@tuned-tensor/cli 0.11.0 → 0.12.0-beta.1

package/dist/index.js

@@ -1,10 +1,11 @@
 #!/usr/bin/env node
 import {
+  TRAINING_MODELS,
   assertLocalRunInputReady,
   loadLocalRunInput,
   loadLocalRunnerConfig,
   runLocalPipeline
-} from "./chunk-DAP2Z7EZ.js";
+} from "./chunk-HDGBDHOB.js";
 
 // src/cli.ts
 import { spawn as spawn4 } from "child_process";
@@ -24,76 +25,6 @@ import { join } from "path";
 import { Type } from "typebox";
 import { Value } from "typebox/value";
 
-// src/base-models.ts
-var SUPPORTED_BASE_MODELS = [
-  "google/gemma-4-E2B-it",
-  "google/gemma-4-E4B-it",
-  "Qwen/Qwen3.5-2B",
-  "Qwen/Qwen3-VL-2B-Instruct",
-  "Qwen/Qwen3.5-4B",
-  "meta-llama/Llama-3.2-3B-Instruct",
-  "microsoft/Phi-4-mini-instruct",
-  "ibm-granite/granite-3.3-2b-instruct",
-  "bigcode/starcoder2-3b"
-];
-var BASE_MODEL_ALIASES = /* @__PURE__ */ new Map();
-for (const model of SUPPORTED_BASE_MODELS) {
-  BASE_MODEL_ALIASES.set(model.toLowerCase(), model);
-}
-for (const [alias, model] of [
-  ["google/gemma-4-E2B", "google/gemma-4-E2B-it"],
-  ["google/gemma-4-e2b", "google/gemma-4-E2B-it"],
-  ["google/gemma-4-e2b-it", "google/gemma-4-E2B-it"],
-  ["google/gemma-4-E4B", "google/gemma-4-E4B-it"],
-  ["google/gemma-4-e4b", "google/gemma-4-E4B-it"],
-  ["google/gemma-4-e4b-it", "google/gemma-4-E4B-it"],
-  ["qwen/qwen3.5-2b", "Qwen/Qwen3.5-2B"],
-  ["Qwen/Qwen3.5-2B-Base", "Qwen/Qwen3.5-2B"],
-  ["qwen/qwen3.5-2b-base", "Qwen/Qwen3.5-2B"],
-  ["qwen/qwen3-vl-2b-instruct", "Qwen/Qwen3-VL-2B-Instruct"],
-  ["Qwen/Qwen3-VL-2B", "Qwen/Qwen3-VL-2B-Instruct"],
-  ["qwen/qwen3-vl-2b", "Qwen/Qwen3-VL-2B-Instruct"],
-  ["qwen/qwen3.5-4b", "Qwen/Qwen3.5-4B"],
-  ["Qwen/Qwen3.5-4B-Base", "Qwen/Qwen3.5-4B"],
-  ["qwen/qwen3.5-4b-base", "Qwen/Qwen3.5-4B"],
-  ["meta-llama/llama-3.2-3b-instruct", "meta-llama/Llama-3.2-3B-Instruct"],
-  ["meta-llama/Llama-3.2-3B", "meta-llama/Llama-3.2-3B-Instruct"],
-  ["meta-llama/llama-3.2-3b", "meta-llama/Llama-3.2-3B-Instruct"],
-  ["microsoft/phi-4-mini-instruct", "microsoft/Phi-4-mini-instruct"],
-  ["phi-4-mini-instruct", "microsoft/Phi-4-mini-instruct"],
-  ["ibm-granite/granite-3.3-2b-instruct", "ibm-granite/granite-3.3-2b-instruct"],
-  ["granite-3.3-2b-instruct", "ibm-granite/granite-3.3-2b-instruct"],
-  ["bigcode/starcoder2-3b", "bigcode/starcoder2-3b"],
-  ["starcoder2-3b", "bigcode/starcoder2-3b"]
-]) {
-  BASE_MODEL_ALIASES.set(alias.toLowerCase(), model);
-}
-var UnsupportedBaseModelError = class extends Error {
-  constructor(model) {
-    const supported = SUPPORTED_BASE_MODELS.join(", ");
-    super(
-      `Unsupported base_model "${String(model)}". Supported base models: ${supported}`
-    );
-    this.name = "UnsupportedBaseModelError";
-  }
-};
-function canonicalizeBaseModel(model) {
-  if (typeof model !== "string" || !model.trim()) {
-    throw new UnsupportedBaseModelError(model);
-  }
-  const canonical = BASE_MODEL_ALIASES.get(model.trim().toLowerCase());
-  if (!canonical) {
-    throw new UnsupportedBaseModelError(model);
-  }
-  return canonical;
-}
-function canonicalizeSpecBaseModel(body) {
-  if ("base_model" in body && body.base_model !== void 0) {
-    return { ...body, base_model: canonicalizeBaseModel(body.base_model) };
-  }
-  return body;
-}
-
 // src/eval/rules.ts
 function validateSpec(spec) {
   const checks = [];
@@ -258,15 +189,15 @@ var FolderName = Type.String({
   pattern: "^[A-Za-z0-9][A-Za-z0-9._-]*$",
   description: "One new folder name directly beneath the current workspace"
 });
-var BaseModel = Type.Unsafe({
+var LocalBaseModel = Type.String({
   type: "string",
-  enum: [...SUPPORTED_BASE_MODELS]
+  enum: TRAINING_MODELS.map((model) => model.id)
 });
 var NonEmptyText = Type.String({ minLength: 1, maxLength: 1e3 });
 var LocalProjectSpecSchema = Type.Object({
   name: Type.String({ minLength: 1, maxLength: 255 }),
   description: Type.Optional(Type.String({ maxLength: 5e3 })),
-  base_model: BaseModel,
+  base_model: LocalBaseModel,
   system_prompt: Type.String({ minLength: 1, maxLength: 1e4 }),
   guidelines: Type.Array(NonEmptyText, { minItems: 1, maxItems: 50 }),
   constraints: Type.Optional(Type.Array(NonEmptyText, { maxItems: 50 })),
@@ -652,7 +583,7 @@ function resolveAgentModel(runtime, selection) {
   const resolved = resolveAgentModelDefinition(runtime, selection);
   if (!runtime.hasConfiguredAuth(selection.provider)) {
     throw new Error(
-      `Authenticate Pi provider "${selection.provider}" first (for example with Pi's /login), then retry. Provider secrets are never accepted by tt flags.`
+      `Authenticate provider "${selection.provider}" first, then retry. Provider secrets are never accepted by tt flags.`
     );
   }
   return resolved;
