@aliou/pi-neuralwatt 0.2.0 → 0.4.1

package/README.md

@@ -1,4 +1,4 @@
-![banner](https://assets.aliou.me/pi-extensions/banners/pi-neuralwatt.png)
+![banner](https://assets.aliou.me/github/aliou/pi-neuralwatt/banner.png)
 
 # Pi Neuralwatt Extension
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aliou/pi-neuralwatt",
-  "version": "0.2.0",
+  "version": "0.4.1",
   "license": "MIT",
   "type": "module",
   "private": false,
@@ -20,7 +20,7 @@
       "./src/extensions/quota-warnings/index.ts",
       "./src/extensions/sub-bar-integration/index.ts"
     ],
-    "video": "https://assets.aliou.me/pi-extensions/demos/pi-neuralwatt.mp4"
+    "video": "https://assets.aliou.me/github/aliou/pi-neuralwatt/demo.mp4"
   },
   "publishConfig": {
     "access": "public"
@@ -31,20 +31,20 @@
     "README.md"
   ],
   "dependencies": {
-    "@aliou/pi-utils-settings": "^0.9.0",
-    "@aliou/pi-utils-ui": "^0.1.0"
+    "@aliou/pi-utils-settings": "^0.15.0",
+    "@aliou/pi-utils-ui": "^0.4.0"
   },
   "peerDependencies": {
-    "@mariozechner/pi-coding-agent": ">=0.67.68",
-    "@mariozechner/pi-tui": ">=0.67.68",
+    "@earendil-works/pi-coding-agent": "0.74.0",
+    "@earendil-works/pi-tui": "0.74.0",
     "@sinclair/typebox": ">=0.34.0"
   },
   "devDependencies": {
     "@aliou/biome-plugins": "^0.8.1",
     "@biomejs/biome": "^2.4.12",
     "@changesets/cli": "^2.27.11",
-    "@mariozechner/pi-coding-agent": "0.67.68",
-    "@mariozechner/pi-tui": "0.67.68",
+    "@earendil-works/pi-coding-agent": "0.74.0",
+    "@earendil-works/pi-tui": "0.74.0",
     "@types/node": "^25.0.10",
     "husky": "^9.1.7",
     "ts-json-schema-generator": "^2.4.0",
@@ -52,10 +52,10 @@
     "vitest": "^4.0.18"
   },
   "peerDependenciesMeta": {
-    "@mariozechner/pi-coding-agent": {
+    "@earendil-works/pi-coding-agent": {
       "optional": true
     },
-    "@mariozechner/pi-tui": {
+    "@earendil-works/pi-tui": {
       "optional": true
     },
     "@sinclair/typebox": {

package/src/config.ts

@@ -3,8 +3,8 @@ import {
   registerSettingsCommand,
   type SettingsSection,
 } from "@aliou/pi-utils-settings";
-import type { ExtensionAPI } from "@mariozechner/pi-coding-agent";
-import type { SettingItem } from "@mariozechner/pi-tui";
+import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
+import type { SettingItem } from "@earendil-works/pi-tui";
 
 export type NeuralwattFeatureId =
   | "quotaCommand"

package/src/extensions/command-quotas/command.ts

@@ -1,5 +1,8 @@
 import { join } from "node:path";
-import { type ExtensionAPI, getAgentDir } from "@mariozechner/pi-coding-agent";
+import {
+  type ExtensionAPI,
+  getAgentDir,
+} from "@earendil-works/pi-coding-agent";
 import { getNeuralwattApiKey } from "../../lib/env";
 import { fetchQuotas } from "../../utils/quotas";
 import { QuotasComponent } from "./components/quotas-display";

package/src/extensions/command-quotas/components/quota-tabs.ts

@@ -1,5 +1,5 @@
-import type { Theme } from "@mariozechner/pi-coding-agent";
-import { truncateToWidth } from "@mariozechner/pi-tui";
+import type { Theme } from "@earendil-works/pi-coding-agent";
+import { truncateToWidth } from "@earendil-works/pi-tui";
 import type { NeuralwattQuotas } from "../../../types/quota-api";
 import {
   percentCreditsRemaining,

