@tuned-tensor/cli 0.26.2 → 0.28.0

package/dist/index.js

@@ -49,6 +49,7 @@ import {
   readHardwareSnapshot,
   readJsonlTail,
   readSpecHistory,
+  renameAtomicFile,
   resolveProjectConfig,
   runGpuProcess,
   runLocalPipeline,
@@ -64,7 +65,7 @@ import {
   warningsFromSnapshot,
   withAppendLock,
   writePrivateJsonAtomic
-} from "./chunk-VQWJ2XKR.js";
+} from "./chunk-7CO5DBCY.js";
 import {
   DECISION_MODELS,
   DEFAULT_ARTIFACT_ROOT,
@@ -96,10 +97,10 @@ import {
   specLockPath,
   withDecisionPythonEnvironment,
   withFoundationPythonEnvironment
-} from "./chunk-2AIP63QO.js";
+} from "./chunk-FSVTMX4L.js";
 
 // src/cli.ts
-import { spawn as spawn6 } from "child_process";
+import { spawn as spawn7 } from "child_process";
 import { constants as osConstants2 } from "os";
 import { Command } from "commander";
 import chalk15 from "chalk";
@@ -661,7 +662,38 @@ import { randomUUID as randomUUID3 } from "crypto";
 import { closeSync, openSync } from "fs";
 import { mkdir, open, readdir, readFile as readFile2, rm as rm3, writeFile as writeFile3 } from "fs/promises";
 import { hostname, userInfo } from "os";
-import { basename as basename2, dirname as dirname3, join as join3 } from "path";
+import { basename as basename2, dirname as dirname3, join as join4 } from "path";
+
+// src/local-runtime/windows-child-signals.ts
+import { spawn } from "child_process";
+import { join as join3 } from "path";
+function windowsChildSignals(child) {
+  let timer;
+  const dispose = () => {
+    if (timer) clearTimeout(timer);
+  };
+  child.once("close", dispose);
+  return {
+    dispose,
+    forward(signal) {
+      if (process.platform !== "win32" || !child.connected) return false;
+      child.send({ ttSignal: signal }, () => {
+      });
+      timer ??= setTimeout(() => {
+        if (child.pid) {
+          spawn(
+            join3(process.env.SystemRoot ?? "C:\\Windows", "System32", "taskkill.exe"),
+            ["/PID", String(child.pid), "/T", "/F"],
+            { windowsHide: true, stdio: "ignore" }
+          ).on("error", () => child.kill());
+        }
+      }, 5e3);
+      return true;
+    }
+  };
+}
+
+// src/local-runtime/background-runs.ts
 var ACTIVE = /* @__PURE__ */ new Set(["starting", "running", "cancel_requested"]);
 var DEFAULT_MAX_PARALLEL_RUNS = 1;
 var MAX_PARALLEL_RUNS_LIMIT = 8;
@@ -675,7 +707,7 @@ var BackgroundRunError = class extends Error {
   }
 };
 function backgroundRunsRoot(env = process.env) {
-  return join3(getTunedTensorHome(env), "background-runs");
+  return join4(getTunedTensorHome(env), "background-runs");
 }
 function maxParallelRuns(env = process.env) {
   const raw = env.TT_MAX_PARALLEL_RUNS?.trim();
@@ -736,17 +768,17 @@ function plainLogText(text3) {
 }
 function createBackgroundRunStore(root = backgroundRunsRoot()) {
   const paths = (id) => {
-    const dir = join3(root, id);
+    const dir = join4(root, id);
     return {
       dir,
-      record: join3(dir, "record.json"),
-      spec: join3(dir, "spec.json"),
-      events: join3(dir, "events.jsonl"),
-      log: join3(dir, "output.log"),
-      result: join3(dir, "result.json")
+      record: join4(dir, "record.json"),
+      spec: join4(dir, "spec.json"),
+      events: join4(dir, "events.jsonl"),
+      log: join4(dir, "output.log"),
+      result: join4(dir, "result.json")
     };
   };
-  const lockPath = (id) => join3(paths(id).dir, ".lock");
+  const lockPath = (id) => join4(paths(id).dir, ".lock");
   async function readRecord(id) {
     return JSON.parse(await readFile2(paths(id).record, "utf8"));
   }
@@ -845,7 +877,7 @@ function createBackgroundRunStore(root = backgroundRunsRoot()) {
     async create(input, options = {}) {
       const limit = options.maxParallel ?? DEFAULT_MAX_PARALLEL_RUNS;
       await mkdir(root, { recursive: true, mode: 448 });
-      return await withAppendLock(join3(root, ".launch.lock"), async () => {
+      return await withAppendLock(join4(root, ".launch.lock"), async () => {
         const active = (await list()).filter(isActiveBackgroundRun);
         if (active.length >= limit) {
           throw new BackgroundRunError(
@@ -896,7 +928,7 @@ function createBackgroundRunStore(root = backgroundRunsRoot()) {
         details: { reason }
       });
       const pid = record6.process?.supervisor_pid;
-      if (wasRunning && pid !== void 0 && record6.process?.host === hostname()) {
+      if (process.platform !== "win32" && wasRunning && pid !== void 0 && record6.process?.host === hostname()) {
         try {
           process.kill(pid, "SIGTERM");
         } catch {
@@ -1050,9 +1082,9 @@ async function launchBackgroundSupervisor(args2) {
       stdio: "ignore"
     });
     if (child.pid === void 0) {
-      await new Promise((resolve13, reject) => {
+      await new Promise((resolve14, reject) => {
         child.once("error", reject);
-        child.once("spawn", () => resolve13());
+        child.once("spawn", () => resolve14());
       });
     }
   } catch (error) {
@@ -1069,8 +1101,8 @@ async function launchBackgroundSupervisor(args2) {
   });
 }
 function workspaceSnapshotPath(run) {
-  const specFile = join3(run.choices.workspace_root, run.choices.spec_path);
-  return join3(dirname3(specFile), `.${basename2(specFile)}.tt-agent-${run.id}.json`);
+  const specFile = join4(run.choices.workspace_root, run.choices.spec_path);
+  return join4(dirname3(specFile), `.${basename2(specFile)}.tt-agent-${run.id}.json`);
 }
 async function superviseBackgroundRun(options) {
   const { store } = options;
@@ -1095,37 +1127,43 @@ async function superviseBackgroundRun(options) {
     return await finishEarly("failed", `Could not prepare the run: ${error.message}`);
   }
   let child;
+  let windowsSignals;
+  let cancellationPoll;
   const state = { exitCode: null, exitSignal: null };
   const stop = (signal) => {
     state.stopSignal ??= signal;
     if (!child?.pid) return;
     try {
       if (process.platform !== "win32") process.kill(-child.pid, "SIGTERM");
-      else child.kill("SIGTERM");
+      else if (!windowsSignals?.forward("SIGTERM")) child.kill("SIGTERM");
     } catch {
     }
   };
-  const onSignal = (signal) => stop(signal);
-  process.on("SIGTERM", onSignal);
-  process.on("SIGINT", onSignal);
-  if (process.platform !== "win32") process.on("SIGHUP", onSignal);
+  const onSigterm = () => stop("SIGTERM");
+  const onSigint = () => stop("SIGINT");
+  const onSighup = () => stop("SIGHUP");
+  process.on("SIGTERM", onSigterm);
+  process.on("SIGINT", onSigint);
+  if (process.platform !== "win32") process.on("SIGHUP", onSighup);
   try {
     const spawnImpl = options.spawn ?? nodeSpawn;
     child = spawnImpl(options.execPath ?? process.execPath, [options.entrypoint, "--json", "pipeline", "run", "--spec", snapshot], {
       cwd: run.choices.workspace_root,
-      env: options.env ?? process.env,
+      env: process.platform === "win32" ? { ...options.env ?? process.env, TT_PARENT_IPC: "1" } : options.env ?? process.env,
       detached: process.platform !== "win32",
-      stdio: ["ignore", resultFd, logFd]
+      stdio: ["ignore", resultFd, logFd, ...process.platform === "win32" ? ["ipc"] : []],
+      windowsHide: true
     });
-    const closed = new Promise((resolve13) => {
+    windowsSignals = windowsChildSignals(child);
+    const closed = new Promise((resolve14) => {
       child.once("error", (error) => {
         state.spawnError = error;
-        resolve13();
+        resolve14();
       });
       child.once("close", (code2, signal) => {
         state.exitCode = code2;
         state.exitSignal = signal;
-        resolve13();
+        resolve14();
       });
     });
     if (child.pid !== void 0) {
@@ -1135,12 +1173,27 @@ async function superviseBackgroundRun(options) {
         current.process = { ...current.process, host: hostname(), supervisor_pid: process.pid, pid: child.pid };
       }, { type: "started", message: `Pipeline process ${child.pid} started.` });
       if ((await store.get(run.id)).status === "cancel_requested") stop("SIGTERM");
+      if (process.platform === "win32") {
+        let polling = false;
+        cancellationPoll = setInterval(() => {
+          if (polling || state.stopSignal) return;
+          polling = true;
+          void store.get(run.id).then((current) => {
+            if (current.status === "cancel_requested") stop("SIGTERM");
+          }).catch(() => {
+          }).finally(() => {
+            polling = false;
+          });
+        }, 250);
+      }
     }
     await closed;
   } finally {
-    process.off("SIGTERM", onSignal);
-    process.off("SIGINT", onSignal);
-    if (process.platform !== "win32") process.off("SIGHUP", onSignal);
+    if (cancellationPoll) clearInterval(cancellationPoll);
+    windowsSignals?.dispose();
+    process.off("SIGTERM", onSigterm);
+    process.off("SIGINT", onSigint);
+    if (process.platform !== "win32") process.off("SIGHUP", onSighup);
     closeSync(logFd);
     closeSync(resultFd);
     await rm3(snapshot, { force: true });
@@ -1325,7 +1378,7 @@ function createBackgroundRunExecutor(args2) {
 // src/agent-labeling.ts
 import { createHash as createHash4 } from "crypto";
 import { existsSync as existsSync3, lstatSync as lstatSync2, readFileSync as readFileSync4 } from "fs";
-import { isAbsolute as isAbsolute2, join as join5, relative as relative4 } from "path";
+import { isAbsolute as isAbsolute2, join as join6, relative as relative4 } from "path";
 import { Type } from "typebox";
 