@@ -663,7 +594,7 @@ function resolveAgentModelDefinition(runtime, selection) {
   );
   if (!provider) {
     throw new Error(
-      `Unknown Pi provider "${selection.provider}". Run \`tt agent models --all\` to list providers and models.`
+      `Unknown provider "${selection.provider}". Run \`tt agent models --all\` to list providers and models.`
     );
   }
   const model = runtime.getModel(selection.provider, selection.model);
@@ -721,6 +652,83 @@ function ensureOpenRouterAppHeaders(modelsPath) {
 import { randomUUID } from "crypto";
 import { Type as Type2 } from "typebox";
 
+// src/base-models.ts
+var SUPPORTED_BASE_MODELS = [
+  "google/gemma-4-E2B-it",
+  "google/gemma-4-E4B-it",
+  "Qwen/Qwen3.5-2B",
+  "Qwen/Qwen3-VL-2B-Instruct",
+  "Qwen/Qwen3.5-4B",
+  "nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
+  "meta-models/Muse-Glimmer-30B",
+  "meta-llama/Llama-3.2-3B-Instruct",
+  "microsoft/Phi-4-mini-instruct",
+  "ibm-granite/granite-3.3-2b-instruct",
+  "bigcode/starcoder2-3b"
+];
+var BASE_MODEL_ALIASES = /* @__PURE__ */ new Map();
+for (const model of SUPPORTED_BASE_MODELS) {
+  BASE_MODEL_ALIASES.set(model.toLowerCase(), model);
+}
+for (const [alias, model] of [
+  ["google/gemma-4-E2B", "google/gemma-4-E2B-it"],
+  ["google/gemma-4-e2b", "google/gemma-4-E2B-it"],
+  ["google/gemma-4-e2b-it", "google/gemma-4-E2B-it"],
+  ["google/gemma-4-E4B", "google/gemma-4-E4B-it"],
+  ["google/gemma-4-e4b", "google/gemma-4-E4B-it"],
+  ["google/gemma-4-e4b-it", "google/gemma-4-E4B-it"],
+  ["qwen/qwen3.5-2b", "Qwen/Qwen3.5-2B"],
+  ["Qwen/Qwen3.5-2B-Base", "Qwen/Qwen3.5-2B"],
+  ["qwen/qwen3.5-2b-base", "Qwen/Qwen3.5-2B"],
+  ["qwen/qwen3-vl-2b-instruct", "Qwen/Qwen3-VL-2B-Instruct"],
+  ["Qwen/Qwen3-VL-2B", "Qwen/Qwen3-VL-2B-Instruct"],
+  ["qwen/qwen3-vl-2b", "Qwen/Qwen3-VL-2B-Instruct"],
+  ["qwen/qwen3.5-4b", "Qwen/Qwen3.5-4B"],
+  ["Qwen/Qwen3.5-4B-Base", "Qwen/Qwen3.5-4B"],
+  ["qwen/qwen3.5-4b-base", "Qwen/Qwen3.5-4B"],
+  [
+    "nvidia/nvidia-nemotron-3.5-lightning-30b-a3b-bf16",
+    "nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16"
+  ],
+  ["meta-models/muse-glimmer-30b", "meta-models/Muse-Glimmer-30B"],
+  ["meta-llama/llama-3.2-3b-instruct", "meta-llama/Llama-3.2-3B-Instruct"],
+  ["meta-llama/Llama-3.2-3B", "meta-llama/Llama-3.2-3B-Instruct"],
+  ["meta-llama/llama-3.2-3b", "meta-llama/Llama-3.2-3B-Instruct"],
+  ["microsoft/phi-4-mini-instruct", "microsoft/Phi-4-mini-instruct"],
+  ["phi-4-mini-instruct", "microsoft/Phi-4-mini-instruct"],
+  ["ibm-granite/granite-3.3-2b-instruct", "ibm-granite/granite-3.3-2b-instruct"],
+  ["granite-3.3-2b-instruct", "ibm-granite/granite-3.3-2b-instruct"],
+  ["bigcode/starcoder2-3b", "bigcode/starcoder2-3b"],
+  ["starcoder2-3b", "bigcode/starcoder2-3b"]
+]) {
+  BASE_MODEL_ALIASES.set(alias.toLowerCase(), model);
+}
+var UnsupportedBaseModelError = class extends Error {
+  constructor(model) {
+    const supported = SUPPORTED_BASE_MODELS.join(", ");
+    super(
+      `Unsupported base_model "${String(model)}". Supported base models: ${supported}`
+    );
+    this.name = "UnsupportedBaseModelError";
+  }
+};
+function canonicalizeBaseModel(model) {
+  if (typeof model !== "string" || !model.trim()) {
+    throw new UnsupportedBaseModelError(model);
+  }
+  const canonical = BASE_MODEL_ALIASES.get(model.trim().toLowerCase());
+  if (!canonical) {
+    throw new UnsupportedBaseModelError(model);
+  }
+  return canonical;
+}
+function canonicalizeSpecBaseModel(body) {
+  if ("base_model" in body && body.base_model !== void 0) {
+    return { ...body, base_model: canonicalizeBaseModel(body.base_model) };
+  }
+  return body;
+}
+
 // src/pipeline.ts
 import {
   canonicalPipeline as canonicalPortablePipeline,
@@ -837,7 +845,7 @@ var RunBody = Type2.Object({
   }, { additionalProperties: false })),
   split_ratios: Type2.Optional(SplitRatios)
 }, { additionalProperties: false });
-var BaseModel2 = Type2.Union(
+var BaseModel = Type2.Union(
   SUPPORTED_BASE_MODELS.map((model) => Type2.Literal(model))
 );
 var PipelineDocument = Type2.Object({
@@ -864,7 +872,7 @@ var CreateSpecBody = Type2.Object({
   guidelines: Type2.Optional(Guidelines),
   examples: Examples,
   constraints: Type2.Optional(Guidelines),
-  base_model: Type2.Optional(BaseModel2)
+  base_model: Type2.Optional(BaseModel)
 }, { additionalProperties: false });
 var UpdateSpecBody = Type2.Object({
   name: Type2.Optional(Type2.String({ minLength: 1, maxLength: 255 })),
@@ -873,7 +881,7 @@ var UpdateSpecBody = Type2.Object({
   guidelines: Type2.Optional(Guidelines),
   examples: Type2.Optional(Examples),
   constraints: Type2.Optional(Guidelines),
-  base_model: Type2.Optional(BaseModel2)