package/src/extensions/command-quotas/components/quotas-display.ts

@@ -1,7 +1,7 @@
-import type { Theme } from "@mariozechner/pi-coding-agent";
-import { DynamicBorder } from "@mariozechner/pi-coding-agent";
-import type { Component, TUI } from "@mariozechner/pi-tui";
-import { Loader, matchesKey, truncateToWidth } from "@mariozechner/pi-tui";
+import type { Theme } from "@earendil-works/pi-coding-agent";
+import { DynamicBorder } from "@earendil-works/pi-coding-agent";
+import type { Component, TUI } from "@earendil-works/pi-tui";
+import { Loader, matchesKey, truncateToWidth } from "@earendil-works/pi-tui";
 import type { NeuralwattQuotas } from "../../../types/quota-api";
 import {
   renderCreditsTab,

package/src/extensions/command-quotas/index.ts

@@ -1,4 +1,4 @@
-import type { ExtensionAPI } from "@mariozechner/pi-coding-agent";
+import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
 import {
   configLoader,
   NEURALWATT_EXTENSIONS_REGISTER_EVENT,

package/src/extensions/provider/context-overflow.ts

@@ -0,0 +1,31 @@
+interface AssistantErrorLike {
+  role: string;
+  stopReason?: string;
+  provider?: string;
+  errorMessage?: string;
+}
+
+const NEURALWATT_CONTEXT_OVERFLOW_PATTERN =
+  /request exceeds model'?s maximum context length/i;
+
+/**
+ * Normalize Neuralwatt context overflow errors so Pi's native overflow
+ * compaction path can detect them and perform compact-and-retry.
+ */
+export function normalizeNeuralwattContextOverflowError<
+  TMessage extends AssistantErrorLike,
+>(message: TMessage, currentProvider?: string): TMessage | undefined {
+  if (message.role !== "assistant") return;
+  if (message.stopReason !== "error") return;
+  if (message.provider !== "neuralwatt" && currentProvider !== "neuralwatt")
+    return;
+
+  const errorMessage = message.errorMessage ?? "";
+  if (errorMessage.includes("context_length_exceeded")) return;
+  if (!NEURALWATT_CONTEXT_OVERFLOW_PATTERN.test(errorMessage)) return;
+
+  return {
+    ...message,
+    errorMessage: `context_length_exceeded: ${errorMessage}`,
+  };
+}

package/src/extensions/provider/index.ts

@@ -1,4 +1,4 @@
-import type { AuthStorage, ExtensionAPI } from "@mariozechner/pi-coding-agent";
+import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
 import {
   configLoader,
   emitConfigUpdated,
@@ -14,22 +14,16 @@ import {
   NEURALWATT_QUOTAS_REQUEST_EVENT,
   NEURALWATT_QUOTAS_UPDATED_EVENT,
   type NeuralwattQuotasUpdatedPayload,
-  parseQuotaHeaders,
 } from "../../types/quota-events";
+import { isOffline } from "../../utils/is-offline";
 import { fetchQuotas } from "../../utils/quotas";
+import { normalizeNeuralwattContextOverflowError } from "./context-overflow";
 import type { NeuralwattModelConfig } from "./models";
 import { NEURALWATT_MODELS_CACHE } from "./models";
+import { buildModelsPayload } from "./provider-payload";
+import { buildQuotasFromHeaders, fetchRequestedQuotas } from "./quota-store";
 
-function buildModelsPayload(models: NeuralwattModelConfig[]) {
-  return models.map(({ fast: _fast, ...model }) => ({
-    ...model,
-    compat: {
-      supportsDeveloperRole: false,
-      maxTokensField: "max_tokens",
-      ...model.compat,
-    },
-  }));
-}
+const HEADER_EMIT_THROTTLE_MS = 5_000;
 
 function registerNeuralwattProvider(
   pi: ExtensionAPI,
@@ -54,17 +48,15 @@ export default async function (pi: ExtensionAPI) {
   // Register with hardcoded cache immediately so models are available on startup
   registerNeuralwattProvider(pi, NEURALWATT_MODELS_CACHE);
 