 // src/commands/label.ts
@@ -1345,7 +1398,7 @@ import {
   readFileSync,
   writeFileSync
 } from "fs";
-import { join as join4 } from "path";
+import { join as join5 } from "path";
 var AGENT_THINKING_LEVELS = [
   "off",
   "minimal",
@@ -1371,7 +1424,7 @@ function validateAccessToken(value) {
 }
 var CONFIG_FILE_NAME = "config.json";
 function getConfigPath() {
-  return join4(getConfigDir(), CONFIG_FILE_NAME);
+  return join5(getConfigDir(), CONFIG_FILE_NAME);
 }
 function readConfig() {
   const path = getConfigPath();
@@ -1705,7 +1758,7 @@ function formatCents(cents) {
   return `$${(cents / 100).toFixed(2)}`;
 }
 function sleep(ms) {
-  return new Promise((resolve13) => setTimeout(resolve13, ms));
+  return new Promise((resolve14) => setTimeout(resolve14, ms));
 }
 function isTransientError(error) {
   if (error instanceof ApiError) return error.status === 429 || error.status >= 500;
@@ -2215,10 +2268,10 @@ function registerLabelCommands(parent) {
       console.log(`Follow progress with \`tt cloud label watch ${shortId(job.id)}\`.`);
       return;
     }
-    const finished = await watchJob(job.id, opts);
-    if (isJsonMode()) return printJson(finished);
+    const finished2 = await watchJob(job.id, opts);
+    if (isJsonMode()) return printJson(finished2);
     printSuccess(
-      `Labeling ${finished.status === "awaiting_review" ? "complete" : finished.status}: ${finished.labeled_count}/${finished.row_count} rows labeled.`
+      `Labeling ${finished2.status === "awaiting_review" ? "complete" : finished2.status}: ${finished2.labeled_count}/${finished2.row_count} rows labeled.`
     );
     console.log(
       `Review with \`tt cloud label rows ${shortId(job.id)} --status labeled\`, then \`tt cloud label download ${shortId(job.id)}\` to write local training files.`
@@ -2453,7 +2506,7 @@ async function workspaceFile(workspaceRoot, file) {
   }
   let current = root;
   for (const part of parts) {
-    current = join5(current, part);
+    current = join6(current, part);
     let info;
     try {
       info = lstatSync2(current);
@@ -2629,7 +2682,7 @@ async function revalidateLabelDownload(workspaceRoot, args2) {
   if (isAbsolute2(String(args2.out_dir)) || parts.some((part) => !part || part === "." || part === "..")) {
     throw new Error("Download directory must be inside the current workspace.");
   }
-  const outDir = join5(spec.root, ...parts);
+  const outDir = join6(spec.root, ...parts);
   assertLabelDownloadPaths(outDir, args2.overwrite === true);
   return {
     workspaceRoot,
@@ -3063,7 +3116,7 @@ async function rejectPreparedAction(action, persist) {
 // src/agent-model.ts
 import { ModelRuntime } from "@earendil-works/pi-coding-agent";
 import { chmodSync as chmodSync2, existsSync as existsSync4, mkdirSync as mkdirSync3, readFileSync as readFileSync5, writeFileSync as writeFileSync3 } from "fs";
-import { dirname as dirname5, join as join6 } from "path";
+import { dirname as dirname5, join as join7 } from "path";
 function missingProviderAuthMessage() {
   return "No API key is saved for the selected model. Run /login, or /model to pick a different one.";
 }
@@ -3162,10 +3215,10 @@ function resolveAgentModelDefinition(runtime, selection) {
   return { model, thinking: selection.thinking };
 }
 function getAgentAuthPath() {
-  return join6(getAgentConfigDir(), "auth.json");
+  return join7(getAgentConfigDir(), "auth.json");
 }
 function getAgentModelsPath() {
-  return join6(getAgentConfigDir(), "models.json");
+  return join7(getAgentConfigDir(), "models.json");
 }
 async function createPiModelRuntime(env = process.env) {
   const agentDir = getAgentConfigDir();
@@ -3577,7 +3630,7 @@ async function fetchHuggingFaceSearch(kind, query, limit, filters, options) {
   let lastError = new Error("Hugging Face search failed.");
   for (let attempt = 0; attempt < 2; attempt += 1) {
     if (attempt > 0) {
-      await new Promise((resolve13) => setTimeout(resolve13, options.huggingFaceRetryDelayMs ?? 750));
+      await new Promise((resolve14) => setTimeout(resolve14, options.huggingFaceRetryDelayMs ?? 750));
     }
     try {
       const payload = await fetchHuggingFaceJson(url, options);
@@ -3958,7 +4011,7 @@ function createTunedTensorTools(api, options = {}) {
             dry_run: "tt pipeline run --dry-run --spec tunedtensor.json",
             run: "tt pipeline run --spec tunedtensor.json"
           },
-          note: "Runs locally on CPU, Apple MPS or CUDA; runtime.gpu (AWS) is not used. The tuned model directory loads in Python with laya.load(<model_dir>).",
+          note: "Runs locally on CPU, Apple MPS or CUDA; runtime.gpu (AWS) is not used. The tuned model directory loads in Python with laya.load(<model_dir>). For paid serverless CPU hosting, the user can run tt cloud endpoints create --model <model_dir> --name <name>; see docs/local-runtime/hosting.md. This uploads weights to the TT account, returns an authenticated inference URL and scoped key, and charges prepaid credits per successful request.",
           ...host ? { host } : {}
         };
       }
@@ -4142,11 +4195,21 @@ function createTunedTensorTools(api, options = {}) {
       }, { additionalProperties: false }),
       async (p) => {
         const current = await inspectLocalSpec(workspaceRoot, p.spec_path);
+        const spec = record4(current.document);
+        const inline = current.validation.valid && spec && (spec.engine ?? "adapter") === "adapter" && !spec.dataset_prebuilt && Array.isArray(spec.examples) ? spec.examples : void 0;
+        const split = inline ? splitSpecExamples(inline, 0) : void 0;
         return {
           spec_path: current.displayPath,
           sha256: current.sha256,
           spec: current.document ?? current.source,
-          validation: current.validation
+          validation: current.validation,
+          ...split ? { inline_data_split: {
+            method: "deterministic_holdout",
+            training_examples: split.train.length,
+            heldout_examples: split.holdout.length,
+            real_training_minimum_examples: 2,
+            note: "Inline adapter data is split automatically. A small held-out set is not evidence of model quality."
+          } } : {}
         };
       }
     ),
@@ -4485,6 +4548,8 @@ function createTunedTensorTools(api, options = {}) {
 // src/local-agent-client.ts
 var MAX_TOOL_CALLS_PER_TURN = 12;
 var SYSTEM_PROMPT = `You are the local Tuned Tensor assistant running on the user's laptop.
+Use get_local_spec's inline_data_split counts when describing adapter training and evaluation data. An automatically held-out set is distinct from training, but a small set is not evidence of model quality.
+Downloads may also write TT manifests, reports and logs. Summarize the proposed effects without claiming the cache and lock are the only writes.
 The workspace tunedtensor.json behavior spec is the source of truth for model behavior, training, evaluation policy, runtime placement and optional pipeline settings. For requests to change the model's behavior, inspect it with get_local_spec first, then prepare_update_local_spec with the returned SHA-256 and only the requested field changes; use add_examples to append examples. Preserve unrelated examples, instructions, engine and training settings. Explain the proposed diff and validation; never claim an edit was saved before approval. Read-only questions about the current spec use get_local_spec without proposing an edit. If no spec exists, help prepare a new one with prepare_create_local_spec; it is created as tunedtensor.json in the current workspace. For an initial draft, default to three concise representative examples unless the user requests otherwise; more examples can be added later. Malformed JSON must be repaired in an editor before prepare_update_local_spec can operate; never offer that tool as a way to reconstruct a truncated file. For direct review, point users to /spec (a readable overview), /spec <section> such as /spec examples, /spec show for the JSON, and /spec diff, /spec validate and /spec history. General educational or inspection questions do not require a spec mutation.
 Local workflow commands need no TT access token. For local inspection commands such as runs, models, or doctor, tell the user to run the matching TT command in this shell (for example \`runs list\` or \`doctor\`); those commands execute outside the agent. User-owned AWS GPU training uses \`tt pipeline run\` with runtime.gpu configuration in tunedtensor.json, without a TT token. \`tt cloud\` is for TT account records, not GPU execution; hosted training start/estimate are retired. AWS credentials authorize instance lookup; separate SSH access is required. The instance must already be running, and the user manages quota, capacity, costs and shutdown. Keep the laptop running until outputs return. Model-provider choice is independent of GPU placement.
 When the user requests commands to inspect a run or serve a result, provide the direct CLI handoff: \`tt runs list\` to find a run, \`tt runs get <run-id>\` for its details, \`tt runs report <run-id>\` for its evaluation report, \`tt models list\` to find the resulting model ID, and \`tt serve <model-id>\` to serve that model. Replace placeholders with actual IDs from those commands; never invent results or IDs. Until the user supplies command output, keep literal <run-id> and <model-id> placeholders; do not fabricate example IDs or prefixes. Local run IDs are UUIDs and adapter model IDs are local-<run-id>. Do not refresh hardware or propose spec edits merely to supply those commands; perform those actions only when requested.
@@ -4602,11 +4667,11 @@ function createLocalAgentClient(options) {
     const runs = options.backgroundRuns?.store;
     if (!runs) return void 0;
     try {
-      const finished = (await runs.list()).filter((run) => run.requested_by.thread_id === threadId && !isActiveBackgroundRun(run) && !run.reported_at);
-      if (finished.length === 0) return void 0;
-      await runs.markReported(finished.map((run) => run.id));
+      const finished2 = (await runs.list()).filter((run) => run.requested_by.thread_id === threadId && !isActiveBackgroundRun(run) && !run.reported_at);
+      if (finished2.length === 0) return void 0;
+      await runs.markReported(finished2.map((run) => run.id));
       return `[Background run outcomes, recorded by tt]
-${finished.map((run) => JSON.stringify(describeBackgroundRun(run))).join("\n")}
+${finished2.map((run) => JSON.stringify(describeBackgroundRun(run))).join("\n")}
 These runs finished since your last turn. Mention them briefly when relevant; use get_background_run for logs.`;
     } catch {
       return void 0;
@@ -4851,13 +4916,12 @@ import {
   mkdir as mkdir2,
   readFile as readFile3,
   readdir as readdir2,
-  rename,
   stat,
   unlink,
   writeFile as writeFile4
 } from "fs/promises";
 import { randomUUID as randomUUID6 } from "crypto";
-import { join as join7 } from "path";
+import { join as join8 } from "path";
 var UUID = /^[0-9a-f]{8}-[0-9a-f]{4}-[1-8][0-9a-f]{3}-[89ab][0-9a-f]{3}-[0-9a-f]{12}$/i;
 var DEFAULT_MAX_THREAD_BYTES = 1e6;
 var DEFAULT_MAX_MESSAGES = 200;
@@ -4888,8 +4952,8 @@ function scrub(value, secrets) {
 var LocalAgentStore = class {
   constructor(rootDir, options = {}) {
     this.rootDir = rootDir;
-    this.threadsDir = join7(rootDir, "threads");
-    this.claimsDir = join7(rootDir, "action-claims");
+    this.threadsDir = join8(rootDir, "threads");
+    this.claimsDir = join8(rootDir, "action-claims");
     this.secrets = options.secretValues?.filter(Boolean) ?? [];
     this.secretValueProvider = options.secretValueProvider;
     this.maxThreadBytes = options.maxThreadBytes ?? DEFAULT_MAX_THREAD_BYTES;
@@ -4927,7 +4991,7 @@ var LocalAgentStore = class {
   }
   path(id) {
     assertThreadId(id);
-    return join7(this.threadsDir, `${id}.json`);
+    return join8(this.threadsDir, `${id}.json`);
   }
   serializeWithinQuota(state) {
     const active = state.actions.filter(
@@ -4966,11 +5030,11 @@ var LocalAgentStore = class {
     return serialized;
   }
   async pruneDirectory(directory, suffix, maximum, preserve) {
-    const entries = await Promise.all((await readdir2(directory)).filter((name) => suffix.test(name) && name !== preserve).map(async (name) => ({ name, modified: (await stat(join7(directory, name))).mtimeMs })));
+    const entries = await Promise.all((await readdir2(directory)).filter((name) => suffix.test(name) && name !== preserve).map(async (name) => ({ name, modified: (await stat(join8(directory, name))).mtimeMs })));
     entries.sort((a, b) => a.modified - b.modified);
     const keepOther = Math.max(0, maximum - (preserve ? 1 : 0));
     for (const entry of entries.slice(0, Math.max(0, entries.length - keepOther))) {
-      await unlink(join7(directory, entry.name));
+      await unlink(join8(directory, entry.name));
     }
   }
   async pruneThreads(preserve) {
@@ -4990,19 +5054,19 @@ var LocalAgentStore = class {
     }
     const candidates = await Promise.all(names.filter((name) => !protectedNames.has(name)).map(async (name) => ({
       name,
-      modified: (await stat(join7(this.threadsDir, name))).mtimeMs
+      modified: (await stat(join8(this.threadsDir, name))).mtimeMs
     })));
     candidates.sort((a, b) => a.modified - b.modified);
     const keep = Math.max(0, this.maxThreads - protectedNames.size);
     for (const entry of candidates.slice(0, Math.max(0, candidates.length - keep))) {
-      await unlink(join7(this.threadsDir, entry.name));
+      await unlink(join8(this.threadsDir, entry.name));
     }
   }
   async save(state) {
     assertThreadId(state.thread.id);
     await this.ensureDirectories();
     const destination = this.path(state.thread.id);
-    const temporary = join7(
+    const temporary = join8(
       this.threadsDir,
       `.${state.thread.id}.${randomUUID6()}.tmp`
     );
@@ -5015,14 +5079,13 @@ var LocalAgentStore = class {
     }
     await writeFile4(temporary, serialized, { mode: 384, flag: "wx" });
     await chmod(temporary, 384);
-    await rename(temporary, destination);
-    await chmod(destination, 384);
+    await renameAtomicFile(temporary, destination);
     await this.pruneThreads(`${state.thread.id}.json`);
   }
   async claimAction(actionId) {
     assertThreadId(actionId);
     await this.ensureDirectories();
-    const path = join7(this.claimsDir, `${actionId}.claim`);
+    const path = join8(this.claimsDir, `${actionId}.claim`);
     try {
       await writeFile4(path, `${(/* @__PURE__ */ new Date()).toISOString()}
 `, {
@@ -6331,7 +6394,7 @@ function createCommandCompleter(getMode = () => "local") {
 
