimate, nativeWindow) {
-  const margin = Math.max(OUTPUT_CLAMP_MIN_MARGIN, Math.ceil(inputEstimate * OUTPUT_CLAMP_MARGIN_PCT));
-  const cap = nativeWindow - inputEstimate - margin;
-  if (cap < OUTPUT_CLAMP_FLOOR || cap >= requested) return void 0;
-  return cap;
-}
-function emergencyNudge(nudge, escalationPct = EMERGENCY_NUDGE_ESCALATION_PCT) {
-  if (!nudge || nudge.shouldInject) return false;
-  if (nudge.compressibleRanges.length === 0) return false;
-  return nudge.contextUsage >= escalationPct;
-}
-function clampOutgoingOutput(rebuilt, field, ctx, sessionId, log2) {
-  const raw = rebuilt[field];
-  if (typeof raw !== "number") return;
-  const inputEstimate = estimateInputTokens(ctx.processedMessages, ctx.systemText, ctx.tools, ctx.lastInputTokens) + ctx.imageTokens;
-  const capped = clampOutputBudget(raw, inputEstimate, ctx.nativeWindow);
-  if (capped !== void 0) {
-    rebuilt[field] = capped;
-    log2("info", `[${sessionId}] output budget clamped ${raw} -> ${capped} (input~${inputEstimate}, window=${ctx.nativeWindow}); prevents input+output overflow (#453)`);
-  }
-}
-
-// src/server/stream-io.ts
-async function readStreamToBuffer(stream2, maxBytes = 1 << 20) {
-  const reader = stream2.getReader();
-  const chunks = [];
-  let kept = 0;
-  try {
-    for (; ; ) {
-      const { done, value } = await reader.read();
-      if (done) break;
-      if (value && kept < maxBytes) {
-        const take = Math.min(value.length, maxBytes - kept);
-        chunks.push(Buffer.from(value.subarray(0, take)));
-        kept += take;
-      }
-    }
-  } finally {
-    reader.releaseLock();
-  }
-  return Buffer.concat(chunks);
-}
-function bufferToStream(buf) {
-  return new ReadableStream({
-    start(controller) {
-      controller.enqueue(new Uint8Array(buf));
-      controller.close();
-    }
-  });
-}
-async function pipeThrough(stream2, res) {
-  const reader = stream2.getReader();
-  try {
-    for (; ; ) {
-      const { done, value } = await reader.read();
-      if (done) break;
-      if (!res.write(Buffer.from(value))) {
-        await new Promise((r) => res.once("drain", () => r()));
-      }
-    }
-  } finally {
-    reader.releaseLock();
-    res.end();
-  }
-}
-async function dumpStreamToFile(stream2, dir, name) {
-  const { mkdirSync: mkdirSync7, createWriteStream: createWriteStream2 } = await import("fs");
-  const { join: join6 } = await import("path");
-  try {
-    mkdirSync7(dir, { recursive: true });
-    const ws2 = createWriteStream2(join6(dir, name));
-    ws2.on("error", (e) => {
-      logDumpFailure("SSE stream dump", e);
-    });
-    const reader = stream2.getReader();
-    try {
-      for (; ; ) {
-        const { done, value } = await reader.read();
-        if (done) break;
-        ws2.write(Buffer.from(value));
-      }
-    } finally {
-      reader.releaseLock();
-      ws2.end();
-    }
-  } catch (err2) {
-    logDumpFailure("SSE stream dump", err2);
-  }
-}
-
-// src/server.ts
 function resolveUpstream(_opts, reqUrl, req) {
   const mitmUpstream = readMitmUpstream(req?.socket);
   if (mitmUpstream) {
@@ -61360,6 +61187,48 @@ Content-Length: ${Buffer.byteLength(body)}\r
   }
   return server;
 }
+var SIDE_REQUEST_MAX_TOKENS = 200;
+function isSideRequest(parsed) {
+  if (!parsed || typeof parsed !== "object") return false;
+  const p2 = parsed;
+  if (Array.isArray(p2.tools) && p2.tools.length > 0) return false;
+  const raw = p2.max_tokens ?? p2.max_completion_tokens ?? p2.max_output_tokens;
+  return typeof raw === "number" && raw > 0 && raw <= SIDE_REQUEST_MAX_TOKENS;
+}
+function outputBudgetField(parsed) {
+  if (!parsed || typeof parsed !== "object") return null;
+  const p2 = parsed;