-  // Track which feature extensions loaded
   const loadedFeatures = new Set<NeuralwattFeatureId>();
 
-  // Register settings (in the provider, so it's always available)
+  // Register settings in the provider so it is always available.
   registerNeuralwattSettings(pi, {
     getLoadedFeatures: () => loadedFeatures,
   });
 
-  // --- Quota store (event-based) ---
   let lastHeaderEmitAt = 0;
-  const HEADER_EMIT_THROTTLE_MS = 5_000;
+  let quotaRequestInFlight = false;
 
   function emitQuotas(
     quotas: NeuralwattQuotas,
@@ -77,94 +69,67 @@ export default async function (pi: ExtensionAPI) {
     pi.events.emit(NEURALWATT_QUOTAS_UPDATED_EVENT, { quotas, source });
   }
 
-  // Ingest quotas from response headers
+  pi.on("message_end", (event, ctx) => {
+    const message = normalizeNeuralwattContextOverflowError(
+      event.message,
+      ctx.model?.provider,
+    );
+    if (!message) return;
+    return { message };
+  });
+
   pi.on("after_provider_response", (event, ctx) => {
     if (ctx.model?.provider !== "neuralwatt") return;
-    const headerQuotas = parseQuotaHeaders(event.headers);
-    if (!headerQuotas) return;
-
-    const quotas: NeuralwattQuotas = {
-      snapshot_at: new Date().toISOString(),
-      balance: {
-        credits_remaining_usd: headerQuotas.allowanceRemainingUsd,
-        total_credits_usd: 0,
-        credits_used_usd: 0,
-        accounting_method: "token",
-      },
-      usage: {
-        lifetime: { cost_usd: 0, requests: 0, tokens: 0, energy_kwh: 0 },
-        current_month: { cost_usd: 0, requests: 0, tokens: 0, energy_kwh: 0 },
-      },
-      limits: { overage_limit_usd: null, rate_limit_tier: "standard" },
-      subscription:
-        headerQuotas.subscriptionPlan !== "none" &&
-        headerQuotas.energyRemaining !== undefined
-          ? {
-              plan: headerQuotas.subscriptionPlan,
-              status: "active",
-              billing_interval: "month",
-              current_period_start: "",
-              current_period_end: "",
-              auto_renew: false,
-              kwh_included: headerQuotas.energyIncluded ?? 0,
-              kwh_used: headerQuotas.energyUsed ?? 0,
-              kwh_remaining: headerQuotas.energyRemaining,
-              in_overage: false,
-            }
-          : null,
-      key: { name: "", allowance: null },
-    };
-
+    const quotas = buildQuotasFromHeaders(event.headers);
+    if (!quotas) return;
     emitQuotas(quotas, "header");
   });
 
-  // Respond to quota requests from other extensions
-  let quotaRequestInFlight = false;
   pi.events.on(NEURALWATT_QUOTAS_REQUEST_EVENT, async (data: unknown) => {
     if (quotaRequestInFlight) return;
     quotaRequestInFlight = true;
     try {
-      if (!data || typeof data !== "object") return;
-      const { authStorage } = data as { authStorage?: AuthStorage };
-      if (!authStorage) return;
-      const apiKey = await getNeuralwattApiKey(authStorage);
-      if (!apiKey) return;
-      const result = await fetchQuotas(apiKey);
-      if (result.success) emitQuotas(result.data.quotas, "api");
+      const quotas = await fetchRequestedQuotas(data);
+      if (quotas)
+        pi.events.emit(NEURALWATT_QUOTAS_UPDATED_EVENT, {
+          quotas,
+          source: "api",
+        });
     } finally {
       quotaRequestInFlight = false;
     }
   });
 
-  // Collect which feature extensions are loaded
   pi.events.on(NEURALWATT_EXTENSIONS_REGISTER_EVENT, (data: unknown) => {
     const { feature } = data as { feature: NeuralwattFeatureId };
     loadedFeatures.add(feature);
   });
 