 // src/shell-context.ts
 import { readFile as readFile4, readdir as readdir3, stat as stat2 } from "fs/promises";
-import { basename as basename5, dirname as dirname6, join as join8, resolve as resolve3 } from "path";
+import { basename as basename5, dirname as dirname6, join as join9, resolve as resolve3 } from "path";
 async function readJsonObject(path) {
   try {
     const parsed = JSON.parse(await readFile4(path, "utf8"));
@@ -6358,7 +6421,7 @@ function expandPath(value, baseDirectory, env) {
   return expandUserPath(value, env, baseDirectory);
 }
 function configPath(env) {
-  return join8(getConfigDir(env), "config.json");
+  return join9(getConfigDir(env), "config.json");
 }
 function agentSelectionFrom(env, configAgent, hasToken) {
   const stored = configAgent && typeof configAgent === "object" && !Array.isArray(configAgent) ? configAgent : {};
@@ -6376,14 +6439,14 @@ async function adjacentFile(path) {
   }
 }
 async function readActiveModelId(storeRoot) {
-  const result2 = await readJsonObject(join8(storeRoot, "active-model.json"));
+  const result2 = await readJsonObject(join9(storeRoot, "active-model.json"));
   if (!result2.found || result2.invalid) return void 0;
   const modelId = result2.value?.model_id;
   if (modelId === null) return "base";
   return typeof modelId === "string" && modelId ? modelId : void 0;
 }
 async function readLatestRun(storeRoot) {
-  const runsRoot = join8(storeRoot, "runs");
+  const runsRoot = join9(storeRoot, "runs");
   let names;
   try {
     names = (await readdir3(runsRoot, { withFileTypes: true })).filter((entry) => entry.isDirectory()).map((entry) => entry.name);
@@ -6391,7 +6454,7 @@ async function readLatestRun(storeRoot) {
     return void 0;
   }
   const states = await Promise.all(names.map(async (name) => {
-    const result2 = await readJsonObject(join8(runsRoot, name, "state.json"));
+    const result2 = await readJsonObject(join9(runsRoot, name, "state.json"));
     if (!result2.found || result2.invalid || !result2.value) return void 0;
     const id = stringField(result2.value, "id") ?? name;
     const state = {
@@ -6428,7 +6491,7 @@ async function discoverShellContext(options = {}) {
   const env = options.env ?? process.env;
   const cwd = resolve3(options.cwd ?? process.cwd());
   const warnings = [];
-  const specPath = join8(cwd, "tunedtensor.json");
+  const specPath = join9(cwd, "tunedtensor.json");
   const specJson = await readJsonObject(specPath);
   let spec;
   if (specJson.found) {
@@ -6453,7 +6516,7 @@ async function discoverShellContext(options = {}) {
       };
     }
   }
-  const localConfigPath = join8(cwd, "local-runner.json");
+  const localConfigPath = join9(cwd, "local-runner.json");
   const hasAdjacentLocalConfig = await adjacentFile(localConfigPath);
   const localConfigJson = hasAdjacentLocalConfig ? await readJsonObject(localConfigPath) : { found: false, invalid: false };
   if (localConfigJson.invalid) {
@@ -6787,12 +6850,12 @@ async function promptKeyChoice(message, input = stdin, output = stdout) {
       terminal.setRawMode(true);
       input.resume();
       output.write(`${message} ${hint()}`);
-      const choice = await new Promise((resolve13) => {
+      const choice = await new Promise((resolve14) => {
         const finish = (decision) => {
           input.removeListener("keypress", onKeypress);
           input.removeListener("end", onEnd);
           input.removeListener("close", onEnd);
-          resolve13(decision);
+          resolve14(decision);
         };
         const onKeypress = (text3, key) => {
           const decided = keyChoiceFor(key, text3);
@@ -6834,7 +6897,7 @@ function operatorError(operator) {
     `Shell operator ${JSON.stringify(operator)} is not supported. Run TT commands directly; pipes, redirects, and command substitution are disabled.`
   );
 }
-function tokenizeShellInput(input) {
+function tokenizeShellInput(input, platform = process.platform) {
   if (input.includes("\0")) {
     throw new ShellParseError("Command input must not contain NUL bytes.");
   }
@@ -6869,7 +6932,7 @@ function tokenizeShellInput(input) {
         quote = "none";
         continue;
       }
-      if (character === "\\") {
+      if (character === "\\" && platform !== "win32") {
         const escaped = input[index + 1];
         if (escaped === void 0) {
           throw new ShellParseError("Command ends with an unfinished escape.");
@@ -6901,7 +6964,7 @@ function tokenizeShellInput(input) {
       tokenStarted = true;
       continue;
     }
-    if (character === "\\") {
+    if (character === "\\" && platform !== "win32") {
       const escaped = input[index + 1];
       if (escaped === void 0) {
         throw new ShellParseError("Command ends with an unfinished escape.");
@@ -8446,7 +8509,7 @@ var TunedTensorAgentSession = class {
 // src/local-runner.ts
 import { constants as osConstants } from "os";
 import { randomUUID as randomUUID7 } from "crypto";
-import { spawn } from "child_process";
+import { spawn as spawn2 } from "child_process";
 import { existsSync as existsSync5 } from "fs";
 import { fileURLToPath as fileURLToPath2 } from "url";
 import {
@@ -8459,7 +8522,7 @@ import {
   basename as basename6,
   dirname as dirname7,
   isAbsolute as isAbsolute4,
-  join as join9,
+  join as join10,
   resolve as resolve5
 } from "path";
 
@@ -9248,17 +9311,17 @@ function positionalSpecArgument(args2, start, defaultPath) {
 function selectedSpecArgument(args2, cwd) {
   const command2 = args2[0];
   if (["doctor", "validate", "run"].includes(command2 ?? "")) {
-    return positionalSpecArgument(args2, 1, join9(cwd, DEFAULT_SPEC_NAME));
+    return positionalSpecArgument(args2, 1, join10(cwd, DEFAULT_SPEC_NAME));
   }
   if (command2 === "models" && ["prefetch", "verify-base"].includes(args2[1] ?? "")) {
-    return positionalSpecArgument(args2, 2, join9(cwd, DEFAULT_SPEC_NAME));
+    return positionalSpecArgument(args2, 2, join10(cwd, DEFAULT_SPEC_NAME));
   }
   if (command2 === "datasets") {
     if (args2[1] === void 0 || args2[1].startsWith("-") || args2[1].endsWith(".json")) {
       args2.splice(1, 0, "prefetch");
     }
     if (["prefetch", "verify"].includes(args2[1] ?? "")) {
-      return positionalSpecArgument(args2, 2, join9(cwd, DEFAULT_SPEC_NAME));
+      return positionalSpecArgument(args2, 2, join10(cwd, DEFAULT_SPEC_NAME));
     }
   }
   const serves = command2 === "serve" || command2 === "models" && args2[1] === "serve";
@@ -9346,7 +9409,7 @@ async function projectSpecArguments(originalArgs, cwd) {
       }
     };
   }
-  const temporaryPath = join9(
+  const temporaryPath = join10(
     dirname7(sourcePath),
     `.${basename6(sourcePath)}.tt-local-${process.pid}-${randomUUID7()}.json`
   );
@@ -9416,18 +9479,20 @@ async function spawnLocalCli(args2) {
   const childEnv = errorTerminal.isTTY && !args2.jsonMode && args2.env.TT_PROGRESS_TTY === void 0 ? { ...args2.env, TT_PROGRESS_TTY: "1", TT_PROGRESS_COLUMNS: String(errorTerminal.columns ?? 100) } : args2.env;
   const pipeInput = Boolean(args2.stdin && args2.stdin !== process.stdin);
   const pipeStreamingOutput = args2.streamingStdout && Boolean(args2.stdout && args2.stdout !== process.stdout);
-  const child = spawn(
+  const child = spawn2(
     args2.nodeExecutable,
     [args2.entrypoint, ...args2.commandArgs],
     {
       cwd: args2.cwd,
-      env: childEnv,
+      env: process.platform === "win32" ? { ...childEnv, TT_PARENT_IPC: "1" } : childEnv,
       stdio: [
         pipeInput ? "pipe" : "inherit",
         args2.streamingStdout && !pipeStreamingOutput ? "inherit" : "pipe",
-        "pipe"
+        "pipe",
+        ...process.platform === "win32" ? ["ipc"] : []
       ],
-      detached: process.platform !== "win32"
+      detached: process.platform !== "win32",
+      windowsHide: true
     }
   );
   if (pipeInput && args2.stdin && child.stdin) args2.stdin.pipe(child.stdin);
@@ -9456,8 +9521,10 @@ async function spawnLocalCli(args2) {
   });
   return new Promise((resolveChild) => {
     let spawnError;
+    const windowsSignals = windowsChildSignals(child);
     const forwardSignal = (signal) => {
       const childSignal = signal === "SIGHUP" ? "SIGTERM" : signal;
+      if (windowsSignals.forward(childSignal)) return;
       if (child.pid && process.platform !== "win32") {
         try {
           process.kill(-child.pid, childSignal);
@@ -9470,14 +9537,16 @@ async function spawnLocalCli(args2) {
     const onSigint = () => forwardSignal("SIGINT");
     const onSigterm = () => forwardSignal("SIGTERM");
     const onSighup = () => forwardSignal("SIGHUP");
+    const signals = args2.signals ?? process;
     const cleanup = () => {
-      process.off("SIGINT", onSigint);
-      process.off("SIGTERM", onSigterm);
-      if (process.platform !== "win32") process.off("SIGHUP", onSighup);
+      windowsSignals.dispose();
+      signals.off("SIGINT", onSigint);
+      signals.off("SIGTERM", onSigterm);
+      if (process.platform !== "win32") signals.off("SIGHUP", onSighup);
     };
-    process.on("SIGINT", onSigint);
-    process.on("SIGTERM", onSigterm);
-    if (process.platform !== "win32") process.on("SIGHUP", onSighup);
+    signals.on("SIGINT", onSigint);
+    signals.on("SIGTERM", onSigterm);
+    if (process.platform !== "win32") signals.on("SIGHUP", onSighup);
     child.once("error", (error) => {
       spawnError = error;
     });
@@ -9528,6 +9597,7 @@ async function runLocalCommand(args2, options = {}) {
       stdin: options.stdin,
       stdout: options.stdout,
       stderr: options.stderr,
+      signals: options.signals,
       pathReplacements: projected.pathReplacements
     });
     const parsed = parseJsonOutput(result2.stdout);
@@ -9715,11 +9785,11 @@ function registerAgentCommands(parent, options) {
 
 // src/commands/pipeline.ts
 import { existsSync as existsSync6, readFileSync as readFileSync7, writeFileSync as writeFileSync4 } from "fs";
-import { basename as basename8, dirname as dirname11, join as join13, resolve as resolve8 } from "path";
+import { basename as basename8, dirname as dirname11, join as join14, resolve as resolve8 } from "path";
 
