@tekmidian/pai 0.39.0 → 0.40.0

package/dist/daemon-mcp/index.mjs

@@ -1,8 +1,10 @@
 #!/usr/bin/env node
-import { t as PaiClient } from "../ipc-client-BmypMNYk.mjs";
-import { o as loadConfig } from "../config-B64vFg14.mjs";
-import { B as workersLogDir, F as ledgerSummary, J as readWorkersSection, M as loadStatuses, a as testProvider, g as sayToWorker, h as readInbox, i as runWorker, j as loadStatus, k as alive, m as handoffFromInside, t as runChain, z as ledgerPath } from "../chain-CTtHligI.mjs";
-import { a as setClass, c as unsetClass, f as psOutput, i as removeProvider, l as updateProvider, n as classTargetText, o as setProviderEnabled, p as replayOutput, r as describeProviders, s as setWorkersEnabled, t as addProvider, u as useProvider } from "../providers-FYkZjn_C.mjs";
+import { o as loadConfig } from "../config-D1G9IFpn.mjs";
+import { t as PaiClient } from "../ipc-client-C7AVpsyv.mjs";
+import { m as readWorkersSection, n as MODEL_CAPABILITIES } from "../config-Dl8lT4Lu.mjs";
+import { _ as workersLogDir, p as ledgerPath } from "../server-BOuAOj9b.mjs";
+import { D as loadStatus, O as loadStatuses, a as testProvider, d as handoffFromInside, f as readInbox, i as runWorker, j as ledgerSummary, p as sayToWorker, t as runChain, w as alive } from "../chain-DhVHVnmT.mjs";
+import { _ as updateProvider, a as addProvider, b as psOutput, c as describeProviders, d as resolveProviderName, f as setClass, g as unsetClass, h as setWorkersEnabled, i as fallbackStatusText, l as modelPrefsText, m as setProviderModel, n as fallbackOn, o as classTargetText, p as setProviderEnabled, r as fallbackStatus, s as describeModels, t as fallbackOff, u as removeProvider, v as useProvider, x as replayOutput } from "../fallback-CdXv-Np0.mjs";
 import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
 import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
 
@@ -4582,10 +4584,17 @@ pai worker run --label "short task label" --class research \\
   parked in ~/.config/pai/keys/<name>, mode 0600). update changes cost_tier
   (1 cheapest … 5 most expensive) and tags (code, vision, image-gen,
   long-context, fast, reasoning).
-- \`worker_classes\` — list/set/unset class → provider[/fast], or constraints
+- \`worker_classes\` — list/set/unset class → provider[/<capability>], or constraints
   only (max_cost_tier, require_tags) so auto-routing picks a qualifying provider.
+- \`worker_model\` — show or set a provider's model ids per capability:
+  default, fast, image (get without provider lists every provider's ids).
 - \`worker_run\` — start a worker or chain from chat; returns the id immediately.
 - \`worker_toggle\` — off makes Agent subagents run on Anthropic again.
+- \`worker_fallback\` — machine-wide switch: every NEW Claude Code process
+  (interactive, task-bus, daemon summarizer) runs on a worker provider until
+  switched back. on writes the provider env + model pin into the Claude
+  settings file (the token sits there until off), off restores it exactly,
+  status lists running sessions and the note. Subagent routing is unchanged.
 
 ### Preferences from chat (map the phrase, use the tool, never mention files)
 
@@ -4593,10 +4602,21 @@ pai worker run --label "short task label" --class research \\
   worker_classes set <class>=<provider>. One line back.
 - "prefer the flash model for simple tasks" / "cheap only for drafts" →
   worker_classes set simple|draft=<provider>/fast (or a max_cost_tier). One line.
+- "run everything on <model>" / "make <model> the fast model" / "<model> is
+  the image model" → worker_model set model=<model> (capability=fast|image for
+  the latter two). One line back.
 - "reviews should use a reasoning model" → worker_classes set review with
   require_tags ["reasoning"] (drop the pin if one exists). One line.
 - "what handles reviews" / "show the routing table" → worker_classes list
   (plus worker_status for providers). One or two lines.
+- "switch everything to <provider>" / "fallback on" (plan exhausted, every
+  new session must run on the worker provider) → worker_fallback on. Confirm
+  with one line: provider, that running sessions keep their provider until
+  restarted, and how to switch back.
+- "back to anthropic" / "fallback off" → worker_fallback off. One line: what
+  was restored.
+- "is fallback on" / "are we on <provider>" → worker_fallback status. One or
+  two lines.
 - Never tell the user to edit a config file — the tools do it.`
 };
 
@@ -5408,6 +5428,56 @@ At 200 message sessions, that is 1.3MB of context saved — equivalent to 5-10 e
 `
 };
 
