pi-ollama-cloud 0.2.1 → 0.3.1

package/CHANGELOG.md

@@ -2,6 +2,20 @@
 
 All notable changes to this project will be documented in this file.
 
+## [0.3.1] - 2026-05-05
+
+- Fix `OLLAMA_API_KEY` env var not being respected by `fetchModels` and web tools. pi-ai does not know about the `ollama-cloud` provider ID, so `AuthStorage.getApiKey()` alone misses the env var. Added explicit `process.env.OLLAMA_API_KEY` fallback.
+- Switch web tools to `AuthStorage.create()` for API key lookup, matching the `models.ts` auth pattern from v0.2.1.
+- Add null-safe access to `data.details?.family` in `resolveThinkingLevelMap`.
+- Change `OLLAMA_BASE` from `export let` to `export const` to prevent accidental mutation.
+- Fix fallback model IDs to use real Ollama Cloud identifiers (`glm-5.1`, `gemma4:31b`) instead of synthetic `:cloud` suffixes.
+- Add smoke test workflow for CI.
+
+## [0.3.0] - 2026-05-04
+
+- Derive `thinkingLevelMap` from pi's built-in model definitions instead of hardcoding model-family mappings. The extension now picks up thinking level metadata automatically when pi-mono adds or updates it for any model.
+- Add family-based fallback matching: when an Ollama Cloud model ID doesn't match a pi model ID exactly, the extension now tries matching by model family (via Ollama's `details.family` field). For example, `gemma4:31b` correctly picks up Gemma 4's thinking level map from pi.
+
 ## [0.2.1] - 2026-04-29
 
 - Fix API key retrieval by using `AuthStorage` instead of `ctx.modelRegistry.getApiKeyForProvider`. The provider-level API key lookup was failing, causing auth to only work when an environment variable was set. Now reads from `auth.json` directly via the pi `AuthStorage` class.

package/models.ts

@@ -1,7 +1,12 @@
 import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs";
 import { join } from "node:path";
-import { type ExtensionCommandContext, getAgentDir, type ProviderModelConfig } from "@mariozechner/pi-coding-agent";
-import { AuthStorage } from "@mariozechner/pi-coding-agent";
+import { getModels, getProviders } from "@mariozechner/pi-ai";
+import {
+  AuthStorage,
+  type ExtensionCommandContext,
+  getAgentDir,
+  type ProviderModelConfig,
+} from "@mariozechner/pi-coding-agent";
 
 // --- Constants ---
 const CACHE_DIR = join(getAgentDir(), "cache");
@@ -9,7 +14,7 @@ const CACHE_FILE = join(CACHE_DIR, "ollama-cloud-models.json");
 const FETCH_TIMEOUT_MS = 10000;
 
 // --- API fetch ---
-export let OLLAMA_BASE = (process.env.OLLAMA_API_BASE || "https://ollama.com").replace(/\/+$/, "");
+export const OLLAMA_BASE = (process.env.OLLAMA_API_BASE || "https://ollama.com").replace(/\/+$/, "");
 
 // Initialize AuthStorage
 const authStorage = AuthStorage.create();
@@ -46,6 +51,54 @@ function getContextLength(modelInfo: Record<string, unknown>): number {
   return 128000;
 }
 
+// --- Built-in model knowledge index ---
+// Build a lookup of model ID -> thinkingLevelMap from pi's built-in models.
+// This avoids hardcoding model-family mappings: when pi-mono updates its
+// model definitions (e.g. DeepSeek V4's thinking levels), the extension
+// picks up the changes automatically.
+const BUILTIN_THINKING_MAP: Record<string, ProviderModelConfig["thinkingLevelMap"]> = {};
+// Fallback: family stem -> [stem, thinkingLevelMap] pairs for models whose Ollama Cloud
+// ID doesn't match exactly.  The stem is derived by stripping provider prefixes and
+// non-alphanumeric characters (e.g. "gemma-4-31b-it" -> "gemma431bit").
+// When looking up an Ollama model by its details.family field, we search for a pi stem
+// that starts with the family stem (e.g. family "gemma4" -> pi "gemma431bit").
+// Entries are sorted longest-first so the most specific match wins.
+const BUILTIN_FAMILY_ENTRIES: [string, NonNullable<ProviderModelConfig["thinkingLevelMap"]>][] = [];
+for (const provider of getProviders()) {
+  for (const model of getModels(provider as any)) {
+    if (model.thinkingLevelMap) {
+      BUILTIN_THINKING_MAP[model.id] = model.thinkingLevelMap;
+      const stem = model.id
+        .replace(/^[a-z0-9-]+\//, "") // strip provider prefix (e.g. "zai/", "deepseek/")
+        .replace(/[^a-zA-Z0-9]/g, "") // strip non-alphanumeric
+        .toLowerCase();
+      BUILTIN_FAMILY_ENTRIES.push([stem, model.thinkingLevelMap]);
+    }
+  }
+}
+// Longest stems first so a more specific match (e.g. "gemma431bit") wins over a generic one (e.g. "gemma4").
+BUILTIN_FAMILY_ENTRIES.sort((a, b) => b[0].length - a[0].length);
+
+function resolveThinkingLevelMap(modelId: string, data: OllamaShowResponse): ProviderModelConfig["thinkingLevelMap"] {
+  // 1. Exact ID match (e.g. "deepseek-v4-pro")
+  const exact = BUILTIN_THINKING_MAP[modelId];
+  if (exact) return exact;
+
+  // 2. Family-based fallback: match Ollama's details.family against pi model stems
+  if (data.capabilities?.includes("thinking")) {
+    const familyStem = data.details?.family?.replace(/[^a-zA-Z0-9]/g, "").toLowerCase() ?? "";
+    if (familyStem) {
+      for (const [stem, tlm] of BUILTIN_FAMILY_ENTRIES) {
+        if (stem.startsWith(familyStem)) {
+          return tlm;
+        }
+      }
+    }
+  }
+
+  return undefined;
+}
+
 export function assembleModels(raw: Record<string, OllamaShowResponse>): ProviderModelConfig[] {
   return Object.entries(raw)
     .filter(([, data]) => data.capabilities?.includes("tools"))
@@ -53,6 +106,7 @@ export function assembleModels(raw: Record<string, OllamaShowResponse>): Provide
       id,
       name: id,
       reasoning: data.capabilities?.includes("thinking") ?? false,
+      thinkingLevelMap: resolveThinkingLevelMap(id, data),
       input: (data.capabilities?.includes("vision") ? ["text", "image"] : ["text"]) as ("text" | "image")[],
       cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
       contextWindow: getContextLength(data.model_info ?? {}),
@@ -63,18 +117,18 @@ export function assembleModels(raw: Record<string, OllamaShowResponse>): Provide
 // --- Fallback models (cold cache) ---
 export const FALLBACK_MODELS: ProviderModelConfig[] = [
   {
-    id: "glm-5.1:cloud",
-    name: "GLM 5.1 Cloud",
-    reasoning: true,
+    id: "glm-5.1",
+    name: "GLM 5.1",
+    reasoning: false,
     input: ["text"],
     cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
     contextWindow: 202752,
     maxTokens: 32768,
   },
   {
-    id: "gemma4:cloud",
-    name: "Gemma 4 Cloud",
-    reasoning: true,
+    id: "gemma4:31b",
+    name: "Gemma 4 31B",
+    reasoning: false,
     input: ["text"],
     cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
     contextWindow: 262144,
@@ -105,17 +159,18 @@ export function writeCache(models: Record<string, OllamaShowResponse>): void {
 