 // src/local-runtime/project-migration.ts
-import { lstat as lstat3, readFile as readFile6, rename as rename2, writeFile as writeFile6, unlink as unlink2 } from "fs/promises";
-import { dirname as dirname8, join as join10 } from "path";
+import { lstat as lstat3, readFile as readFile6, rename, writeFile as writeFile6, unlink as unlink2 } from "fs/promises";
+import { dirname as dirname8, join as join11 } from "path";
 async function migrateProjectWorkflow(specPath) {
   const info = await lstat3(specPath);
   if (!info.isFile() || info.isSymbolicLink()) throw new Error("Migration requires a regular spec file.");
@@ -9730,7 +9800,7 @@ async function migrateProjectWorkflow(specPath) {
   const files = [];
   const sources = /* @__PURE__ */ new Map([[specPath, original]]);
   for (const name of ["tunedtensor.pipeline.json", "local-runner.json"]) {
-    const path = join10(dirname8(specPath), name);
+    const path = join11(dirname8(specPath), name);
     const entry = await lstat3(path).catch((error) => {
       if (error.code === "ENOENT") return void 0;
       throw error;
@@ -9779,7 +9849,7 @@ async function migrateProjectWorkflow(specPath) {
         throw new Error(`Configuration changed during migration: ${path}; retry after reviewing it.`);
       }
     }
-    await rename2(temp, specPath);
+    await rename(temp, specPath);
   } catch (error) {
     if (createdTemp) await unlink2(temp).catch(() => {
     });
@@ -9797,8 +9867,8 @@ async function migrateProjectWorkflow(specPath) {
 // src/local-runtime/foundation-runner.ts
 import { createHash as createHash7, randomUUID as randomUUID8 } from "crypto";
 import { createReadStream } from "fs";
-import { chmod as chmod2, lstat as lstat4, mkdir as mkdir3, open as open2, readdir as readdir4, readFile as readFile7, rename as rename3, rm as rm5 } from "fs/promises";
-import { basename as basename7, dirname as dirname9, join as join11, resolve as resolve6 } from "path";
+import { chmod as chmod2, lstat as lstat4, mkdir as mkdir3, open as open2, readdir as readdir4, readFile as readFile7, rename as rename2, rm as rm5 } from "fs/promises";
+import { basename as basename7, dirname as dirname9, join as join12, resolve as resolve6 } from "path";
 var INSTRUCTION_CORPUS_VERSION = 1;
 function isRef(value) {
   return Boolean(value) && typeof value === "object" && !Array.isArray(value) && typeof value.from === "string";
@@ -9827,7 +9897,7 @@ async function hashPath(path) {
   if (info.isDirectory()) {
     for (const name of (await readdir4(path)).sort()) {
       hash.update(name);
-      hash.update(await hashPath(join11(path, name)));
+      hash.update(await hashPath(join12(path, name)));
     }
     return hash.digest("hex");
   }
@@ -9850,7 +9920,7 @@ async function makePrivateTree(path) {
   if (info.isDirectory()) {
     await chmod2(path, 448);
     for (const name of await readdir4(path)) {
-      await makePrivateTree(join11(path, name));
+      await makePrivateTree(join12(path, name));
     }
     return;
   }
@@ -9879,7 +9949,7 @@ async function readRequiredJsonObject(path, description) {
   return parsed;
 }
 async function writePrivateTextAtomic(path, content) {
-  const temporary = join11(dirname9(path), `.${basename7(path)}.${randomUUID8()}.tmp`);
+  const temporary = join12(dirname9(path), `.${basename7(path)}.${randomUUID8()}.tmp`);
   const file = await open2(temporary, "wx", 384);
   let closed = false;
   try {
@@ -9888,7 +9958,7 @@ async function writePrivateTextAtomic(path, content) {
     await file.sync();
     await file.close();
     closed = true;
-    await rename3(temporary, path);
+    await rename2(temporary, path);
   } finally {
     if (!closed) await file.close().catch(() => void 0);
     await rm5(temporary, { force: true }).catch(() => void 0);
@@ -9995,7 +10065,7 @@ async function runFoundationPipeline(args2) {
   assertFoundationPlanSupported(args2.spec, args2.plan);
   assertFoundationSettingsMatch(args2.spec.foundation, args2.plan);
   const outputDir = resolve6(
-    args2.outputDir ?? (args2.spec.runtime?.artifactRoot ? join11(resolve6(dirname9(args2.specPath), args2.spec.runtime.artifactRoot), "foundation-runs", randomUUID8()) : join11(dirname9(resolve6(args2.specPath)), DEFAULT_FOUNDATION_RUNS_DIR, randomUUID8()))
+    args2.outputDir ?? (args2.spec.runtime?.artifactRoot ? join12(resolve6(dirname9(args2.specPath), args2.spec.runtime.artifactRoot), "foundation-runs", randomUUID8()) : join12(dirname9(resolve6(args2.specPath)), DEFAULT_FOUNDATION_RUNS_DIR, randomUUID8()))
   );
   if (args2.gpu && args2.spec.foundation.checkpoint_backup_dir && localPathsOverlap(outputDir, args2.spec.foundation.checkpoint_backup_dir)) {
     throw new Error("AWS checkpoint_backup_dir must be separate from the foundation run directory; overlapping backups can overwrite returned checkpoints.");
@@ -10020,7 +10090,7 @@ async function runFoundationPipeline(args2) {
     await mkdir3(outputDir, { recursive: true, mode: 448 });
   }
   await makePrivateTree(outputDir);
-  const workflowPath = join11(outputDir, "resolved-workflow.json");
+  const workflowPath = join12(outputDir, "resolved-workflow.json");
   const workflow = { spec: args2.spec, plan: args2.plan };
   if (outputExists) {
     const previous = await readFile7(workflowPath, "utf8").catch((error) => {
@@ -10053,16 +10123,16 @@ async function runFoundationPipeline(args2) {
     throw new Error(`Missing ${kind} artifact.`);
   };
   for (const step of args2.plan.steps) {
-    const stepDir = join11(outputDir, step.id);
-    const output = join11(stepDir, "output");
+    const stepDir = join12(outputDir, step.id);
+    const output = join12(stepDir, "output");
     await mkdir3(stepDir, { recursive: true, mode: 448 });
     await chmod2(stepDir, 448);
     await mkdir3(output, { recursive: true, mode: 448 });
     await chmod2(output, 448);
-    const configPath2 = join11(stepDir, "config.json");
-    const logPath = join11(stepDir, "step.log");
-    const metricsPath = join11(output, "metrics.json");
-    const completionPath = join11(stepDir, "completion.json");
+    const configPath2 = join12(stepDir, "config.json");
+    const logPath = join12(stepDir, "step.log");
+    const metricsPath = join12(output, "metrics.json");
+    const completionPath = join12(stepDir, "completion.json");
     const fields = stepConfig(step);
     let entrypoint;
     let config;
@@ -10083,7 +10153,7 @@ async function runFoundationPipeline(args2) {
       config = {
         tokenizer_dir: tokenizerDir,
         output_dir: output,
-        work_dir: join11(stepDir, "recovery"),
+        work_dir: join12(stepDir, "recovery"),
         depth: fields.depth ?? hp.depth,
         steps: fields.steps ?? hp.pretrain_steps,
         batch_size: fields.batchSize ?? hp.batch_size,
@@ -10188,15 +10258,15 @@ async function runFoundationPipeline(args2) {
     const artifacts = {};
     const record6 = {};
     if (step.uses === "tokenize") {
-      const tokenizerPath = join11(output, "tokenizer.json");
+      const tokenizerPath = join12(output, "tokenizer.json");
       await requireRegularNonemptyFile(tokenizerPath, `Foundation step "${step.id}" tokenizer.json`);
       await readRequiredJsonObject(tokenizerPath, `Foundation step "${step.id}" tokenizer.json`);
       record6.tokenizer = output;
       tokenizerDir = output;
       artifacts.tokenizer = await artifactFor(output);
     } else if (step.uses === "pretrain" || step.uses === "finetune" || step.uses === "rl") {
-      const modelPath = join11(output, "model.safetensors");
-      const modelConfigPath = join11(output, "config.json");
+      const modelPath = join12(output, "model.safetensors");
+      const modelConfigPath = join12(output, "config.json");
       await requireRegularNonemptyFile(modelPath, `Foundation step "${step.id}" model.safetensors`);
       const modelConfig = await readRequiredJsonObject(
         modelConfigPath,
@@ -10211,7 +10281,7 @@ async function runFoundationPipeline(args2) {
       modelDir = output;
       artifacts.model = await artifactFor(output);
     } else if (step.uses === "evaluate") {
-      const reportPath2 = join11(output, "report.json");
+      const reportPath2 = join12(output, "report.json");
       const stepReport = await readRequiredJsonObject(
         reportPath2,
         `Foundation step "${step.id}" report.json`
@@ -10251,7 +10321,7 @@ async function runFoundationPipeline(args2) {
     spec: args2.specPath,
     steps
   };
-  const reportPath = join11(outputDir, "report.json");
+  const reportPath = join12(outputDir, "report.json");
   await writePrivateTextAtomic(reportPath, `${JSON.stringify(report, null, 2)}
 `);
   return {
@@ -10264,8 +10334,8 @@ async function runFoundationPipeline(args2) {
 