-  // On session start: fetch live models, request extensions, emit config, fetch quotas
   pi.on("session_start", async (_event, ctx) => {
     loadedFeatures.clear();
     pi.events.emit(NEURALWATT_EXTENSIONS_REQUEST_EVENT, undefined);
     emitConfigUpdated(pi);
 
-    // Fetch live models from the API and re-register if successful
-    const result = await fetchModels();
-    if (result.success) {
-      const cacheIds = new Set(NEURALWATT_MODELS_CACHE.map((m) => m.id));
-      const liveIds = new Set(result.models.map((m) => m.id));
-      const added = result.models.filter((m) => !cacheIds.has(m.id));
-      const removed = NEURALWATT_MODELS_CACHE.filter((m) => !liveIds.has(m.id));
-      if (added.length > 0 || removed.length > 0) {
-        const parts: string[] = [];
-        if (added.length > 0) parts.push(`${added.length} new`);
-        if (removed.length > 0) parts.push(`${removed.length} removed`);
-        ctx.ui.notify(
-          `Neuralwatt models updated (${parts.join(", ")})`,
-          "info",
+    if (!isOffline()) {
+      const result = await fetchModels();
+      if (result.success) {
+        const cacheIds = new Set(NEURALWATT_MODELS_CACHE.map((m) => m.id));
+        const liveIds = new Set(result.models.map((m) => m.id));
+        const added = result.models.filter((m) => !cacheIds.has(m.id));
+        const removed = NEURALWATT_MODELS_CACHE.filter(
+          (m) => !liveIds.has(m.id),
         );
+        if (added.length > 0 || removed.length > 0) {
+          const parts: string[] = [];
+          if (added.length > 0) parts.push(`${added.length} new`);
+          if (removed.length > 0) parts.push(`${removed.length} removed`);
+          ctx.ui.notify(
+            `Neuralwatt models updated (${parts.join(", ")})`,
+            "info",
+          );
+        }
+        registerNeuralwattProvider(pi, result.models);
       }
-      registerNeuralwattProvider(pi, result.models);
     }
 
     if (ctx.model?.provider !== "neuralwatt") return;

package/src/extensions/provider/models.test.ts

@@ -207,9 +207,13 @@ describe("Neuralwatt models", () => {
     expect(result.input).toEqual(["text"]);
     expect(result.cost.input).toBe(0.03);
     expect(result.cost.output).toBe(0.16);
-    expect(
-      (result.compat as Record<string, unknown>)?.supportsReasoningEffort,
-    ).toBe(true);
+    expect(result.thinkingLevelMap).toEqual({
+      minimal: "low",
+      low: "low",
+      medium: "medium",
+      high: "high",
+      xhigh: null,
+    });
     expect(result.fast).toBeUndefined();
   });
 
@@ -308,9 +312,13 @@ describe("Neuralwatt models", () => {
     const result = mapApiModel(visionModel);
     expect(result.input).toEqual(["text", "image"]);
     expect(result.reasoning).toBe(true);
-    expect(
-      (result.compat as Record<string, unknown>)?.supportsReasoningEffort,
-    ).toBeUndefined();
+    expect(result.thinkingLevelMap).toEqual({
+      minimal: null,
+      low: null,
+      medium: "medium",
+      high: null,
+      xhigh: null,
+    });
   });
 
   it("should use defaults when metadata is missing", () => {

package/src/extensions/provider/models.ts

@@ -3,19 +3,19 @@
 // Pricing: https://portal.neuralwatt.com/pricing
 // max_model_len from /v1/models, pricing from /pricing page
 
-import type { ProviderModelConfig } from "@mariozechner/pi-coding-agent";
+import type { ProviderModelConfig } from "@earendil-works/pi-coding-agent";
 
 export interface NeuralwattModelConfig extends ProviderModelConfig {
   /** Fast variant of a parent model (e.g. "glm-5-fast" is the fast variant of "zai-org/GLM-5.1-FP8"). */
   fast?: boolean;
 }
 