+  for (const field of ["max_tokens", "max_completion_tokens", "max_output_tokens"]) {
+    if (typeof p2[field] === "number" && p2[field] > 0) return field;
+  }
+  return null;
+}
+function restoreOutputBudget(parsed, session, log2) {
+  const field = outputBudgetField(parsed);
+  if (!field) return;
+  const p2 = parsed;
+  const value = p2[field];
+  if (value > SIDE_REQUEST_MAX_TOKENS) {
+    session.metadata.outputBudgetHighWater = value;
+    return;
+  }
+  if (!Array.isArray(p2.tools) || p2.tools.length === 0) return;
+  const highWater = session.metadata.outputBudgetHighWater;
+  if (typeof highWater === "number" && highWater > SIDE_REQUEST_MAX_TOKENS) {
+    p2[field] = highWater;
+    log2("info", `[${session.id}] output budget restored ${value} -> ${highWater} (#546: client shrank it from its raw-history estimate)`);
+  }
+}
+var SIDE_REQUEST_GUARD_TOLERANCE = 1.15;
+function sideRequestGuard(parsed, protocol, modelContextLimit, learnedLimit, imageBilling = "bytes") {
+  let limit = modelContextLimit;
+  if (typeof learnedLimit === "number" && learnedLimit > 0 && learnedLimit < limit) limit = learnedLimit;
+  const field = outputBudgetField(parsed);
+  const maxOut = field ? parsed[field] : 0;
+  if (limit > 0 && shouldReserveOutputHeadroom(protocol)) limit = reserveOutputHeadroom(limit, maxOut);
+  const estimate = estimateRawBodyTokens(parsed) + imageTokensInParsedBody(protocol, parsed, imageBilling);
+  return { blocked: limit > 0 && estimate >= limit * SIDE_REQUEST_GUARD_TOLERANCE, estimate, limit };
+}
 function imageBillingFor(opts, upstreamUrl) {
   const env = process.env.BILI_IMAGE_BILLING;
   const configured = env === "pixels" || env === "bytes" ? env : findRoute(opts.routes, upstreamUrl)?.imageBilling ?? opts.imageBilling ?? "auto";
@@ -62186,6 +62055,59 @@ function prepareAnthropic(parsed, req, opts, core, config, prompts, surface, log
   delete rebuilt.prompt_cache_key;
   return { body: JSON.stringify(rebuilt), session, processedMessages, originalMessages, anthropicSystem: parsed.system, protocol: "anthropic", stream: stream2, compressInjected: injectTools, pluginMode, nudge, prompts, surface, renderTags: "text-only" };
 }
+var OUTPUT_CLAMP_MARGIN_PCT = 0.05;
+var OUTPUT_CLAMP_MIN_MARGIN = 2048;
+var OUTPUT_CLAMP_FLOOR = 1024;
+var EMERGENCY_NUDGE_ESCALATION_PCT = 0.7;
+function countSystemAndToolsTokens(systemText, tools) {
+  return defaultCountTokens(systemText ?? "") + defaultCountTokens(JSON.stringify(tools ?? []));
+}
+function estimateInputTokens(processedMessages, systemText, tools, lastInputTokens) {
+  const est = estimateCoreMessages(processedMessages) + countSystemAndToolsTokens(systemText, tools);
+  return Math.max(lastInputTokens > 0 ? lastInputTokens : 0, est);
+}
+function estimateWireOverhead(protocol, body) {
+  let parsed;
+  try {
+    parsed = JSON.parse(typeof body === "string" ? body : body.toString("utf8"));
+  } catch {
+    return 0;
+  }
+  const sysRaw = protocol === "responses" ? parsed.instructions : parsed.system;
+  let sysText = "";
+  if (typeof sysRaw === "string") {
+    sysText = sysRaw;
+  } else if (Array.isArray(sysRaw)) {
+    sysText = sysRaw.map((part) => typeof part?.text === "string" ? part.text : "").join("\n");
+  }
+  if (protocol === "openai" && Array.isArray(parsed.messages)) {
+    const hoisted = parsed.messages.filter((m2) => m2.role === "system" || m2.role === "developer").map((m2) => typeof m2.content === "string" ? m2.content : "").join("\n");
+    sysText = sysText ? `${sysText}
+${hoisted}` : hoisted;
+  }
+  return defaultCountTokens(sysText) + defaultCountTokens(JSON.stringify(parsed.tools ?? []));
+}
+function clampOutputBudget(requested, inputEstimate, nativeWindow) {