+  base_model: Type2.Optional(BaseModel)
 }, { additionalProperties: false });
 function record2(value) {
   return value !== null && typeof value === "object" && !Array.isArray(value) ? value : void 0;
@@ -1119,7 +1127,7 @@ You have no general filesystem tools. The only workspace-scoped local spec capab
 }
 function defaultAgent(options) {
   if (!options.streamSimple) {
-    throw new Error("The configured Pi model runtime cannot stream this model.");
+    throw new Error("The configured model runtime cannot stream this model.");
   }
   let calls = 0;
   return new Agent({
@@ -1562,6 +1570,10 @@ function readConfig() {
     return {};
   }
 }
+var configRevision = 0;
+function getConfigRevision() {
+  return configRevision;
+}
 function writeConfig(config) {
   const dir = getConfigDir();
   if (!existsSync2(dir)) mkdirSync2(dir, { recursive: true, mode: 448 });
@@ -1571,6 +1583,7 @@ function writeConfig(config) {
     mode: 384
   });
   chmodSync(path, 384);
+  configRevision += 1;
 }
 function updateConfig(partial) {
   const current = readConfig();
@@ -1762,9 +1775,7 @@ var SLASH_COMMANDS = [
   { path: "/help", description: "Show commands; add a word to filter." },
   { path: "/status", description: "Show lightweight workflow status." },
   { path: "/context", description: "Show the current project and backend context." },
-  { path: "/mode cloud", description: "Use the managed cloud workflow." },
-  { path: "/mode local", description: "Use the local GPU workflow." },
-  { path: "/model", description: "Show the active model; /model <id> activates a local model." },
+  { path: "/model", description: "Show or change the TT agent model." },
   { path: "/cd", description: "Change the shell's working directory." },
   { path: "/clear", description: "Clear the terminal." },
   { path: "/exit", description: "Exit the TT shell." }
@@ -1874,6 +1885,14 @@ function cloudConfigPath(env, homeDirectory) {
   const configHome = env.XDG_CONFIG_HOME ? resolve(env.XDG_CONFIG_HOME) : join5(homeDirectory, ".config");
   return join5(configHome, "tuned-tensor", "config.json");
 }
+function agentSelectionFrom(env, configAgent) {
+  const stored = configAgent && typeof configAgent === "object" && !Array.isArray(configAgent) ? configAgent : {};
+  const provider = env.TUNED_TENSOR_AGENT_PROVIDER?.trim() || stringField(stored, "provider");
+  const model = env.TUNED_TENSOR_AGENT_MODEL?.trim() || stringField(stored, "model");
+  const thinking = env.TUNED_TENSOR_AGENT_THINKING?.trim() || stringField(stored, "thinking");
+  if (!provider && !model && !thinking) return void 0;
+  return { provider, model, thinking };
+}
 function redactApiKey(key) {
   if (!key) return void 0;
   if (key.length <= 4) return "\u2026";
@@ -1971,12 +1990,13 @@ async function discoverShellContext(options = {}) {
   const storedApiKey = stringField(cloudConfigJson.value, "api_key");
   const apiKey = environmentApiKey || storedApiKey;
   const baseUrl = env.TUNED_TENSOR_URL?.trim() || stringField(cloudConfigJson.value, "base_url") || "https://tunedtensor.com";
+  const agent = agentSelectionFrom(env, cloudConfigJson.value?.agent);
   const environmentTarget = targetFromEnvironment(env);
   if (env.TT_TARGET && !environmentTarget) {
     warnings.push(`Ignoring invalid TT_TARGET=${JSON.stringify(env.TT_TARGET)}; use cloud or local.`);
   }
-  const inferredTarget = environmentTarget ?? (hasAdjacentLocalConfig ? "local" : "cloud");
-  const targetSource = environmentTarget ? "environment" : hasAdjacentLocalConfig ? "adjacent-config" : "default-cloud";
+  const inferredTarget = environmentTarget ?? "local";
+  const targetSource = environmentTarget ? "environment" : "default-local";
   const [activeModelId, latestRun] = await Promise.all([
     readActiveModelId(storeRoot),
     readLatestRun(storeRoot)
@@ -2001,25 +2021,30 @@ async function discoverShellContext(options = {}) {
       activeModelId,
       latestRun
     },
+    agent,
     warnings
   };
 }
 function valueOrDash(value) {
   return value === void 0 || value === "" ? "\u2014" : String(value);
 }
+function agentLabel(context) {
+  const agent = context.agent;
+  if (!agent?.provider && !agent?.model) return "not configured";
+  const id = [agent.provider, agent.model].filter(Boolean).join("/");
+  return agent.thinking ? `${id} (thinking ${agent.thinking})` : id;
+}
 function targetSourceLabel(source) {
   switch (source) {
     case "environment":
       return "TT_TARGET";
-    case "adjacent-config":
-      return "adjacent local-runner.json";
     default:
       return "default";
   }
 }
 function formatShellContext(context, selectedTarget, targetSource) {
   const specLabel = context.spec ? context.spec.parseError ? `${context.spec.path} (invalid JSON)` : context.spec.path : "not found";