-const NEURALWATT_REASONING_EFFORT_MAP = {
-  minimal: "low",
-  low: "low",
+const NEURALWATT_BINARY_THINKING_LEVEL_MAP = {
+  minimal: null,
+  low: null,
   medium: "medium",
-  high: "high",
-  xhigh: "high",
+  high: null,
+  xhigh: null,
 } as const;
 
 /** Hardcoded model cache. Used as a fallback on startup before live models are fetched.
@@ -34,7 +34,7 @@ export const NEURALWATT_MODELS_CACHE: NeuralwattModelConfig[] = [
       cacheRead: 0,
       cacheWrite: 0,
     },
-    contextWindow: 262144,
+    contextWindow: 262128,
     maxTokens: 32768,
     compat: {
       supportsDeveloperRole: false,
@@ -54,7 +54,7 @@ export const NEURALWATT_MODELS_CACHE: NeuralwattModelConfig[] = [
       cacheRead: 0,
       cacheWrite: 0,
     },
-    contextWindow: 202752,
+    contextWindow: 202736,
     maxTokens: 32768,
     compat: {
       supportsDeveloperRole: false,
@@ -73,8 +73,9 @@ export const NEURALWATT_MODELS_CACHE: NeuralwattModelConfig[] = [
       cacheRead: 0,
       cacheWrite: 0,
     },
-    contextWindow: 202752,
+    contextWindow: 202736,
     maxTokens: 32768,
+    thinkingLevelMap: NEURALWATT_BINARY_THINKING_LEVEL_MAP,
     compat: {
       supportsDeveloperRole: false,
       maxTokensField: "max_tokens",
@@ -93,7 +94,7 @@ export const NEURALWATT_MODELS_CACHE: NeuralwattModelConfig[] = [
       cacheRead: 0,
       cacheWrite: 0,
     },
-    contextWindow: 202752,
+    contextWindow: 202736,
     maxTokens: 32768,
     compat: {
       supportsDeveloperRole: false,
@@ -112,12 +113,17 @@ export const NEURALWATT_MODELS_CACHE: NeuralwattModelConfig[] = [
       cacheRead: 0,
       cacheWrite: 0,
     },
-    contextWindow: 16384,
+    contextWindow: 16368,
     maxTokens: 4096,
+    thinkingLevelMap: {
+      minimal: "low",
+      low: "low",
+      medium: "medium",
+      high: "high",
+      xhigh: null,
+    },
     compat: {
       supportsDeveloperRole: false,
-      supportsReasoningEffort: true,
-      reasoningEffortMap: NEURALWATT_REASONING_EFFORT_MAP,
       maxTokensField: "max_tokens",
     },
   },
@@ -133,8 +139,9 @@ export const NEURALWATT_MODELS_CACHE: NeuralwattModelConfig[] = [
       cacheRead: 0,
       cacheWrite: 0,
     },
-    contextWindow: 262144,
+    contextWindow: 262128,
     maxTokens: 65536,
+    thinkingLevelMap: NEURALWATT_BINARY_THINKING_LEVEL_MAP,
     compat: {
       supportsDeveloperRole: false,
       maxTokensField: "max_tokens",
@@ -153,7 +160,7 @@ export const NEURALWATT_MODELS_CACHE: NeuralwattModelConfig[] = [
       cacheRead: 0,
       cacheWrite: 0,
     },
-    contextWindow: 262144,
+    contextWindow: 262128,
     maxTokens: 65536,
     compat: {