 // src/local-runtime/decision-runner.ts
 import { randomUUID as randomUUID9 } from "crypto";
-import { lstat as lstat5, mkdir as mkdir4, readFile as readFile8, rename as rename4, rm as rm6, writeFile as writeFile7 } from "fs/promises";
-import { dirname as dirname10, join as join12, resolve as resolve7 } from "path";
+import { lstat as lstat5, mkdir as mkdir4, readFile as readFile8, rename as rename3, rm as rm6, writeFile as writeFile7 } from "fs/promises";
+import { dirname as dirname10, join as join13, resolve as resolve7 } from "path";
 var PROBABILITY_FLOOR = 1e-6;
 function assertDecisionPlanSupported(plan) {
   for (const step of plan.steps) {
@@ -10428,7 +10498,7 @@ async function writePrivateAtomic(path, content) {
   const temporary = `${path}.${process.pid}.${randomUUID9()}.tmp`;
   try {
     await writeFile7(temporary, content, { encoding: "utf8", mode: 384, flag: "wx" });
-    await rename4(temporary, path);
+    await rename3(temporary, path);
   } finally {
     await rm6(temporary, { force: true });
   }
@@ -10447,7 +10517,7 @@ async function runDecisionPipeline(args2) {
   assertDecisionPlanSupported(args2.plan);
   const model = resolveDecisionModel(args2.spec.base_model);
   const specDir = dirname10(resolve7(args2.specPath));
-  const outputDir = resolve7(args2.outputDir ?? (args2.spec.runtime?.artifactRoot ? join12(resolve7(specDir, args2.spec.runtime.artifactRoot), "decision-runs", randomUUID9()) : join12(specDir, DEFAULT_DECISION_RUNS_DIR, randomUUID9())));
+  const outputDir = resolve7(args2.outputDir ?? (args2.spec.runtime?.artifactRoot ? join13(resolve7(specDir, args2.spec.runtime.artifactRoot), "decision-runs", randomUUID9()) : join13(specDir, DEFAULT_DECISION_RUNS_DIR, randomUUID9())));
   if (await lstat5(outputDir).catch(() => null)) {
     throw new Error(`Decision output directory already exists: ${outputDir}`);
   }
@@ -10466,13 +10536,13 @@ async function runDecisionPipeline(args2) {
   const hp = args2.spec.hyperparameters ?? {};
   const device = hp.device ?? "auto";
   const evalIds = split.holdout.map((_, index) => `eval-${index}`);
-  const dataDir = join12(outputDir, "data");
+  const dataDir = join13(outputDir, "data");
   await mkdir4(dataDir, { recursive: true, mode: 448 });
-  const trainPath = join12(dataDir, "train.jsonl");
-  const evalInputsPath = join12(dataDir, "eval-inputs.jsonl");
+  const trainPath = join13(dataDir, "train.jsonl");
+  const evalInputsPath = join13(dataDir, "eval-inputs.jsonl");
   await writePrivateAtomic(trainPath, jsonl(split.train));
   await writePrivateAtomic(evalInputsPath, jsonl(split.holdout.map((example, index) => ({ id: evalIds[index], input: example.input }))));
-  await writePrivateJson2(join12(outputDir, "resolved-workflow.json"), {
+  await writePrivateJson2(join13(outputDir, "resolved-workflow.json"), {
     spec: args2.spec,
     plan: args2.plan,
     model: { id: model.id, revision: model.revision },
@@ -10485,13 +10555,13 @@ async function runDecisionPipeline(args2) {
   const steps = [];
   let tunedModelDir;
   for (const step of args2.plan.steps) {
-    const stepDir = join12(outputDir, step.id);
+    const stepDir = join13(outputDir, step.id);
     await mkdir4(stepDir, { recursive: true, mode: 448 });
-    const configPath2 = join12(stepDir, "config.json");
-    const logPath = join12(stepDir, "step.log");
-    const metricsPath = join12(stepDir, "metrics.json");
+    const configPath2 = join13(stepDir, "config.json");
+    const logPath = join13(stepDir, "step.log");
+    const metricsPath = join13(stepDir, "metrics.json");
     if (step.uses === "train") {
-      const modelDir = join12(stepDir, "model");
+      const modelDir = join13(stepDir, "model");
       args2.onProgress?.(`${step.id}: fine-tuning ${model.id} on ${split.train.length} examples (log: ${logPath})`);
       await writePrivateJson2(configPath2, {
         model: model.id,
@@ -10511,8 +10581,8 @@ async function runDecisionPipeline(args2) {
       await spawnStep({ entrypoint: "train.py", configPath: configPath2, logPath, stepId: step.id, onProgress: args2.onProgress });
       const metrics = await readJsonObject2(metricsPath, `Decision step "${step.id}" metrics`);
       if (metrics.ok !== true) throw new Error(`Decision step "${step.id}" metrics must report ok: true.`);
-      await requireRegularNonemptyFile2(join12(modelDir, "model.safetensors"), `Decision step "${step.id}" model.safetensors`);
-      await readJsonObject2(join12(modelDir, "rl_agent_config.json"), `Decision step "${step.id}" rl_agent_config.json`);
+      await requireRegularNonemptyFile2(join13(modelDir, "model.safetensors"), `Decision step "${step.id}" model.safetensors`);
+      await readJsonObject2(join13(modelDir, "rl_agent_config.json"), `Decision step "${step.id}" rl_agent_config.json`);
       producedModels.set(step.id, modelDir);
       tunedModelDir = modelDir;
       steps.push({ id: step.id, uses: step.uses, model: modelDir, training: metrics });
@@ -10523,7 +10593,7 @@ async function runDecisionPipeline(args2) {
       const tuned = source !== "base";
       const modelDir = tuned ? producedModels.get(producer(source.from)) : void 0;
       if (tuned && !modelDir) throw new Error(`Step ${step.id} needs a model from ${source.from}.`);
-      const predictionsPath = join12(stepDir, "predictions.jsonl");
+      const predictionsPath = join13(stepDir, "predictions.jsonl");
       args2.onProgress?.(`${step.id}: evaluating ${tuned ? "tuned" : "base"} model on ${split.holdout.length} held-out examples`);
       await writePrivateJson2(configPath2, {
         model: modelDir ?? model.id,
@@ -10537,7 +10607,7 @@ async function runDecisionPipeline(args2) {
       await spawnStep({ entrypoint: "evaluate.py", configPath: configPath2, logPath, stepId: step.id, onProgress: args2.onProgress });
       const metrics = await readJsonObject2(metricsPath, `Decision step "${step.id}" metrics`);
       if (metrics.ok !== true) throw new Error(`Decision step "${step.id}" metrics must report ok: true.`);
-      const reportPath2 = join12(stepDir, "report.json");
+      const reportPath2 = join13(stepDir, "report.json");
       const scored = scoreDecisionPredictions({
         kind: tuned ? "candidate" : "baseline",
         modelId: modelDir ? fileUri(modelDir) : model.id,
@@ -10561,10 +10631,10 @@ async function runDecisionPipeline(args2) {
       log_loss_delta: after.metrics.log_loss - before.metrics.log_loss,
       brier_delta: after.metrics.brier - before.metrics.brier
     };
-    await writePrivateJson2(join12(stepDir, "comparison.json"), comparison);
+    await writePrivateJson2(join13(stepDir, "comparison.json"), comparison);
     steps.push({ id: step.id, uses: step.uses, comparison });
   }
-  const reportPath = join12(outputDir, "report.json");
+  const reportPath = join13(outputDir, "report.json");
   const report = {
     status: "succeeded",
     engine: "decision",
@@ -11190,7 +11260,7 @@ function registerTopupCommands(parent) {
 import { createInterface as createInterface4 } from "readline/promises";
 import { existsSync as existsSync7, readFileSync as readFileSync8 } from "fs";
 import { homedir as homedir2 } from "os";
-import { isAbsolute as isAbsolute5, join as join14, resolve as resolve9 } from "path";
+import { isAbsolute as isAbsolute5, join as join15, resolve as resolve9 } from "path";
 import { stdin as defaultStdin, stdout as defaultStdout } from "process";
 var MAX_PUBLISH_BYTES = 2 * 1024 * 1024;
 function expandPath2(value, baseDirectory, homeDirectory) {
@@ -11202,8 +11272,8 @@ function resolvePublishStoreRoot(options) {
   const cwd = options?.cwd ?? process.cwd();
   const env = options?.env ?? process.env;
   const homeDirectory = options?.homeDirectory ?? (env.HOME ? resolve9(env.HOME) : homedir2());
-  const localConfigPath = join14(cwd, "local-runner.json");
-  const projectPath = join14(cwd, "tunedtensor.json");
+  const localConfigPath = join15(cwd, "local-runner.json");
+  const projectPath = join15(cwd, "tunedtensor.json");
   if (existsSync7(projectPath)) {
     const spec = localBehaviorSpecFileSchema.parse(JSON.parse(readFileSync8(projectPath, "utf8")));
     if (existsSync7(localConfigPath) && (spec.runtime || spec.evaluation)) throw new Error("Conflicting configuration sources; migrate local-runner.json first.");
@@ -12272,31 +12342,166 @@ Eval Results (${data3._evals.length}):`);
   });
 }
 
+// src/commands/endpoints.ts
+import { createHash as createHash8, randomUUID as randomUUID10 } from "crypto";
+import { createReadStream as createReadStream2, readFileSync as readFileSync11, statSync as statSync4, openSync as openSync3, readSync, closeSync as closeSync3 } from "fs";
+import { open as open4 } from "fs/promises";
+import { finished } from "stream/promises";
+import { resolve as resolve10, join as join16 } from "path";
+import { z } from "zod";
+async function inspectCheckpoint(modelPath) {
+  const directory = resolve10(modelPath);
+  const config = JSON.parse(readFileSync11(join16(directory, "rl_agent_config.json"), "utf8"));
+  if (config.encoder !== "answerdotai/ModernBERT-large" || config.head_layers !== 2 || config.max_len !== 512 || config.head_max_len !== 192) {
+    throw new Error("Hosting currently supports only the pinned English Laya checkpoint and TT decision fine-tunes.");
+  }
+  const kind = config.tuned_tensor ? "tuned" : "base";
+  if (kind === "tuned" && (JSON.stringify(config.temperature) !== "[1,1,1]" || config.temperature_by_options)) {
+    throw new Error("Custom calibration is not supported by hosted decision inference.");
+  }
+  const file = join16(directory, "model.safetensors");
+  const size = statSync4(file).size;
+  if (size <= 8 || size > 2e9) throw new Error("Weights must be a safetensors file smaller than 2 GB. GPU and adapter models are not supported.");
+  const fd = openSync3(file, "r");
+  try {
+    const length = Buffer.alloc(8);
+    if (readSync(fd, length, 0, 8, 0) !== 8) throw new Error("Invalid safetensors header.");
+    const headerSize = Number(length.readBigUInt64LE());
+    if (!Number.isSafeInteger(headerSize) || headerSize < 2 || headerSize > 1e6 || headerSize + 8 >= size) throw new Error("Invalid safetensors header.");
+    const header = Buffer.alloc(headerSize);
+    if (readSync(fd, header, 0, headerSize, 8) !== headerSize) throw new Error("Truncated safetensors header.");
+    const tensors = JSON.parse(header.toString("utf8"));
+    for (const prefix of ["encoder.", "type_emb.", "scorer.", "act_head."]) {
+      if (!Object.keys(tensors).some((key) => key.startsWith(prefix))) throw new Error("Not a complete Laya checkpoint.");
+    }
+  } finally {
+    closeSync3(fd);
+  }
+  const hash = createHash8("sha256");
+  for await (const chunk of createReadStream2(file)) hash.update(chunk);
+  return { file, weights_bytes: size, weights_sha256: hash.digest("hex"), checkpoint_kind: kind };
+}
+function readInferencePayload(file) {
+  const path = resolve10(file);
+  if (statSync4(path).size > 131072) throw new Error("Inference JSON must be at most 128 KiB.");
+  const data3 = JSON.parse(readFileSync11(path, "utf8"));
+  return z.record(z.string(), z.json()).refine((value) => Object.keys(value).length > 0).parse(data3);
+}
+async function uploadWeights(endpoint, file) {
+  const url = new URL(endpoint.upload.url);
+  if (url.protocol !== "https:" || url.username || url.password || !/\.s3\.[a-z0-9-]+\.amazonaws\.com$/.test(url.hostname)) {
+    throw new Error("The server returned an invalid model upload URL.");
+  }
+  const handle = await open4(file, "r");
+  const stream = handle.createReadStream();
+  const completion = finished(stream);
+  void completion.catch(() => {
+  });
+  try {
+    const response = await fetch(url, {
+      method: "PUT",
+      headers: endpoint.upload.headers,
+      body: stream,
+      duplex: "half",
+      redirect: "error",
+      signal: AbortSignal.timeout(60 * 6e4)
+    });
+    if (!response.ok) throw new Error(`Model upload failed (${response.status}). Delete endpoint ${endpoint.id} and retry.`);
+  } finally {
+    stream.destroy();
+    await completion.catch(() => {
+    });
+  }
+}
+function show(endpoint, opts) {
+  const result2 = { ...endpoint, invocation_url: new URL(endpoint.invocation_path, getBaseUrl(opts)).toString() };
+  if (isJsonMode()) return printJson(result2);
+  printDetail([
+    ["Endpoint", result2.id],
+    ["Name", result2.name],
+    ["Status", result2.status],
+    ["Inference URL", result2.invocation_url],
+    ["Price per successful request", `${result2.price_cents} cents`],
+    ...result2.inference_key ? [["Inference key (save securely)", result2.inference_key]] : []
+  ]);
+}
+function registerEndpointCommands(parent) {
+  const endpoints = parent.command("endpoints").description("Host small CPU decision models using prepaid TT credits");
+  endpoints.command("create").description("Upload Laya weights and deploy a serverless endpoint").requiredOption("--model <directory>", "Local Laya checkpoint (e.g. a decision run's train/model)").requiredOption("--name <name>", "Endpoint display name").action(async (options) => {
+    const opts = parent.optsWithGlobals();
+    const { file, ...artifact } = await inspectCheckpoint(options.model);
+    const { data: data3 } = await post("/endpoints", { name: options.name, format: "laya-safetensors-v1", ...artifact }, opts);
+    if (!isJsonMode()) console.error(`Endpoint ${data3.id} created. Uploading ${(artifact.weights_bytes / 1e9).toFixed(2)} GB of weights...`);
+    try {
+      await uploadWeights(data3, file);
+      const { data: deployed } = await post(`/endpoints/${data3.id}/deploy`, {}, opts);
+      show({ ...deployed, inference_key: data3.inference_key }, opts);
+    } catch (error) {
+      throw new Error(`Endpoint ${data3.id}: ${error instanceof Error ? error.message : "upload/deployment failed"}. Use tt cloud endpoints get ${data3.id} or delete ${data3.id}. Rotate its key if the create response was lost.`, { cause: error });
+    }
+  });
+  endpoints.command("list").action(async () => {
+    const opts = parent.optsWithGlobals();
+    const { data: data3 } = await get("/endpoints", void 0, opts);
+    if (isJsonMode()) return printJson(data3);
+    for (const row of data3) show(row, opts);
+  });
+  endpoints.command("get <id>").description("Check deployment or deletion status").action(async (id) => {
+    z.uuid().parse(id);
+    const opts = parent.optsWithGlobals();
+    show((await get(`/endpoints/${id}`, void 0, opts)).data, opts);
+  });
+  endpoints.command("delete <id>").description("Stop serving and delete cloud weights and resources").action(async (id) => {
+    z.uuid().parse(id);
+    const opts = parent.optsWithGlobals();
+    show((await del(`/endpoints/${id}`, opts)).data, opts);
+  });
+  endpoints.command("rotate-key <id>").description("Revoke the old inference key and return a new key").action(async (id) => {
+    z.uuid().parse(id);
+    const { data: data3 } = await post(`/endpoints/${id}/key`, {}, parent.optsWithGlobals());
+    if (isJsonMode()) return printJson(data3);
+    printDetail([["Inference key (save securely)", data3.inference_key]]);
+  });
+  endpoints.command("invoke <id>").requiredOption("--payload <file>", "Complete model request as a JSON object (up to 128 KiB)").option("--request-id <uuid>", "Reuse this ID to retry without a second charge").action(async (id, options) => {
+    z.uuid().parse(id);
+    const requestId = options.requestId ?? randomUUID10();
+    z.uuid().parse(requestId);
+    const payload = readInferencePayload(options.payload);
+    if (!isJsonMode()) console.error(`Request ID: ${requestId} (reuse with --request-id when retrying)`);
+    try {
+      const { data: data3 } = await post(`/endpoints/${id}/invoke`, payload, parent.optsWithGlobals(), { "Idempotency-Key": requestId });
+      printJson({ request_id: requestId, data: data3 });
+    } catch (error) {
+      throw new Error(`Request ${requestId}: ${error instanceof Error ? error.message : "inference failed"}. Retry with --request-id ${requestId}.`, { cause: error });
+    }
+  });
+}
+
 // src/commands/models.ts
-import { spawn as spawn3, execFileSync } from "child_process";
-import { createHash as createHash8 } from "crypto";
+import { spawn as spawn4, execFileSync } from "child_process";
+import { createHash as createHash9 } from "crypto";
 import {
   existsSync as existsSync10,
   mkdirSync as mkdirSync5,
-  statSync as statSync4,
+  statSync as statSync5,
   createWriteStream,
   unlinkSync,
-  readFileSync as readFileSync12,
+  readFileSync as readFileSync13,
   writeFileSync as writeFileSync6,
   readdirSync,
   rmSync
 } from "fs";
 import { homedir as homedir3 } from "os";
-import { dirname as dirname13, join as join15, resolve as resolve11, basename as basename9, normalize, isAbsolute as isAbsolute6 } from "path";
+import { dirname as dirname13, join as join17, resolve as resolve12, basename as basename9, normalize, isAbsolute as isAbsolute6 } from "path";
 import { Readable, Transform } from "stream";
 import { pipeline } from "stream/promises";
 