-  const sourceLabel = targetSource === "session" ? "session" : targetSource === "command-option" ? "--target" : targetSourceLabel(targetSource);
+  const sourceLabel = targetSource === "command-option" ? "--target" : targetSourceLabel(targetSource);
   const lines = [
     `Target         ${selectedTarget} (${sourceLabel})`,
     `Project        ${context.projectName}`,
@@ -2028,6 +2053,7 @@ function formatShellContext(context, selectedTarget, targetSource) {
     `Spec ID        ${valueOrDash(context.spec?.id)}`,
     `Base model     ${valueOrDash(context.spec?.baseModel)}`,
     `Examples       ${valueOrDash(context.spec?.exampleCount)}`,
+    `Agent model    ${agentLabel(context)}`,
     `Cloud endpoint ${context.cloud.baseUrl}`,
     `Cloud auth     ${context.cloud.authenticated ? `yes (${context.cloud.keyPrefix})` : "no"}`,
     `Local config   ${context.local.configPath ?? "not found"}`,
@@ -2040,7 +2066,8 @@ function formatShellContext(context, selectedTarget, targetSource) {
 function formatShellStatus(context, selectedTarget) {
   const lines = [
     `Workflow       ${selectedTarget}`,
-    `Spec           ${context.spec?.name ?? (context.spec ? "unnamed" : "not found")}`
+    `Spec           ${context.spec?.name ?? (context.spec ? "unnamed" : "not found")}`,
+    `Agent model    ${agentLabel(context)}`
   ];
   if (selectedTarget === "cloud") {
     lines.push(
@@ -2059,6 +2086,103 @@ function formatShellStatus(context, selectedTarget) {
   return lines;
 }
 
+// src/agent-control.ts
+function modelKey(model) {
+  return `${model.provider}/${model.id}`;
+}
+function modelRelevance(model, query) {
+  const q = query.trim().toLowerCase();
+  if (!q) return void 0;
+  const fullId = modelKey(model).toLowerCase();
+  const bareId = model.id.toLowerCase();
+  const name = model.name.toLowerCase();
+  if (fullId === q || bareId === q) return 0;
+  if (fullId.startsWith(q)) return 1;
+  if (bareId.startsWith(q)) return 2;
+  if (fullId.includes(q)) return 3;
+  if (name.includes(q)) return 4;
+  return void 0;
+}
+function listAgentModels(runtime, options = {}) {
+  let models = runtime.getModels(options.provider).filter(
+    (model) => options.includeUnauthenticated || runtime.hasConfiguredAuth(model.provider)
+  ).map((model) => ({
+    provider: model.provider,
+    id: model.id,
+    name: model.name ?? model.id,
+    authenticated: runtime.hasConfiguredAuth(model.provider),
+    thinking: model.reasoning !== false
+  }));
+  const query = options.query?.trim().toLowerCase();
+  if (query) {
+    const scored = [];
+    for (const model of models) {
+      const score = modelRelevance(model, query);
+      if (score !== void 0) scored.push({ model, score });
+    }
+    scored.sort(
+      (left, right) => left.score - right.score || modelKey(left.model).localeCompare(modelKey(right.model))
+    );
+    models = scored.map((entry) => entry.model);
+  } else {
+    models.sort((left, right) => modelKey(left).localeCompare(modelKey(right)));
+  }
+  if (options.limit !== void 0 && options.limit > 0) {
+    models = models.slice(0, options.limit);
+  }
+  return models;
+}
+function describeAgentModel(runtime, env) {
+  const selection = getAgentSelection(env);
+  if (!selection) return void 0;
+  const resolved = resolveAgentModelDefinition(runtime, selection);
+  const providerName = runtime.getProviders().find((provider) => provider.id === selection.provider)?.name;
+  return {
+    provider: selection.provider,
+    providerName,
+    model: selection.model,
+    modelName: resolved.model.name,
+    thinking: selection.thinking,
+    authenticated: runtime.hasConfiguredAuth(selection.provider),
+    supportsThinking: resolved.model.reasoning !== false
+  };
+}
+function setAgentModel(runtime, env, provider, model, options = {}) {
+  const current = getAgentSelection(env);
+  const requestedThinking = options.thinking ?? current?.thinking ?? "medium";
+  if (!AGENT_THINKING_LEVELS.includes(requestedThinking)) {
+    throw new Error(
+      `Agent thinking must be one of: ${AGENT_THINKING_LEVELS.join(", ")}`
+    );
+  }
+  const adjust = options.adjustThinking ?? true;
+  if (!adjust) {
+    const selection2 = {
+      provider,
+      model,
+      thinking: requestedThinking
+    };
+    resolveAgentModel(runtime, selection2);
+    updateConfig({ agent: selection2 });
+    return { selection: selection2, adjustedThinking: false };
+  }
+  const definition = resolveAgentModelDefinition(runtime, {
+    provider,
+    model,
+    thinking: "off"
+  });
+  const supportsThinking = definition.model.reasoning !== false;
+  const effectiveThinking = supportsThinking ? requestedThinking : "off";