       supportsDeveloperRole: false,
@@ -172,8 +179,9 @@ export const NEURALWATT_MODELS_CACHE: NeuralwattModelConfig[] = [
       cacheRead: 0,
       cacheWrite: 0,
     },
-    contextWindow: 262144,
+    contextWindow: 262128,
     maxTokens: 65536,
+    thinkingLevelMap: NEURALWATT_BINARY_THINKING_LEVEL_MAP,
     compat: {
       supportsDeveloperRole: false,
       maxTokensField: "max_tokens",
@@ -192,7 +200,7 @@ export const NEURALWATT_MODELS_CACHE: NeuralwattModelConfig[] = [
       cacheRead: 0,
       cacheWrite: 0,
     },
-    contextWindow: 262144,
+    contextWindow: 262128,
     maxTokens: 65536,
     compat: {
       supportsDeveloperRole: false,
@@ -211,8 +219,9 @@ export const NEURALWATT_MODELS_CACHE: NeuralwattModelConfig[] = [
       cacheRead: 0,
       cacheWrite: 0,
     },
-    contextWindow: 196608,
+    contextWindow: 196592,
     maxTokens: 65536,
+    thinkingLevelMap: NEURALWATT_BINARY_THINKING_LEVEL_MAP,
     compat: {
       supportsDeveloperRole: false,
       maxTokensField: "max_tokens",
@@ -230,8 +239,9 @@ export const NEURALWATT_MODELS_CACHE: NeuralwattModelConfig[] = [
       cacheRead: 0,
       cacheWrite: 0,
     },
-    contextWindow: 262144,
+    contextWindow: 262128,
     maxTokens: 65536,
+    thinkingLevelMap: NEURALWATT_BINARY_THINKING_LEVEL_MAP,
     compat: {
       supportsDeveloperRole: false,
       maxTokensField: "max_tokens",
@@ -250,7 +260,7 @@ export const NEURALWATT_MODELS_CACHE: NeuralwattModelConfig[] = [
       cacheRead: 0,
       cacheWrite: 0,
     },
-    contextWindow: 262144,
+    contextWindow: 262128,
     maxTokens: 65536,
     compat: {
       supportsDeveloperRole: false,
@@ -262,15 +272,16 @@ export const NEURALWATT_MODELS_CACHE: NeuralwattModelConfig[] = [
     id: "Qwen/Qwen3.6-35B-A3B",
     name: "Qwen3.6 35B",
     reasoning: true,
-    input: ["text"],
+    input: ["text", "image"],
     cost: {
       input: 0.05,
       output: 0.1,
       cacheRead: 0,
       cacheWrite: 0,
     },
-    contextWindow: 131072,
+    contextWindow: 131056,
     maxTokens: 32768,
+    thinkingLevelMap: NEURALWATT_BINARY_THINKING_LEVEL_MAP,
     compat: {
       supportsDeveloperRole: false,
       maxTokensField: "max_tokens",
@@ -282,14 +293,14 @@ export const NEURALWATT_MODELS_CACHE: NeuralwattModelConfig[] = [
     name: "Qwen3.6 35B Fast",
     reasoning: false,
     fast: true,
-    input: ["text"],
+    input: ["text", "image"],
     cost: {
       input: 0.05,
       output: 0.1,
       cacheRead: 0,
       cacheWrite: 0,
     },
-    contextWindow: 131072,
+    contextWindow: 131056,
     maxTokens: 32768,
     compat: {
       supportsDeveloperRole: false,