 // src/commands/init.ts
-import { existsSync as existsSync9, readFileSync as readFileSync11, writeFileSync as writeFileSync5 } from "fs";
-import { resolve as resolve10 } from "path";
+import { existsSync as existsSync9, readFileSync as readFileSync12, writeFileSync as writeFileSync5 } from "fs";
+import { resolve as resolve11 } from "path";
 var DEFAULT_SPEC_FILE2 = "tunedtensor.json";
 function loadSpec(filePath) {
-  const resolved = resolve10(filePath || DEFAULT_SPEC_FILE2);
+  const resolved = resolve11(filePath || DEFAULT_SPEC_FILE2);
   if (!existsSync9(resolved)) {
     printError(
       `Spec file not found: ${filePath || DEFAULT_SPEC_FILE2}
@@ -12304,13 +12509,13 @@ Run \`tt init\` to create one.`
     );
     process.exit(1);
   }
-  const raw = JSON.parse(readFileSync11(resolved, "utf-8"));
+  const raw = JSON.parse(readFileSync12(resolved, "utf-8"));
   return raw;
 }
 
 // src/serve-manager.ts
-import { spawn as spawn2 } from "child_process";
-import { randomUUID as randomUUID10 } from "crypto";
+import { spawn as spawn3 } from "child_process";
+import { randomUUID as randomUUID11 } from "crypto";
 import { appendFileSync, mkdirSync as mkdirSync4 } from "fs";
 import { createServer } from "http";
 import { createServer as createNetServer } from "net";
@@ -12327,10 +12532,10 @@ function parsePositiveInteger(value, fallback) {
   return Number.isInteger(value) && value >= 0 ? value : fallback;
 }
 function sleep2(ms) {
-  return new Promise((resolve13) => setTimeout(resolve13, ms));
+  return new Promise((resolve14) => setTimeout(resolve14, ms));
 }
 function readRequestBody(req) {
-  return new Promise((resolve13, reject) => {
+  return new Promise((resolve14, reject) => {
     if (Number(req.headers["content-length"]) > MAX_SERVE_REQUEST_BYTES) {
       reject(new RequestTooLargeError());
       return;
@@ -12360,7 +12565,7 @@ function readRequestBody(req) {
     };
     const onEnd = () => {
       cleanup();
-      resolve13(Buffer.concat(chunks, bytes));
+      resolve14(Buffer.concat(chunks, bytes));
     };
     req.on("data", onData);
     req.on("end", onEnd);
@@ -12437,7 +12642,7 @@ function errorSummary(response, fallback) {
   return typeof message === "string" ? message : fallback;
 }
 async function allocatePort(host) {
-  return new Promise((resolve13, reject) => {
+  return new Promise((resolve14, reject) => {
     const server = createNetServer();
     server.on("error", reject);
     server.listen(0, host, () => {
@@ -12447,14 +12652,14 @@ async function allocatePort(host) {
         return;
       }
       const port = address.port;
-      server.close(() => resolve13(port));
+      server.close(() => resolve14(port));
     });
   });
 }
 var ManagedModelServer = class {
   constructor(options) {
     this.options = options;
-    this.spawnModelServer = options.spawnModelServer ?? spawn2;
+    this.spawnModelServer = options.spawnModelServer ?? spawn3;
     this.idleTimeoutMs = parsePositiveInteger(options.idleTimeoutSeconds, 300) * 1e3;
     this.restartAfterRequests = parsePositiveInteger(options.restartAfterRequests, 100);
     this.logRecordOverride = options.logRecord;
@@ -12488,11 +12693,11 @@ var ManagedModelServer = class {
         sendJson(res, 500, { error: { message: err instanceof Error ? err.message : String(err) } });
       });
     });
-    await new Promise((resolve13, reject) => {
+    await new Promise((resolve14, reject) => {
       this.server?.once("error", reject);
       this.server?.listen(this.options.publicPort, this.options.publicHost, () => {
         this.server?.off("error", reject);
-        resolve13();
+        resolve14();
       });
     });
   }
@@ -12502,15 +12707,15 @@ var ManagedModelServer = class {
     if (!this.server) return;
     const server = this.server;
     this.server = null;
-    await new Promise((resolve13, reject) => {
-      server.close((err) => err ? reject(err) : resolve13());
+    await new Promise((resolve14, reject) => {
+      server.close((err) => err ? reject(err) : resolve14());
     });
   }
   async run() {
     await this.start();
-    await new Promise((resolve13, reject) => {
+    await new Promise((resolve14, reject) => {
       this.server?.once("error", reject);
-      this.server?.once("close", resolve13);
+      this.server?.once("close", resolve14);
     });
   }
   async handle(req, res) {
@@ -12591,7 +12796,7 @@ var ManagedModelServer = class {
     });
   }
   async processGeneration(route, body, res, receivedAt) {
-    const requestId = `ttreq-${randomUUID10()}`;
+    const requestId = `ttreq-${randomUUID11()}`;
     const startedAt = Date.now();
     const queuedMs = startedAt - receivedAt;
     this.activeRequests += 1;
@@ -12697,7 +12902,7 @@ var ManagedModelServer = class {
     try {
       child.kill("SIGTERM");
       await Promise.race([
-        new Promise((resolve13) => child.once("exit", () => resolve13())),
+        new Promise((resolve14) => child.once("exit", () => resolve14())),
         sleep2(5e3).then(() => {
           if (child.exitCode == null && child.signalCode == null) child.kill("SIGKILL");
         })
@@ -13328,8 +13533,8 @@ function validateDownloadFilename(filename) {
 function resolveOutputPath(output, filename) {
   validateDownloadFilename(filename);
   if (!output) return filename;
-  if (existsSync10(output) && statSync4(output).isDirectory()) {
-    return join15(output, filename);
+  if (existsSync10(output) && statSync5(output).isDirectory()) {
+    return join17(output, filename);
   }
   return output;
 }
@@ -13436,15 +13641,15 @@ async function downloadUrlToFile(url, outputPath) {
   return contentLength;
 }
 function defaultCacheDir() {
-  const root = process.env.XDG_CACHE_HOME || join15(homedir3(), ".cache");
-  return join15(root, "tuned-tensor", "models");
+  const root = process.env.XDG_CACHE_HOME || join17(homedir3(), ".cache");
+  return join17(root, "tuned-tensor", "models");
 }
 function runtimeDir(cacheDir) {
-  return join15(cacheDir, "_runtime");
+  return join17(cacheDir, "_runtime");
 }
 function runtimePythonPath(cacheDir) {
   const dir = runtimeDir(cacheDir);
-  return process.platform === "win32" ? join15(dir, "Scripts", "python.exe") : join15(dir, "bin", "python");
+  return process.platform === "win32" ? join17(dir, "Scripts", "python.exe") : join17(dir, "bin", "python");
 }
 function resolveServePython(cacheDir, explicitPython) {
   if (explicitPython) return explicitPython;
@@ -13517,7 +13722,7 @@ print(json.dumps({"missing": missing}))
   }
 }
 function setupRuntimeCommands(python, venvDir, deps) {
-  const venvPython = process.platform === "win32" ? join15(venvDir, "Scripts", "python.exe") : join15(venvDir, "bin", "python");
+  const venvPython = process.platform === "win32" ? join17(venvDir, "Scripts", "python.exe") : join17(venvDir, "bin", "python");
   return [
     [python, "-m", "venv", venvDir],
     [venvPython, "-m", "pip", "install", "--upgrade", "pip"],
@@ -13528,7 +13733,7 @@ function safeSegment(value) {
   return value.replace(/[^a-zA-Z0-9._-]+/g, "-").replace(/^-+|-+$/g, "") || "model";
 }
 function directoryHasFiles(path) {
-  return existsSync10(path) && statSync4(path).isDirectory() && readdirSync(path).length > 0;
+  return existsSync10(path) && statSync5(path).isDirectory() && readdirSync(path).length > 0;
 }
 function validateArchiveEntryPath(entry) {
   const normalized = normalize(entry).replace(/^[\\/]+/, "");
@@ -13555,9 +13760,9 @@ function extractArchive(archivePath, targetDir, force) {
   return targetDir;
 }
 function localArchiveCacheDir(archivePath, cacheDir) {
-  const resolved = resolve11(archivePath);
-  const digest = createHash8("sha256").update(resolved).digest("hex").slice(0, 12);
-  return join15(cacheDir, `local-${safeSegment(basename9(archivePath))}-${digest}`);
+  const resolved = resolve12(archivePath);
+  const digest = createHash9("sha256").update(resolved).digest("hex").slice(0, 12);
+  return join17(cacheDir, `local-${safeSegment(basename9(archivePath))}-${digest}`);
 }
 function buildSystemPrompt(spec) {
   const parts = [];
@@ -13573,40 +13778,40 @@ ${spec.constraints.map((c) => `- ${c}`).join("\n")}`);
   return parts.join("\n\n");
 }
 function findSpecPath(explicitPath, modelPath) {
-  const candidates = explicitPath ? [explicitPath] : [join15(process.cwd(), DEFAULT_SPEC_FILE2), join15(modelPath, DEFAULT_SPEC_FILE2)];
+  const candidates = explicitPath ? [explicitPath] : [join17(process.cwd(), DEFAULT_SPEC_FILE2), join17(modelPath, DEFAULT_SPEC_FILE2)];
   for (const candidate of candidates) {
-    const resolved = resolve11(candidate);
-    if (existsSync10(resolved) && statSync4(resolved).isFile()) return resolved;
+    const resolved = resolve12(candidate);
+    if (existsSync10(resolved) && statSync5(resolved).isFile()) return resolved;
   }
   return null;
 }
 function loadSystemPromptFromSpec(specPath) {
-  const spec = JSON.parse(readFileSync12(specPath, "utf8"));
+  const spec = JSON.parse(readFileSync13(specPath, "utf8"));
   return buildSystemPrompt(spec);
 }
 function loadJsonSchemaForServe(schemaPath) {
-  const resolved = resolve11(schemaPath);
-  if (!existsSync10(resolved) || !statSync4(resolved).isFile()) {
+  const resolved = resolve12(schemaPath);
+  if (!existsSync10(resolved) || !statSync5(resolved).isFile()) {
     throw new Error(`JSON schema file not found: ${schemaPath}`);
   }
   try {
-    const schema = JSON.parse(readFileSync12(resolved, "utf8"));
+    const schema = JSON.parse(readFileSync13(resolved, "utf8"));
     return JSON.stringify(schema);
   } catch (err) {
     throw new Error(`JSON schema file is not valid JSON: ${schemaPath}`);
   }
 }
 function writeReferenceServerScript(cacheDir) {
-  const scriptDir = join15(cacheDir, "_server");
+  const scriptDir = join17(cacheDir, "_server");
   mkdirSync5(scriptDir, { recursive: true });
-  const scriptPath = join15(scriptDir, "openai_reference_server.py");
+  const scriptPath = join17(scriptDir, "openai_reference_server.py");
   writeFileSync6(scriptPath, REFERENCE_SERVER_SCRIPT, "utf8");
   return scriptPath;
 }
 async function resolveServeTarget(target, opts, cacheDir, forceDownload) {
   if (existsSync10(target)) {
-    const resolved = resolve11(target);
-    const stats = statSync4(resolved);
+    const resolved = resolve12(target);
+    const stats = statSync5(resolved);
     if (stats.isDirectory()) {
       return {
         modelPath: resolved,
@@ -13615,7 +13820,7 @@ async function resolveServeTarget(target, opts, cacheDir, forceDownload) {
       };
     }
     if (stats.isFile() && (resolved.endsWith(".tar.gz") || resolved.endsWith(".tgz"))) {