+  const selection = {
+    provider,
+    model,
+    thinking: effectiveThinking
+  };
+  resolveAgentModel(runtime, selection);
+  updateConfig({ agent: selection });
+  return { selection, adjustedThinking: effectiveThinking !== requestedThinking };
+}
+
 // src/shell.ts
 var AGENT_SLASH_COMMANDS = /* @__PURE__ */ new Set([
   "new",
@@ -2189,7 +2313,7 @@ function routeShellCommand(input, activeMode) {
   const override = args2[0];
   if (override === "cloud" || override === "local") {
     if (args2.length === 1) {
-      throw new ShellParseError(`Add a ${override} command, or use \`/mode ${override}\` to switch workflows.`);
+      throw new ShellParseError(`Add a ${override} command, e.g. \`${override} runs list\`.`);
     }
     return { target: override, args: args2.slice(1) };
   }
@@ -2218,7 +2342,6 @@ var SLASH_NAMES = /* @__PURE__ */ new Set([
   "help",
   "status",
   "context",
-  "mode",
   "model",
   "clear",
   "cd",
@@ -2380,7 +2503,12 @@ function helpText(mode, query, palette = false) {
 `;
 }
 function activeModelLabel(snapshot) {
-  return snapshot.mode === "local" ? snapshot.context.local.activeModelId ?? "base" : snapshot.context.spec?.baseModel ?? "\u2014";
+  return snapshot.context.local.activeModelId ?? "base";
+}
+function agentModelLabel(context) {
+  const agent = context.agent;
+  if (!agent?.provider && !agent?.model) return "not configured";
+  return [agent.provider, agent.model].filter(Boolean).join("/") || "not configured";
 }
 function renderShellBanner(snapshot) {
   const spec = snapshot.context.spec?.name ?? snapshot.context.spec?.path ?? "no spec";
@@ -2388,8 +2516,11 @@ function renderShellBanner(snapshot) {
   const lines = [
     heading,
     `${accent(snapshot.mode)}${chalk.dim(
-      ` \xB7 ${snapshot.context.projectName} \xB7 ${spec} \xB7 model ${activeModelLabel(snapshot)}`
+      ` \xB7 ${snapshot.context.projectName} \xB7 ${spec}`
     )}`,
+    chalk.dim(
+      `agent ${agentModelLabel(snapshot.context)} \xB7 workflow model ${activeModelLabel(snapshot)}`
+    ),
     chalk.dim("ctrl+c stop/clear \xB7 ctrl+d exit \xB7 /help commands \xB7 tab complete"),
     "",
     chalk.dim("Ask TT anything. Known commands run directly.")
@@ -2427,20 +2558,19 @@ function resetShellPromptStyle() {
   return "";
 }
 var TunedTensorShellSession = class _TunedTensorShellSession {
-  constructor(runner, io, agent, env, contextProvider, version, initialContext) {
+  constructor(runner, io, agent, env, contextProvider, version, agentModelRuntime, initialContext) {
     this.runner = runner;
     this.io = io;
     this.agent = agent;
     this.env = env;
     this.contextProvider = contextProvider;
     this.version = version;
-    this.mode = initialContext.inferredTarget;
-    this.modeSource = initialContext.targetSource;
+    this.agentModelRuntime = agentModelRuntime;
     this.cwd = initialContext.cwd;
     this.context = initialContext;
   }
-  mode;
-  modeSource;
+  mode = "local";
+  modeSource = "default-local";
   cwd;
   context;
   static async create(options) {
@@ -2455,6 +2585,7 @@ var TunedTensorShellSession = class _TunedTensorShellSession {
       env,
       contextProvider,
       options.version,
+      options.agentModelRuntime,
       initialContext
     );
   }
@@ -2479,26 +2610,67 @@ var TunedTensorShellSession = class _TunedTensorShellSession {
   async refreshContext() {
     this.context = await this.contextProvider({ cwd: this.cwd, env: this.env });
     this.cwd = this.context.cwd;
-    if (this.modeSource !== "session" && this.modeSource !== "environment") {
-      this.mode = this.context.inferredTarget;
-      this.modeSource = this.context.targetSource;
-    }
   }
   writeLines(lines) {
     this.io.write(`${lines.join("\n")}
 `);
   }
-  modelLines() {
-    if (this.mode === "local") {
-      return styleDetailLines([
-        detailLine("Active model", this.context.local.activeModelId ?? "base"),
-        detailLine("Change", "/model <id> to activate a verified local model")
-      ]);
+  async requireAgentModelRuntime() {
+    if (!this.agentModelRuntime) {
+      throw new ShellParseError("The local agent model runtime is unavailable.");
     }
-    return styleDetailLines([
-      detailLine("Base model", this.context.spec?.baseModel ?? "\u2014"),
-      detailLine("Change", "edit base_model in tunedtensor.json, then run push")
+    return await this.agentModelRuntime();
+  }
+  renderAgentModel(runtime, query) {
+    let summary;
+    try {
+      summary = describeAgentModel(runtime, this.env);
+    } catch {
+      summary = void 0;
+    }