package/src/extensions/provider/provider-payload.ts

@@ -0,0 +1,12 @@
+import type { NeuralwattModelConfig } from "./models";
+
+export function buildModelsPayload(models: NeuralwattModelConfig[]) {
+  return models.map(({ fast: _fast, ...model }) => ({
+    ...model,
+    compat: {
+      supportsDeveloperRole: false,
+      maxTokensField: "max_tokens" as const,
+      ...model.compat,
+    },
+  }));
+}

package/src/extensions/provider/quota-store.ts

@@ -0,0 +1,57 @@
+import type { AuthStorage } from "@earendil-works/pi-coding-agent";
+import { getNeuralwattApiKey } from "../../lib/env";
+import type { NeuralwattQuotas } from "../../types/quota-api";
+import { parseQuotaHeaders } from "../../types/quota-events";
+import { fetchQuotas } from "../../utils/quotas";
+
+export function buildQuotasFromHeaders(
+  headers: Record<string, string>,
+): NeuralwattQuotas | undefined {
+  const headerQuotas = parseQuotaHeaders(headers);
+  if (!headerQuotas) return;
+
+  return {
+    snapshot_at: new Date().toISOString(),
+    balance: {
+      credits_remaining_usd: headerQuotas.allowanceRemainingUsd,
+      total_credits_usd: 0,
+      credits_used_usd: 0,
+      accounting_method: "token",
+    },
+    usage: {
+      lifetime: { cost_usd: 0, requests: 0, tokens: 0, energy_kwh: 0 },
+      current_month: { cost_usd: 0, requests: 0, tokens: 0, energy_kwh: 0 },
+    },
+    limits: { overage_limit_usd: null, rate_limit_tier: "standard" },
+    subscription:
+      headerQuotas.subscriptionPlan !== "none" &&
+      headerQuotas.energyRemaining !== undefined
+        ? {
+            plan: headerQuotas.subscriptionPlan,
+            status: "active",
+            billing_interval: "month",
+            current_period_start: "",
+            current_period_end: "",
+            auto_renew: false,
+            kwh_included: headerQuotas.energyIncluded ?? 0,
+            kwh_used: headerQuotas.energyUsed ?? 0,
+            kwh_remaining: headerQuotas.energyRemaining,
+            in_overage: false,
+          }
+        : null,
+    key: { name: "", allowance: null },
+  };
+}
+
+export async function fetchRequestedQuotas(
+  data: unknown,
+): Promise<NeuralwattQuotas | undefined> {
+  if (!data || typeof data !== "object") return;
+  const { authStorage } = data as { authStorage?: AuthStorage };
+  if (!authStorage) return;
+  const apiKey = await getNeuralwattApiKey(authStorage);
+  if (!apiKey) return;
+  const result = await fetchQuotas(apiKey);
+  if (!result.success) return;
+  return result.data.quotas;
+}

package/src/extensions/quota-warnings/index.ts

@@ -1,7 +1,7 @@
 import type {
   ExtensionAPI,
   ExtensionContext,
-} from "@mariozechner/pi-coding-agent";
+} from "@earendil-works/pi-coding-agent";
 import {
   configLoader,
   NEURALWATT_CONFIG_UPDATED_EVENT,

package/src/extensions/quota-warnings/notifier.ts

@@ -1,4 +1,4 @@
-import type { ExtensionContext } from "@mariozechner/pi-coding-agent";
+import type { ExtensionContext } from "@earendil-works/pi-coding-agent";
 import type { NeuralwattQuotas } from "../../types/quota-api";
 import { formatKwh, formatUsd } from "../../utils/quota-format";
 

package/src/extensions/sub-bar-integration/index.ts

@@ -3,7 +3,7 @@ import type {
   ExtensionAPI,
   ExtensionContext,
   Theme,
-} from "@mariozechner/pi-coding-agent";
+} from "@earendil-works/pi-coding-agent";
 import {
   configLoader,
   NEURALWATT_CONFIG_UPDATED_EVENT,

package/src/lib/env.ts

@@ -1,4 +1,4 @@
-import type { AuthStorage } from "@mariozechner/pi-coding-agent";
+import type { AuthStorage } from "@earendil-works/pi-coding-agent";
 
 const PROVIDER_ID = "neuralwatt";
 

package/src/lib/fetch-models.ts

@@ -2,6 +2,22 @@ import type { NeuralwattModelConfig } from "../extensions/provider/models";
 
 const FETCH_TIMEOUT_MS = 15_000;
 
+const NEURALWATT_BINARY_THINKING_LEVEL_MAP = {
+  minimal: null,
+  low: null,
+  medium: "medium",
+  high: null,
+  xhigh: null,
+} as const;
+
+const GPT_OSS_THINKING_LEVEL_MAP = {
+  minimal: "low",
+  low: "low",
+  medium: "medium",
+  high: "high",
+  xhigh: null,
+} as const;
+
 export interface ApiModelMetadata {
   display_name: string;
   description: string | null;
@@ -50,14 +66,6 @@ export interface ApiResponse {
   data: ApiModel[];
 }
 