-      const extractDir2 = join15(localArchiveCacheDir(resolved, cacheDir), "model");
+      const extractDir2 = join17(localArchiveCacheDir(resolved, cacheDir), "model");
       return {
         modelPath: extractArchive(resolved, extractDir2, forceDownload),
         modelName: basename9(resolved).replace(/\.t(ar\.)?gz$/, ""),
@@ -13629,9 +13834,9 @@ async function resolveServeTarget(target, opts, cacheDir, forceDownload) {
   const { data: data3 } = await get(`/models/${fullId}/download`, void 0, opts);
   validateDownloadFilename(fullId);
   validateDownloadFilename(data3.filename);
-  const modelCacheDir = join15(cacheDir, fullId);
-  const archivePath = join15(modelCacheDir, data3.filename);
-  const extractDir = join15(modelCacheDir, "model");
+  const modelCacheDir = join17(cacheDir, fullId);
+  const archivePath = join17(modelCacheDir, data3.filename);
+  const extractDir = join17(modelCacheDir, "model");
   if (!existsSync10(archivePath) || forceDownload) {
     await downloadUrlToFile(data3.url, archivePath);
   }
@@ -13736,8 +13941,8 @@ function ollamaSlug(name) {
 }
 function firstExisting(candidates) {
   for (const candidate of candidates) {
-    if (candidate && existsSync10(candidate) && statSync4(candidate).isFile()) {
-      return resolve11(candidate);
+    if (candidate && existsSync10(candidate) && statSync5(candidate).isFile()) {
+      return resolve12(candidate);
     }
   }
   return null;
@@ -13747,14 +13952,14 @@ function llamaCppDir(explicit) {
 }
 function resolveConvertScript(opts, required) {
   if (opts.convertScript) {
-    const resolved = resolve11(opts.convertScript);
+    const resolved = resolve12(opts.convertScript);
     if (required && !existsSync10(resolved)) {
       throw new Error(`Conversion script not found: ${opts.convertScript}`);
     }
     return resolved;
   }
   const dir = llamaCppDir(opts.llamaCpp);
-  const candidates = dir ? [join15(dir, "convert_hf_to_gguf.py"), join15(dir, "convert-hf-to-gguf.py")] : [];
+  const candidates = dir ? [join17(dir, "convert_hf_to_gguf.py"), join17(dir, "convert-hf-to-gguf.py")] : [];
   const found = firstExisting(candidates);
   if (found) return found;
   if (required) {
@@ -13762,11 +13967,11 @@ function resolveConvertScript(opts, required) {
       "Could not find convert_hf_to_gguf.py. Pass --llama-cpp <dir> (a llama.cpp checkout) or --convert-script <path>. Get it from https://github.com/ggml-org/llama.cpp."
     );
   }
-  return dir ? join15(dir, "convert_hf_to_gguf.py") : "convert_hf_to_gguf.py";
+  return dir ? join17(dir, "convert_hf_to_gguf.py") : "convert_hf_to_gguf.py";
 }
 function resolveQuantizeBin(opts, required) {
   if (opts.quantizeBin) {
-    const resolved = resolve11(opts.quantizeBin);
+    const resolved = resolve12(opts.quantizeBin);
     if (required && !existsSync10(resolved)) {
       throw new Error(`llama-quantize binary not found: ${opts.quantizeBin}`);
     }
@@ -13775,10 +13980,10 @@ function resolveQuantizeBin(opts, required) {
   const dir = llamaCppDir(opts.llamaCpp);
   if (dir) {
     const candidates = [
-      join15(dir, "build", "bin", "llama-quantize"),
-      join15(dir, "build", "bin", "quantize"),
-      join15(dir, "llama-quantize"),
-      join15(dir, "quantize")
+      join17(dir, "build", "bin", "llama-quantize"),
+      join17(dir, "build", "bin", "quantize"),
+      join17(dir, "llama-quantize"),
+      join17(dir, "quantize")
     ];
     const found = firstExisting(candidates);
     if (found) return found;
@@ -13926,7 +14131,7 @@ function registerModelsCommands(parent) {
     }
     const quantPlan = planQuant(String(cmdOpts.quant));
     const printOnly = Boolean(cmdOpts.printCommand);
-    const cacheDir = resolve11(cmdOpts.cacheDir || defaultCacheDir());
+    const cacheDir = resolve12(cmdOpts.cacheDir || defaultCacheDir());
     mkdirSync5(cacheDir, { recursive: true });
     const serveTarget = await resolveServeTarget(
       target,
@@ -13935,9 +14140,9 @@ function registerModelsCommands(parent) {
       Boolean(cmdOpts.forceDownload)
     );
     const slug = ollamaSlug(serveTarget.modelName);
-    const outputDir = resolve11(cmdOpts.output || join15(process.cwd(), `${slug}-gguf`));
-    const finalPath = join15(outputDir, `${slug}.${quantPlan.quant}.gguf`);
-    const intermediatePath = quantPlan.requiresQuantize ? join15(outputDir, `${slug}.f16.gguf`) : void 0;
+    const outputDir = resolve12(cmdOpts.output || join17(process.cwd(), `${slug}-gguf`));
+    const finalPath = join17(outputDir, `${slug}.${quantPlan.quant}.gguf`);
+    const intermediatePath = quantPlan.requiresQuantize ? join17(outputDir, `${slug}.f16.gguf`) : void 0;
     const convertScript = resolveConvertScript(cmdOpts, !printOnly);
     const convertOutfile = intermediatePath ?? finalPath;
     const steps = [
@@ -13975,7 +14180,7 @@ function registerModelsCommands(parent) {
         }
         if (specPath) systemPrompt2 = loadSystemPromptFromSpec(specPath);
       }
-      modelfilePath = join15(outputDir, "Modelfile");
+      modelfilePath = join17(outputDir, "Modelfile");
       modelfileContent = buildModelfile(basename9(finalPath), systemPrompt2);
       const createModel = cmdOpts.ollamaCreate !== false;
       if (createModel) {
@@ -14062,7 +14267,7 @@ function registerModelsCommands(parent) {
     }
   });
   models.command("setup-runtime").description("Install an isolated Python runtime for local model serving").option("--python <path>", "Python 3.10-3.12 executable to create the runtime").option("--cache-dir <path>", "Cache directory for the managed runtime").option("-f, --force", "Recreate the runtime if it already exists").option("--print-command", "Print the setup commands without running them").action(async (cmdOpts) => {
-    const cacheDir = resolve11(cmdOpts.cacheDir || defaultCacheDir());
+    const cacheDir = resolve12(cmdOpts.cacheDir || defaultCacheDir());
     const venvDir = runtimeDir(cacheDir);
     const venvPython = runtimePythonPath(cacheDir);
     const python = pickRuntimePython(cmdOpts.python);
@@ -14108,7 +14313,7 @@ function registerModelsCommands(parent) {
   });
   models.command("serve").description("Serve a downloaded model with an OpenAI-compatible local API").argument("<target>", "Model ID/prefix, model directory, or .tar.gz artifact").option("--spec <path>", "Behavior spec JSON to apply as the default system prompt").option("--no-spec-prompt", "Do not auto-apply a behavior spec system prompt").option("--host <host>", "Host to bind", "127.0.0.1").option("--port <port>", "Port to bind", "8000").option("--python <path>", "Python executable").option("--cache-dir <path>", "Cache directory for downloaded/extracted models").option("--device <device>", "Inference device: auto, cpu, cuda, or mps", "auto").option("--force-download", "Re-download and re-extract model artifacts").option("--max-tokens <n>", "Default max completion tokens", "512").option("--temperature <n>", "Default sampling temperature", "0.7").option("--json-schema <path>", "JSON Schema file to enforce by default for chat completions").option("--json-repair-attempts <n>", "Number of JSON/schema repair retries", "1").option("--trust-remote-code", "Pass trust_remote_code=True to transformers").option("--managed", "Run a local lifecycle manager in front of the model server").option("--idle-timeout <seconds>", "Managed mode idle timeout before stopping the model", "300").option("--restart-after-requests <n>", "Managed mode restart threshold; 0 disables request-count restarts", "100").option("--gate-field <field>", "Response JSON field to log as the managed serving gate result", "should_process").option("--log-file <path>", "Write managed serving JSONL request logs to a file").option("--print-command", "Print the underlying Python command without starting it").action(async (target, cmdOpts) => {
     const opts = parent.optsWithGlobals();
-    const cacheDir = resolve11(cmdOpts.cacheDir || defaultCacheDir());
+    const cacheDir = resolve12(cmdOpts.cacheDir || defaultCacheDir());
     mkdirSync5(cacheDir, { recursive: true });
     const python = resolveServePython(cacheDir, cmdOpts.python);
     if (!cmdOpts.printCommand) ensureServingRuntime(python, cacheDir);
@@ -14157,7 +14362,7 @@ function registerModelsCommands(parent) {
       idle_timeout_seconds: idleTimeoutSeconds,
       restart_after_requests: restartAfterRequests,
       gate_field: String(cmdOpts.gateField),
-      log_file: cmdOpts.logFile ? resolve11(cmdOpts.logFile) : void 0
+      log_file: cmdOpts.logFile ? resolve12(cmdOpts.logFile) : void 0
     } : void 0;
     if (cmdOpts.printCommand) return printServeCommand(python, args2, env, managedConfig);
     if (!isJsonMode()) {
@@ -14194,12 +14399,12 @@ function registerModelsCommands(parent) {
         idleTimeoutSeconds,
         restartAfterRequests,
         gateField: String(cmdOpts.gateField),
-        logFile: cmdOpts.logFile ? resolve11(cmdOpts.logFile) : void 0
+        logFile: cmdOpts.logFile ? resolve12(cmdOpts.logFile) : void 0
       });
       return;
     }
     await new Promise((resolvePromise, reject) => {
-      const child = spawn3(python, args2, {
+      const child = spawn4(python, args2, {
         stdio: "inherit",
         env: { ...process.env, ...env }
       });
@@ -14220,15 +14425,15 @@ function registerModelsCommands(parent) {
 }
 
 // src/commands/push.ts
-import { readFileSync as readFileSync13, writeFileSync as writeFileSync7 } from "fs";
-import { resolve as resolve12 } from "path";
+import { readFileSync as readFileSync14, writeFileSync as writeFileSync7 } from "fs";
+import { resolve as resolve13 } from "path";
 function persistRemoteId(filePath, id) {
-  const raw = JSON.parse(readFileSync13(filePath, "utf-8"));
+  const raw = JSON.parse(readFileSync14(filePath, "utf-8"));
   raw.id = id;
   writeFileSync7(filePath, JSON.stringify(raw, null, 2) + "\n");
 }
 async function pushLocalSpec(file, opts) {
-  const filePath = resolve12(file);
+  const filePath = resolve13(file);
   const spec = loadSpec(file);
   const evalCaseErrors = validateEvalCases(spec);
   if (evalCaseErrors.length > 0) {
@@ -14287,6 +14492,7 @@ function registerCloudCommands(parent) {
   registerLabelCommands(cloud);
   registerRunsCommands(cloud);
   registerModelsCommands(cloud);
+  registerEndpointCommands(cloud);
 }
 function weeklyRows(data3) {
   if (typeof data3.weekly_remaining_percent !== "number" || !Number.isFinite(data3.weekly_remaining_percent) || typeof data3.weekly_cost_limit_usd !== "number" || !Number.isFinite(data3.weekly_cost_limit_usd) || data3.weekly_cost_limit_usd <= 0) return [];
@@ -14325,15 +14531,15 @@ Use \`tt balance\` for cloud training credits.`);
 }
 