+    const label = summary ? `${summary.provider}/${summary.model}${summary.modelName && summary.modelName !== summary.model ? ` (${summary.modelName})` : ""} \u2014 thinking ${summary.thinking}, auth ${summary.authenticated ? "configured" : "required"}` : "not configured";
+    const lines = styleDetailLines([
+      detailLine("Agent model", label),
+      detailLine("Change", "/model <provider>/<model>"),
+      detailLine("Search", "/model <query>")
     ]);
+    const all = listAgentModels(runtime);
+    if (all.length === 0) {
+      lines.push("", chalk.dim("  no authenticated models \u2014 run `tt agent models --all` for the full catalog"));
+      return lines;
+    }
+    if (query) {
+      const matches = listAgentModels(runtime, { query });
+      lines.push("", chalk.bold(`Models matching ${JSON.stringify(query)}`));
+      if (matches.length === 0) {
+        lines.push(chalk.dim("  no matches"));
+      } else {
+        const shown2 = matches.slice(0, 10);
+        for (const model of shown2) {
+          lines.push(this.formatAgentModelChoice(model, summary));
+        }
+        if (matches.length > shown2.length) {
+          lines.push(chalk.dim(`  \u2026 ${matches.length - shown2.length} more \u2014 refine the query`));
+        }
+      }
+      return lines;
+    }
+    const preferred = summary ? all.filter((model) => model.provider === summary.provider) : [];
+    const rest = summary ? all.filter((model) => model.provider !== summary.provider) : all;
+    const shown = [...preferred, ...rest].slice(0, 8);
+    lines.push("", chalk.bold(summary ? "Suggestions" : "Available models"));
+    for (const model of shown) {
+      lines.push(this.formatAgentModelChoice(model, summary));
+    }
+    if (all.length > shown.length) {
+      lines.push(chalk.dim(`  \u2026 ${all.length - shown.length} more \u2014 /model <query> to search`));
+    }
+    return lines;
+  }
+  formatAgentModelChoice(model, summary) {
+    const active = summary?.provider === model.provider && summary?.model === model.id;
+    const displayName = model.name !== model.id ? `  ${model.name}` : "";
+    return `  ${accent(`${model.provider}/${model.id}`)}${displayName}${active ? chalk.dim(" (active)") : ""}`;
   }
   async handleSlash(command) {
     switch (command.name) {
@@ -2520,41 +2692,37 @@ var TunedTensorShellSession = class _TunedTensorShellSession {
           formatShellContext(this.context, this.mode, this.modeSource)
         ));
         return "continue";
-      case "mode": {
-        if (command.args.length === 0) {
-          this.io.write(`Current workflow: ${this.mode}
-`);
-          return "continue";
-        }
-        if (command.args.length !== 1 || !["cloud", "local"].includes(command.args[0])) {
-          throw new ShellParseError("Usage: /mode cloud|local");
-        }
-        this.mode = command.args[0];
-        this.modeSource = "session";
-        this.io.write(`${successMark()} Workflow switched to ${this.mode}.
-`);
-        return "continue";
-      }
       case "model": {
+        const runtime = await this.requireAgentModelRuntime();
         if (command.args.length === 0) {
           await this.refreshContext();
-          this.writeLines(this.modelLines());
+          this.writeLines(this.renderAgentModel(runtime));
           return "continue";
         }
         if (command.args.length !== 1) {
-          throw new ShellParseError("Usage: /model [model-id]");
+          throw new ShellParseError("Usage: /model [<query> | <provider>/<model>]");
+        }
+        const target = command.args[0];
+        const slash = target.indexOf("/");
+        if (slash <= 0) {
+          await this.refreshContext();
+          this.writeLines(this.renderAgentModel(runtime, target));
+          return "continue";
         }
-        if (this.mode !== "local") {
+        if (slash === target.length - 1) {
           throw new ShellParseError(
-            "Activating models is a local workflow action. Use /mode local first; cloud specs change base_model in tunedtensor.json."
+            "Usage: /model <provider>/<model>, e.g. /model anthropic/claude-sonnet-4-5"
           );
         }
-        await this.runner({
-          target: "local",
-          args: ["models", "activate", command.args[0]],
-          cwd: this.cwd
-        });
+        const provider = target.slice(0, slash);
+        const model = target.slice(slash + 1);
+        const result2 = setAgentModel(runtime, this.env, provider, model);
         await this.refreshContext();
+        const note = result2.adjustedThinking ? " \u2014 thinking set to off for this model" : "";
+        this.io.write(
+          `${successMark()} Agent model: ${result2.selection.provider}/${result2.selection.model} (thinking ${result2.selection.thinking})${note}.