-const NEURALWATT_REASONING_EFFORT_MAP = {
-  minimal: "low",
-  low: "low",
-  medium: "medium",
-  high: "high",
-  xhigh: "high",
-} as const;
-
 /** Identify fast variants by their owned_by field or naming convention. */
 function isFastModel(model: ApiModel): boolean {
   if (model.owned_by === "neuralwatt") return true;
@@ -98,13 +106,11 @@ export function mapApiModel(model: ApiModel): NeuralwattModelConfig {
     result.maxTokens = meta.limits.max_output_tokens;
   }
 
-  // Reasoning effort support
-  if (meta?.capabilities.reasoning_effort) {
-    result.compat = {
-      ...result.compat,
-      supportsReasoningEffort: true,
-      reasoningEffortMap: NEURALWATT_REASONING_EFFORT_MAP,
-    };
+  if (result.reasoning) {
+    result.thinkingLevelMap =
+      model.id === "openai/gpt-oss-20b"
+        ? GPT_OSS_THINKING_LEVEL_MAP
+        : NEURALWATT_BINARY_THINKING_LEVEL_MAP;
   }
 
   return result;

package/src/utils/is-offline.test.ts

@@ -0,0 +1,60 @@
+import { describe, expect, it } from "vitest";
+import { isOffline } from "./is-offline";
+
+describe("isOffline", () => {
+  it("returns true when PI_OFFLINE is 1", () => {
+    const original = process.env.PI_OFFLINE;
+    process.env.PI_OFFLINE = "1";
+    expect(isOffline()).toBe(true);
+    process.env.PI_OFFLINE = original;
+  });
+
+  it("returns true when PI_OFFLINE is true", () => {
+    const original = process.env.PI_OFFLINE;
+    process.env.PI_OFFLINE = "true";
+    expect(isOffline()).toBe(true);
+    process.env.PI_OFFLINE = original;
+  });
+
+  it("returns true when PI_OFFLINE is yes", () => {
+    const original = process.env.PI_OFFLINE;
+    process.env.PI_OFFLINE = "yes";
+    expect(isOffline()).toBe(true);
+    process.env.PI_OFFLINE = original;
+  });
+
+  it("returns false when PI_OFFLINE is unset", () => {
+    const original = process.env.PI_OFFLINE;
+    delete process.env.PI_OFFLINE;
+    expect(isOffline()).toBe(false);
+    process.env.PI_OFFLINE = original;
+  });
+
+  it("returns false when PI_OFFLINE is 0", () => {
+    const original = process.env.PI_OFFLINE;
+    process.env.PI_OFFLINE = "0";
+    expect(isOffline()).toBe(false);
+    process.env.PI_OFFLINE = original;
+  });
+
+  it("returns false when PI_OFFLINE is false", () => {
+    const original = process.env.PI_OFFLINE;
+    process.env.PI_OFFLINE = "false";
+    expect(isOffline()).toBe(false);
+    process.env.PI_OFFLINE = original;
+  });
+
+  it("returns false when PI_OFFLINE is no", () => {
+    const original = process.env.PI_OFFLINE;
+    process.env.PI_OFFLINE = "no";
+    expect(isOffline()).toBe(false);
+    process.env.PI_OFFLINE = original;
+  });
+
+  it("returns false for other values", () => {
+    const original = process.env.PI_OFFLINE;
+    process.env.PI_OFFLINE = "maybe";
+    expect(isOffline()).toBe(false);
+    process.env.PI_OFFLINE = original;
+  });
+});

package/src/utils/is-offline.ts

@@ -0,0 +1,4 @@
+export function isOffline(): boolean {
+  const value = process.env.PI_OFFLINE;
+  return value === "1" || value === "true" || value === "yes";
+}

package/src/utils/quota-bar.ts

@@ -1,4 +1,4 @@
-import type { Theme } from "@mariozechner/pi-coding-agent";
+import type { Theme } from "@earendil-works/pi-coding-agent";
 
 export type Severity = "success" | "warning" | "error";