 // src/commands/upgrade.ts
-import { spawn as spawn5 } from "child_process";
-import { accessSync, constants as constants2, existsSync as existsSync11, readFileSync as readFileSync15, realpathSync } from "fs";
-import { dirname as dirname15, join as join17 } from "path";
+import { spawn as spawn6 } from "child_process";
+import { accessSync, constants as constants2, existsSync as existsSync11, readFileSync as readFileSync16, realpathSync } from "fs";
+import { dirname as dirname15, join as join19, win32 } from "path";
 import chalk14 from "chalk";
 
 // src/update-check.ts
-import { spawn as spawn4 } from "child_process";
-import { mkdirSync as mkdirSync6, readFileSync as readFileSync14, writeFileSync as writeFileSync8 } from "fs";
-import { dirname as dirname14, join as join16 } from "path";
+import { spawn as spawn5 } from "child_process";
+import { mkdirSync as mkdirSync6, readFileSync as readFileSync15, writeFileSync as writeFileSync8 } from "fs";
+import { dirname as dirname14, join as join18 } from "path";
 import chalk13 from "chalk";
 var CLI_PACKAGE_NAME = "@tuned-tensor/cli";
 var PACKAGE_LATEST_URL = "https://registry.npmjs.org/@tuned-tensor%2fcli/latest";
@@ -14382,14 +14588,14 @@ function compareVersions(left, right) {
   return comparePrerelease(left.prerelease, right.prerelease);
 }
 function getCliUpdateCacheFile(env = process.env) {
-  return join16(getTunedTensorHome(env), "update-check.json");
+  return join18(getTunedTensorHome(env), "update-check.json");
 }
 function cliUpdateChecksDisabled(env = process.env) {
   return [env.TT_NO_UPDATE_CHECK, env.NO_UPDATE_NOTIFIER, env.CI].some((value) => value !== void 0 && value !== "" && value !== "0" && value !== "false");
 }
 function readUpdateCache(file) {
   try {
-    const payload = JSON.parse(readFileSync14(file, "utf8"));
+    const payload = JSON.parse(readFileSync15(file, "utf8"));
     if (typeof payload.checkedAt !== "number" || typeof payload.latestVersion !== "string" || !parseVersion(payload.latestVersion)) return null;
     return { checkedAt: payload.checkedAt, latestVersion: payload.latestVersion };
   } catch {
@@ -14441,10 +14647,10 @@ async function fetchLatestCliVersion(options = {}) {
       return null;
     }
   })();
-  const deadline = new Promise((resolve13) => {
+  const deadline = new Promise((resolve14) => {
     timeout = setTimeout(() => {
       controller.abort();
-      resolve13(null);
+      resolve14(null);
     }, timeoutMs);
   });
   try {
@@ -14484,7 +14690,7 @@ function spawnCliUpdateRefresh(cacheFile) {
   const cached = readUpdateCache(cacheFile);
   if (cached) recordLatestCliVersion(cacheFile, cached.latestVersion);
   try {
-    const child = spawn4(
+    const child = spawn5(
       process.execPath,
       ["-e", REFRESH_SCRIPT, PACKAGE_LATEST_URL, cacheFile],
       { detached: true, stdio: "ignore", windowsHide: true }
@@ -14526,10 +14732,11 @@ function resolveGlobalInstall(entrypoint, platform = process.platform) {
   if (/[\\/]_npx[\\/]/.test(script)) return null;
   const match = platform === "win32" ? script.match(/^(.*)\\node_modules\\@tuned-tensor\\cli\\dist\\index\.js$/i) : script.match(/^(.*)\/lib\/node_modules\/@tuned-tensor\/cli\/dist\/index\.js$/);
   if (!match?.[1]) return null;
-  return { prefix: match[1], packageDir: dirname15(dirname15(script)) };
+  const pathApi = platform === "win32" ? win32 : { dirname: dirname15 };
+  return { prefix: match[1], packageDir: pathApi.dirname(pathApi.dirname(script)) };
 }
 function prefixIsWritable(prefix, platform) {
-  const modules = platform === "win32" ? join17(prefix, "node_modules") : join17(prefix, "lib", "node_modules");
+  const modules = platform === "win32" ? join19(prefix, "node_modules") : join19(prefix, "lib", "node_modules");
   try {
     accessSync(existsSync11(modules) ? modules : prefix, constants2.W_OK);
     return true;
@@ -14539,32 +14746,50 @@ function prefixIsWritable(prefix, platform) {
 }
 function installedVersion(packageDir) {
   try {
-    const manifest = JSON.parse(readFileSync15(join17(packageDir, "package.json"), "utf8"));
+    const manifest = JSON.parse(readFileSync16(join19(packageDir, "package.json"), "utf8"));
     return typeof manifest.version === "string" ? manifest.version : void 0;
   } catch {
     return void 0;
   }
 }
-function quoteForCmd(arg) {
-  return /[\s"&|<>^]/.test(arg) ? `"${arg.replace(/"/g, '""')}"` : arg;
+function resolveWindowsNpmCli(env, execPath) {
+  const path = Object.entries(env).find(([key]) => key.toLowerCase() === "path")?.[1] ?? "";
+  const directories = [win32.dirname(execPath), ...path.split(";")];
+  for (const directory of directories) {
+    if (!directory) continue;
+    const candidate = win32.join(directory.replace(/^"|"$/g, ""), "node_modules", "npm", "bin", "npm-cli.js");
+    if (existsSync11(candidate)) return candidate;
+  }
+  throw new Error("Could not locate npm-cli.js beside Node.js or on PATH. Reinstall Node.js with npm, then retry.");
 }
-function defaultNpmRunner(platform) {
-  return async (args2) => await new Promise((resolve13, reject) => {
+function isWindowsInstaller(prefix) {
+  try {
+    return JSON.parse(readFileSync16(join19(prefix, "tt-install.json"), "utf8")).manager === "windows-installer";
+  } catch {
+    return false;
+  }
+}
+function defaultNpmRunner(platform, env) {
+  return async (args2) => await new Promise((resolve14, reject) => {
     const windows = platform === "win32";
-    const child = spawn5(
-      windows ? "npm.cmd" : "npm",
-      windows ? args2.map(quoteForCmd) : args2,
+    const child = spawn6(
+      windows ? process.execPath : "npm",
+      windows ? [resolveWindowsNpmCli(env, process.execPath), ...args2] : args2,
       {
-        shell: windows,
+        env,
+        shell: false,
         // Keep stdout clean for --json; npm progress goes to stderr.
         stdio: ["ignore", isJsonMode() ? 2 : "inherit", "inherit"]
       }
     );
     child.once("error", reject);
-    child.once("close", (code2) => resolve13(code2 ?? 1));
+    child.once("close", (code2) => resolve14(code2 ?? 1));
   });
 }
-function formatCommand(args2) {
+function formatCommand(args2, platform) {
+  if (platform === "win32") {
+    return ["npm.cmd", ...args2].map((arg) => /^[a-zA-Z0-9_@./:\\+=-]+$/.test(arg) ? arg : "'" + arg.replaceAll("'", "''") + "'").join(" ");
+  }
   return ["npm", ...args2].map((arg) => /\s/.test(arg) ? JSON.stringify(arg) : arg).join(" ");
 }
 function registerUpgradeCommand(program, deps) {
@@ -14584,6 +14809,15 @@ Set TT_NO_UPDATE_CHECK=1 to silence update notices.
     const cacheFile = getCliUpdateCacheFile(deps.env);
     const current = deps.version;
     const target = requested?.trim() || "latest";
+    if (!/^[a-zA-Z0-9][a-zA-Z0-9.+-]*$/.test(target)) {
+      throw new Error("Choose an exact version or npm dist-tag, such as 0.27.0, latest, or beta.");
+    }
+    const install = resolveGlobalInstall(deps.entrypoint, platform);
+    if (!options.check && install && isWindowsInstaller(install.prefix)) {
+      throw new Error(
+        "This tt is managed by the Windows installer. Upgrade with:\n  winget upgrade --id TunedTensor.CLI --exact --source winget\nOr download and run the newer Windows setup from https://github.com/tunedtensor/tuned-tensor-cli/releases.\nWinGet requires the release to be available in its catalog; use setup for prereleases."
+      );
+    }
     let latest;
     if (target === "latest") {
       latest = await fetchLatestCliVersion({ fetchImpl: deps.fetchImpl, timeoutMs: 1e4 }) ?? void 0;
@@ -14614,7 +14848,6 @@ Set TT_NO_UPDATE_CHECK=1 to silence update notices.
       }
       return;
     }
-    const install = resolveGlobalInstall(deps.entrypoint, platform);
     const packageSpec = `${CLI_PACKAGE_NAME}@${target}`;
     if (!install) {
       throw new Error(
@@ -14626,26 +14859,27 @@ Upgrade it the way you installed it, for example:
     const npmArgs = ["install", "-g", "--ignore-scripts", "--prefix", install.prefix, packageSpec];
     if (!prefixIsWritable(install.prefix, platform)) {
       throw new Error(
-        `tt is installed in ${install.prefix}, which this user can't write to. Run:
-  sudo ${formatCommand(npmArgs)}`
+        `tt is installed in ${install.prefix}, which this user can't write to. ` + (platform === "win32" ? `Use an account with access to that directory and run:
+  ${formatCommand(npmArgs, platform)}` : `Run:
+  sudo ${formatCommand(npmArgs, platform)}`)
       );
     }
     if (!isJsonMode()) {
       console.log(`Upgrading tt ${current} \u2192 ${chalk14.bold(latest ?? target)}`);
-      console.log(chalk14.dim(`$ ${formatCommand(npmArgs)}`));
+      console.log(chalk14.dim(`$ ${formatCommand(npmArgs, platform)}`));
     }
     let code2;
     try {
-      code2 = await (deps.runNpm ?? defaultNpmRunner(platform))(npmArgs);
+      code2 = await (deps.runNpm ?? defaultNpmRunner(platform, deps.env))(npmArgs);
     } catch (error) {
       throw new Error(
         `Could not run npm (${error instanceof Error ? error.message : String(error)}). Run it yourself:
-  ${formatCommand(npmArgs)}`
+  ${formatCommand(npmArgs, platform)}`
       );
     }
     if (code2 !== 0) {
       throw new Error(`npm exited with code ${code2}. Run it yourself to see details:
-  ${formatCommand(npmArgs)}`);
+  ${formatCommand(npmArgs, platform)}`);
     }
     const installed = installedVersion(install.packageDir) ?? latest ?? target;
     if (isJsonMode()) {
@@ -14802,8 +15036,8 @@ async function runSelfCommand(args2, options = {}) {
   if (!entrypoint) {
     throw new Error("Cannot locate the tt CLI entrypoint.");
   }
-  return await new Promise((resolve13, reject) => {
-    const child = spawn6(process.execPath, [entrypoint, ...options.captureJson ? ["--json"] : [], ...args2], {
+  return await new Promise((resolve14, reject) => {
+    const child = spawn7(process.execPath, [entrypoint, ...options.captureJson ? ["--json"] : [], ...args2], {
       cwd: options.cwd,
       env: options.env ?? process.env,
       stdio: options.captureJson ? ["inherit", "pipe", "inherit"] : "inherit",
@@ -14849,7 +15083,7 @@ async function runSelfCommand(args2, options = {}) {
     child.once("close", (exitCode, signal) => {
       cleanup();
       const signalNumber = signal ? osConstants2.signals[signal] : void 0;
-      resolve13({
+      resolve14({
         exitCode: exitCode ?? (typeof signalNumber === "number" ? 128 + signalNumber : 1),
         signal,
         ...options.captureJson ? { stdout: stdout3 } : {}
@@ -15084,7 +15318,7 @@ ${formatCliUpdateNotice(update)}
 }
 
 // src/index.ts
-runCli("0.26.2").catch((err) => {
+runCli("0.28.0").catch((err) => {
   reportError(err);
   process.exit(1);
 });