+//#endregion
+//#region src/daemon-mcp/tools/worker-model.ts
+/**
+* worker_model — the MCP tool behind model selection.
+*
+* Extracted from index.ts (which starts its transport on import, so its
+* inline handlers cannot be imported by tests): this module holds only the
+* handler, index.ts registers it. The mutations themselves live in
+* src/workers/providers.ts, shared with the `pai worker model` CLI.
+*/
+const text = (s) => ({ content: [{
+	type: "text",
+	text: s
+}] });
+const error = (e) => ({
+	content: [{
+		type: "text",
+		text: e instanceof Error ? e.message : String(e)
+	}],
+	isError: true
+});
+/**
+* get: the model ids of one provider (default: the active one), or of every
+* provider when no provider is given. set: write a provider's model id for a
+* capability and report the change.
+*/
+function workerModel(args, configPath) {
+	try {
+		const action = args.action ?? "get";
+		if (action !== "get" && action !== "set") return error(/* @__PURE__ */ new Error(`unknown action "${String(action)}" (expected: get, set)`));
+		if (args.capability !== void 0 && args.slot !== void 0 && args.capability !== args.slot) return error(/* @__PURE__ */ new Error(`capability "${args.capability}" and slot "${args.slot}" disagree — pass one`));
+		const capability = args.capability ?? args.slot ?? "default";
+		const { workers } = readWorkersSection(configPath);
+		if (action === "get") {
+			if (args.provider) {
+				const name = resolveProviderName(workers, args.provider);
+				const p = workers.providers[name];
+				return p ? text(`${name}  ${modelPrefsText(p)}`) : error(/* @__PURE__ */ new Error(`no provider named "${args.provider}"`));
+			}
+			return text(describeModels(workers).join("\n"));
+		}
+		if (!args.model || !args.model.trim()) return error(/* @__PURE__ */ new Error("set needs a non-empty model id"));
+		const name = resolveProviderName(workers, args.provider);
+		const p = setProviderModel(name, capability, args.model, configPath).providers[name];
+		return text(`${name} ${capability} model: ${p?.models[capability] ?? "?"}`);
+	} catch (e) {
+		return error(e);
+	}
+}
+
 //#endregion
 //#region src/daemon-mcp/index.ts
 /**
@@ -5814,7 +5884,7 @@ async function startShim() {
 		upstream_url: string().optional().describe("Chat Completions base URL (add, protocol=openai)."),
 		protocol: _enum(["anthropic", "openai"]).optional().describe("Protocol (add). Default: anthropic."),
 		engine: _enum(["claude", "codex"]).optional().describe("Runner engine (add). Default: claude."),
-		context_window: number().int().positive().optional().describe("Context window for the meter (add). Default: 200000."),
+		context_window: number().int().positive().optional().describe("Context window for the meter (add). No default: unset hides the meter unless the init event announces one."),
 		key_file: string().optional().describe("File holding the API token, 0600 (add)."),
 		key: string().optional().describe("Raw API token (add) — parked in ~/.config/pai/keys/<name>."),
 		model: string().optional().describe("Default model id (add)."),
@@ -5888,6 +5958,44 @@ async function startShim() {
 			return workerError(e);
 		}
 	});
+	server.tool("worker_fallback", [
+		"Machine-wide fallback: switch every NEW Claude Code process on this",
+		"machine (interactive, task-bus, daemon summarizer) to a worker provider,",
+		"or back to the Anthropic login.",
+		"",
+		"action=on switches (provider optional, default: active) by writing the",
+		"provider's env into ~/.claude/settings.json — the token is read from its",
+		"key file and then sits in settings.json until off. action=off restores",
+		"settings.json exactly. action=status (default) shows on/off, provider,",
+		"running sessions and the FALLBACK-ACTIVE.md note. Running sessions keep",
+		"their provider until restarted. CLAUDE_SETTINGS_PATH retargets settings",
+		"for dry runs. Subagent routing is unchanged while on."
+	].join("\n"), {
+		action: _enum([
+			"on",
+			"off",
+			"status"
+		]).optional().describe("Default: status."),
+		provider: string().optional().describe("Provider to switch to (on). Default: active.")
+	}, async (args) => {
+		try {
+			const action = args.action ?? "status";
+			if (action === "status") return workerText(fallbackStatusText(fallbackStatus()).join("\n"));
+			if (action === "on") {
+				const r = fallbackOn(args.provider);
+				return workerText([
+					r.alreadyOn ? `fallback already on — provider ${r.provider}, env re-applied` : `fallback on — provider ${r.provider}; every new Claude Code process uses it`,
+					`settings.json env: ${r.envKeys.join(", ")}`,
+					`model pin: ${r.model}`,
+					"running sessions keep their current provider until restarted"
+				].join("\n"));
+			}
+			const r = fallbackOff();
+			return workerText(`fallback off — provider ${r.provider} released, settings.json restored (${r.envKeys.join(", ")})`);
+		} catch (e) {
+			return workerError(e);
+		}
+	});
 	server.tool("worker_classes", [
 		"Manage worker classes (list, set, unset).",
 		"",
@@ -5939,6 +6047,19 @@ async function startShim() {
 			return workerError(e);
 		}
 	});
+	server.tool("worker_model", [
+		"Show or set a provider's model ids (the config behind `pai worker model`).",
+		"",
+		"action=get (default): the model ids of one provider (provider given,",
+		"default: the active one) or of every provider when provider is omitted.",
+		"action=set needs model; capability picks which one (default, fast, image)."
+	].join("\n"), {
+		action: _enum(["get", "set"]).optional().describe("Default: get."),
+		provider: string().optional().describe("Provider to read or change (default: the active one)."),
+		capability: _enum(MODEL_CAPABILITIES).optional().describe("Which model capability to set (default, fast, image). Default: default."),
+		slot: _enum(["default", "fast"]).optional().describe("Deprecated pre-capability alias of `capability`."),
+		model: string().optional().describe("The new model id (required for set).")
+	}, async (args) => workerModel(args));
 	server.tool("worker_run", [
 		"Start a worker (or a chain) and return its id immediately.",
 		"",