 // --- Fetch Models ---
 export async function fetchModels(ctx: ExtensionCommandContext): Promise<Record<string, OllamaShowResponse> | null> {
-  const apiKey = await authStorage.getApiKey("ollama-cloud");
-  
+  const apiKey = (await authStorage.getApiKey("ollama-cloud")) ?? process.env.OLLAMA_API_KEY;
+
   if (!apiKey) {
     ctx.ui.notify(
       "No Ollama Cloud API key found. \n" +
-      "Please ensure your API key is set in: \n" +
-      "- auth.json file (at ~/.pi/agent/auth.json) under 'ollama-cloud' key,\n" +
-      "- or via the CLI --api-key flag.\n" +
-      "Example auth.json entry: \n" +
-      '{ \"ollama-cloud\": { \"type\": \"api_key\", \"key\": \"YOUR_API_KEY\" } }'
-    , "error");
+        "Please ensure your API key is set in: \n" +
+        "- auth.json file (at ~/.pi/agent/auth.json) under 'ollama-cloud' key,\n" +
+        "- or via the CLI --api-key flag.\n" +
+        "Example auth.json entry: \n" +
+        '{ "ollama-cloud": { "type": "api_key", "key": "YOUR_API_KEY" } }',
+      "error",
+    );
     return null;
   }
 
@@ -176,4 +231,4 @@ export async function fetchModels(ctx: ExtensionCommandContext): Promise<Record<
   ctx.ui.notify(`Fetched ${succeeded} model details${failed ? ` (${failed} failed)` : ""}`, "info");
 
   return results;
-}
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-ollama-cloud",
-  "version": "0.2.1",
+  "version": "0.3.1",
   "type": "module",
   "keywords": [
     "pi-package"

package/web-tools.ts

@@ -1,9 +1,4 @@
-import {
-  type ExtensionAPI,
-  type ExtensionContext,
-  keyHint,
-  truncateToVisualLines,
-} from "@mariozechner/pi-coding-agent";
+import { AuthStorage, type ExtensionAPI, keyHint, truncateToVisualLines } from "@mariozechner/pi-coding-agent";
 import { Text, truncateToWidth } from "@mariozechner/pi-tui";
 import { Type } from "@sinclair/typebox";
 import { OLLAMA_BASE } from "./models.ts";
@@ -26,8 +21,10 @@ interface FetchResponse {
 
 // --- Helpers ---
 
-async function getCloudApiKey(ctx: ExtensionContext): Promise<string | undefined> {
-  return ctx.modelRegistry.getApiKeyForProvider("ollama-cloud");
+const authStorage = AuthStorage.create();
+
+async function getCloudApiKey(): Promise<string | undefined> {
+  return authStorage.getApiKey("ollama-cloud") ?? process.env.OLLAMA_API_KEY;
 }
 
 function noApiKeyError() {
@@ -121,8 +118,8 @@ export function registerWebSearchTool(pi: ExtensionAPI) {
         }),
       ),
     }),
-    async execute(_toolCallId, params, signal, _onUpdate, ctx) {
-      const apiKey = await getCloudApiKey(ctx);
+    async execute(_toolCallId, params, signal, _onUpdate, _ctx) {
+      const apiKey = await getCloudApiKey();
       if (!apiKey) return noApiKeyError();
 
       try {
@@ -180,8 +177,8 @@ export function registerWebFetchTool(pi: ExtensionAPI) {
     parameters: Type.Object({
       url: Type.String({ description: "URL to fetch and extract content from", format: "uri" }),
     }),
-    async execute(_toolCallId, params, signal, _onUpdate, ctx) {
-      const apiKey = await getCloudApiKey(ctx);
+    async execute(_toolCallId, params, signal, _onUpdate, _ctx) {
+      const apiKey = await getCloudApiKey();
       if (!apiKey) return noApiKeyError();
 
       try {