+  const margin = Math.max(OUTPUT_CLAMP_MIN_MARGIN, Math.ceil(inputEstimate * OUTPUT_CLAMP_MARGIN_PCT));
+  const cap = nativeWindow - inputEstimate - margin;
+  if (cap < OUTPUT_CLAMP_FLOOR || cap >= requested) return void 0;
+  return cap;
+}
+function emergencyNudge(nudge, escalationPct = EMERGENCY_NUDGE_ESCALATION_PCT) {
+  if (!nudge || nudge.shouldInject) return false;
+  if (nudge.compressibleRanges.length === 0) return false;
+  return nudge.contextUsage >= escalationPct;
+}
+function clampOutgoingOutput(rebuilt, field, ctx, sessionId, log2) {
+  const raw = rebuilt[field];
+  if (typeof raw !== "number") return;
+  const inputEstimate = estimateInputTokens(ctx.processedMessages, ctx.systemText, ctx.tools, ctx.lastInputTokens) + ctx.imageTokens;
+  const capped = clampOutputBudget(raw, inputEstimate, ctx.nativeWindow);
+  if (capped !== void 0) {
+    rebuilt[field] = capped;
+    log2("info", `[${sessionId}] output budget clamped ${raw} -> ${capped} (input~${inputEstimate}, window=${ctx.nativeWindow}); prevents input+output overflow (#453)`);
+  }
+}
 function prepareOpenai(parsed, req, opts, core, config, prompts, surface, log2, session, pluginMode, upstreamOrigin, nativeWindow, reasoning, billingUpstream) {
   const sessionId = session.id;
   const stream2 = parsed.stream === true;
@@ -63367,6 +63289,33 @@ ${buildAbsorbSystemPrompt(absorbToolName(loopConfig))}` : "";
     clearUpstreamTimer();
   }
 }
+async function readStreamToBuffer(stream2, maxBytes = 1 << 20) {
+  const reader = stream2.getReader();
+  const chunks = [];
+  let kept = 0;
+  try {
+    for (; ; ) {
+      const { done, value } = await reader.read();
+      if (done) break;
+      if (value && kept < maxBytes) {
+        const take = Math.min(value.length, maxBytes - kept);
+        chunks.push(Buffer.from(value.subarray(0, take)));
+        kept += take;
+      }
+    }
+  } finally {
+    reader.releaseLock();
+  }
+  return Buffer.concat(chunks);
+}
+function bufferToStream(buf) {
+  return new ReadableStream({
+    start(controller) {
+      controller.enqueue(new Uint8Array(buf));
+      controller.close();
+    }
+  });
+}
 async function resolveFakeCompletion(stream2, opts) {
   let buffer = await readStreamToBuffer(stream2, fakeBufCap());
   const max = maxFakeCompletionRetries();
@@ -63411,6 +63360,45 @@ async function resolveFakeCompletion(stream2, opts) {
   markDirty(opts.session);
   return buffer;
 }
+async function pipeThrough(stream2, res) {
+  const reader = stream2.getReader();
+  try {
+    for (; ; ) {
+      const { done, value } = await reader.read();
+      if (done) break;
+      if (!res.write(Buffer.from(value))) {
+        await new Promise((r) => res.once("drain", () => r()));
+      }
+    }
+  } finally {
+    reader.releaseLock();
+    res.end();
+  }
+}
+async function dumpStreamToFile(stream2, dir, name) {
+  const { mkdirSync: mkdirSync7, createWriteStream: createWriteStream2 } = await import("fs");
+  const { join: join6 } = await import("path");
+  try {
+    mkdirSync7(dir, { recursive: true });
+    const ws2 = createWriteStream2(join6(dir, name));
+    ws2.on("error", (e) => {
+      logDumpFailure("SSE stream dump", e);
+    });
+    const reader = stream2.getReader();
+    try {
+      for (; ; ) {
+        const { done, value } = await reader.read();
+        if (done) break;
+        ws2.write(Buffer.from(value));
+      }
+    } finally {
+      reader.releaseLock();
+      ws2.end();
+    }
+  } catch (err2) {
+    logDumpFailure("SSE stream dump", err2);
+  }
+}
 function deriveTitle(messages) {
   for (const m2 of messages) {
     if (m2.role !== "user" || m2.contentType !== "text") continue;