+`
+        );
         return "continue";
       }
       case "clear":
@@ -2694,7 +2862,8 @@ async function startInteractiveShell(options) {
     agent: options.agent,
     cwd: options.cwd,
     env: options.env,
-    version: options.version
+    version: options.version,
+    agentModelRuntime: options.agentModelRuntime
   });
   const completer = createCommandCompleter(() => session.snapshot().mode);
   readline = createInterface({
@@ -8064,55 +8233,47 @@ function registerPushCommand(parent) {
 }
 
 // src/commands/agent.ts
-function selectionOrHint(env) {
-  const selection = getAgentSelection(env);
-  if (!selection) {
-    throw new Error(
-      "The local agent is not configured. Run `tt agent configure --provider <provider> --model <model>`."
-    );
-  }
-  return selection;
-}
 function registerAgentCommands(parent, options) {
-  const agent = parent.command("agent").description("Configure the laptop-local Pi agent");
+  const agent = parent.command("agent").description("Configure the laptop-local TT agent");
   agent.command("status").description("Show the selected local provider, model, thinking, and auth state").action(async () => {
-    const selection = selectionOrHint(options.env);
     const runtime = await options.getRuntime();
-    const resolved = resolveAgentModelDefinition(runtime, selection);
-    const authenticated = runtime.hasConfiguredAuth(selection.provider);
+    const summary = describeAgentModel(runtime, options.env);
+    if (!summary) {
+      throw new Error(
+        "The local agent is not configured. Run `tt agent configure --provider <provider> --model <model>`."
+      );
+    }
     const value = {
       execution: "local",
-      provider: selection.provider,
-      model: selection.model,
-      thinking: selection.thinking,
-      authenticated,
-      supports_thinking: resolved.model.reasoning !== false
+      provider: summary.provider,
+      provider_name: summary.providerName,
+      model: summary.model,
+      model_name: summary.modelName,
+      thinking: summary.thinking,
+      authenticated: summary.authenticated,
+      supports_thinking: summary.supportsThinking
     };
+    const modelLabel = summary.modelName && summary.modelName !== summary.model ? `${summary.modelName} (${summary.provider}/${summary.model})` : `${summary.provider}/${summary.model}`;
     options.output.write(isJsonMode() ? `${JSON.stringify(value, null, 2)}
-` : `Local agent: ${value.provider}/${value.model} (thinking: ${value.thinking}, auth: ${authenticated ? "configured" : "required"})
+` : `Local agent: ${modelLabel} (thinking: ${value.thinking}, auth: ${value.authenticated ? "configured" : "required"})
 `);
   });
-  agent.command("models").description("List Pi provider models available to the local agent").option("--provider <provider>", "Filter by Pi provider ID").option("--all", "Include models whose provider auth is not configured").action(async (commandOptions) => {
+  agent.command("models").description("List provider models available to the local agent").option("--provider <provider>", "Filter by provider ID").option("--all", "Include models whose provider auth is not configured").action(async (commandOptions) => {
     const runtime = await options.getRuntime();
     if (commandOptions.provider && !runtime.getProviders().some((provider) => provider.id === commandOptions.provider)) {
-      throw new Error(`Unknown Pi provider "${commandOptions.provider}".`);
-    }
-    const models = runtime.getModels(commandOptions.provider).filter(
-      (model) => commandOptions.all || runtime.hasConfiguredAuth(model.provider)
-    ).map((model) => ({
-      provider: model.provider,
-      id: model.id,
-      name: model.name ?? model.id,
-      authenticated: runtime.hasConfiguredAuth(model.provider),
-      thinking: model.reasoning !== false
-    }));
+      throw new Error(`Unknown provider "${commandOptions.provider}".`);
+    }
+    const models = listAgentModels(runtime, {
+      provider: commandOptions.provider,
+      includeUnauthenticated: commandOptions.all
+    });
     if (isJsonMode()) {
       options.output.write(`${JSON.stringify({ data: models }, null, 2)}
 `);
       return;
     }
     if (models.length === 0) {
-      options.output.write("No matching authenticated Pi models. Use --all to inspect the full catalog.\n");
+      options.output.write("No matching authenticated models. Use --all to inspect the full catalog.\n");
       return;
     }
     for (const model of models) {
@@ -8120,20 +8281,17 @@ function registerAgentCommands(parent, options) {
 `);
     }
   });
-  agent.command("configure").description("Select the local Pi provider, model, and thinking level").requiredOption("--provider <provider>", "Pi provider ID").requiredOption("--model <model>", "Pi model ID").option("--thinking <level>", "off, minimal, low, medium, high, xhigh, or max", "medium").action(async (commandOptions) => {
-    if (!AGENT_THINKING_LEVELS.includes(commandOptions.thinking)) {
-      throw new Error(`--thinking must be one of: ${AGENT_THINKING_LEVELS.join(", ")}`);
-    }
-    const selection = {
-      provider: commandOptions.provider,
-      model: commandOptions.model,
-      thinking: commandOptions.thinking
-    };
-    resolveAgentModel(await options.getRuntime(), selection);
-    updateConfig({ agent: selection });
-    const value = { execution: "local", ...selection };
+  agent.command("configure").description("Select the local provider, model, and thinking level").requiredOption("--provider <provider>", "Provider ID").requiredOption("--model <model>", "Model ID").option("--thinking <level>", "off, minimal, low, medium, high, xhigh, or max", "medium").action(async (commandOptions) => {
+    const result2 = setAgentModel(
+      await options.getRuntime(),
+      options.env,
+      commandOptions.provider,
+      commandOptions.model,
+      { thinking: commandOptions.thinking, adjustThinking: false }
+    );
+    const value = { execution: "local", ...result2.selection };
     options.output.write(isJsonMode() ? `${JSON.stringify(value, null, 2)}
-` : `Configured local agent: ${selection.provider}/${selection.model} (thinking: ${selection.thinking}).
