pi-ollama-cloud 0.2.0 → 0.3.0

package/CHANGELOG.md

@@ -2,6 +2,15 @@
 
 All notable changes to this project will be documented in this file.
 
+## [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.
+
 ## [0.2.0] - 2026-04-28
 
 - Add `PI_OLLAMA_WEB_TOOLS` environment variable to optionally disable `ollama_web_search` and `ollama_web_fetch` tool registrations. Set to `0`, `false`, `no`, `off`, or an empty string to opt-out. The model provider and `/ollama-cloud-refresh` command remain active regardless.

package/models.ts

@@ -1,21 +1,21 @@
 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";
 
 // --- Constants ---
-
 const CACHE_DIR = join(getAgentDir(), "cache");
 const CACHE_FILE = join(CACHE_DIR, "ollama-cloud-models.json");
 const FETCH_TIMEOUT_MS = 10000;
 
-/**
- * Base URL for the Ollama Cloud API.
- * Defaults to "https://ollama.com"; override with OLLAMA_API_BASE to point at a proxy or self-hosted instance.
- */
-export const OLLAMA_BASE = (process.env.OLLAMA_API_BASE || "https://ollama.com").replace(/\/+$/, "");
+// --- API fetch ---
+export let OLLAMA_BASE = (process.env.OLLAMA_API_BASE || "https://ollama.com").replace(/\/+$/, "");
 
-// --- Raw API types ---
+// Initialize AuthStorage
+const authStorage = AuthStorage.create();
 
+// --- Raw API types ---
 /** Response from POST /api/show */
 export interface OllamaShowResponse {
   details: {
@@ -38,7 +38,6 @@ interface CachedData {
 }
 
 // --- Assembly: raw API data -> ProviderModelConfig[] ---
-
 function getContextLength(modelInfo: Record<string, unknown>): number {
   for (const [key, value] of Object.entries(modelInfo)) {
     if (key.endsWith(".context_length") && typeof value === "number") {
@@ -48,6 +47,52 @@ 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();
+    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"))
@@ -55,6 +100,7 @@ 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 ?? {}),
@@ -63,7 +109,6 @@ export function assembleModels(raw: Record<string, OllamaShowResponse>): Provide
 }
 
 // --- Fallback models (cold cache) ---
-
 export const FALLBACK_MODELS: ProviderModelConfig[] = [
   {
     id: "glm-5.1:cloud",
@@ -86,7 +131,6 @@ export const FALLBACK_MODELS: ProviderModelConfig[] = [
 ];
 
 // --- Cache I/O ---
-
 export function readCache(): Record<string, OllamaShowResponse> | null {
   try {
     if (!existsSync(CACHE_FILE)) return null;
@@ -107,17 +151,20 @@ export function writeCache(models: Record<string, OllamaShowResponse>): void {
   }
 }
 
-// --- API fetch ---
-
-/**
- * Fetch the full model catalog from Ollama Cloud.
- * Returns null on fatal errors (missing API key, list fetch failed, no models fetched);
- * the caller can rely on fetchModels having already shown a user-facing error.
- */
+// --- Fetch Models ---
 export async function fetchModels(ctx: ExtensionCommandContext): Promise<Record<string, OllamaShowResponse> | null> {
-  const apiKey = await ctx.modelRegistry.getApiKeyForProvider("ollama-cloud");
+  const apiKey = await authStorage.getApiKey("ollama-cloud");
+
   if (!apiKey) {
-    ctx.ui.notify("No Ollama Cloud API key configured (auth.json or OLLAMA_API_KEY env var)", "error");
+    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",
+    );
     return null;
   }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-ollama-cloud",
-  "version": "0.2.0",
+  "version": "0.3.0",
   "type": "module",
   "keywords": [
     "pi-package"