+` : `Configured local agent: ${result2.selection.provider}/${result2.selection.model} (thinking: ${result2.selection.thinking}).
 `);
   });
 }
@@ -8303,14 +8461,14 @@ function formatCliUpdateNotice(update) {
 }
 
 // src/cli.ts
-async function createDefaultAgentClient(runtime, env, cloud, workspaceRoot) {
+async function createDefaultAgentClient(runtime, env, cloud, workspaceRoot, getModelRuntime) {
   const selection = getAgentSelection(env);
   if (!selection) {
     throw new Error(
       "The laptop-local agent is not configured. Run `tt agent models --all`, then `tt agent configure --provider <provider> --model <model>`."
     );
   }
-  const modelRuntime = runtime.modelRuntime ?? await createPiModelRuntime();
+  const modelRuntime = await getModelRuntime();
   const clientOpts = { apiKey: cloud.apiKey, baseUrl: cloud.baseUrl };
   const toolApi = runtime.agentToolApi ?? {
     get: async (path, query) => await get(path, query, clientOpts),
@@ -8343,11 +8501,23 @@ async function createDefaultAgentClient(runtime, env, cloud, workspaceRoot) {
     createAgent: runtime.createPiAgent
   });
 }
-function createLazyDefaultAgentClient(runtime, env, cloud, workspaceRoot) {
+function createModelRuntimeGetter(runtime) {
+  const injected = runtime.modelRuntime;
+  let promise;
+  return async () => {
+    if (injected) return injected;
+    promise ??= createPiModelRuntime();
+    return await promise;
+  };
+}
+function createLazyDefaultAgentClient(runtime, env, cloud, workspaceRoot, getModelRuntime) {
   let pending;
+  let builtRevision = -1;
   const client = () => {
-    if (!pending) {
-      const attempt = createDefaultAgentClient(runtime, env, cloud, workspaceRoot);
+    const revision = getConfigRevision();
+    if (!pending || builtRevision !== revision) {
+      builtRevision = revision;
+      const attempt = createDefaultAgentClient(runtime, env, cloud, workspaceRoot, getModelRuntime);
       pending = attempt;
       attempt.catch(() => {
         if (pending === attempt) pending = void 0;
@@ -8447,13 +8617,7 @@ function createProgram(version, runtime = {}) {
   const invokeLocal = runtime.runLocalCommand ?? executeLocalCommand;
   const launchShell = runtime.startShell ?? startInteractiveShell;
   const cwd = runtime.cwd ?? process.cwd();
-  const injectedModelRuntime = runtime.modelRuntime;
-  let modelRuntimePromise;
-  const getModelRuntime = async () => {
-    if (injectedModelRuntime) return injectedModelRuntime;
-    modelRuntimePromise ??= createPiModelRuntime();
-    return await modelRuntimePromise;
-  };
+  const getModelRuntime = createModelRuntimeGetter(runtime);
   const shellRunner = async (request2) => {
     const root = program.opts();
     const args2 = request2.target === "local" ? ["local", ...request2.args] : request2.args;
@@ -8472,7 +8636,7 @@ function createProgram(version, runtime = {}) {
       client: runtime.agentClient ?? createLazyDefaultAgentClient(runtime, shellEnvironment, {
         apiKey: root.apiKey ?? shellEnvironment.TUNED_TENSOR_API_KEY,
         baseUrl: root.baseUrl ?? shellEnvironment.TUNED_TENSOR_URL
-      }, cwd),
+      }, cwd, getModelRuntime),
       output,
       error
     });
@@ -8484,7 +8648,8 @@ function createProgram(version, runtime = {}) {
       error,
       cwd,
       env: shellEnvironment,
-      version
+      version,
+      agentModelRuntime: getModelRuntime
     });
   };
   program.name("tt").description("Tuned Tensor \u2014 converse, train, and inspect from one terminal").version(version).option("-k, --api-key <key>", "API key (overrides stored key)").option(
@@ -8588,7 +8753,7 @@ function createProgram(version, runtime = {}) {
     });
     applyExitCode(result2);
   });
-  program.command("shell").description("Open the interactive cloud/local terminal").action(openShell);
+  program.command("shell").description("Open the conversational terminal (local by default)").action(openShell);
   program.addHelpText(
     "after",
     `
@@ -8645,11 +8810,12 @@ async function runCli(version, runtime = {}) {
 `);
     }
     const invokeSelf = runtime.runSelfCommand ?? runSelfCommand;
+    const getModelRuntime = createModelRuntimeGetter(runtime);
     const agent = createShellAgent({
       client: runtime.agentClient ?? createLazyDefaultAgentClient(runtime, env, {
         apiKey: env.TUNED_TENSOR_API_KEY,
         baseUrl: env.TUNED_TENSOR_URL
-      }, runtime.cwd ?? process.cwd()),
+      }, runtime.cwd ?? process.cwd(), getModelRuntime),
       output,
       error
     });
@@ -8668,7 +8834,8 @@ async function runCli(version, runtime = {}) {
       error,
       cwd: runtime.cwd ?? process.cwd(),
       env,
-      version
+      version,
+      agentModelRuntime: getModelRuntime
     });
     return;
   }
@@ -8677,7 +8844,7 @@ async function runCli(version, runtime = {}) {
 }
 
 // src/index.ts
-runCli("0.11.0").catch((err) => {
+runCli("0.12.0-beta.1").catch((err) => {
   reportError(err);
   process.exit(1);
 });