pi-free 2.2.2 → 2.2.4

package/lib/telemetry.ts

@@ -105,67 +105,59 @@ const _store = createJSONStore<TelemetryStore>(TELEMETRY_FILE, {
 // =============================================================================
 
 function deriveModelTelemetry(
-	_modelKey: string,
 	entries: TelemetryEntry[],
 ): ModelTelemetry {
 	const recent = entries.slice(-MAX_RECENT_CALLS);
+
+	let successCalls = 0;
+	let totalTokensFromSuccessful = 0;
+	let totalLatencyFromSuccessful = 0;
+	let totalTokens = 0;
+	let totalPromptTokens = 0;
+	let totalCompletionTokens = 0;
+	let totalLatencyMs = 0;
+	let totalCost = 0;
+
+	for (const e of entries) {
+		totalTokens += e.totalTokens;
+		totalPromptTokens += e.promptTokens;
+		totalCompletionTokens += e.completionTokens;
+		totalLatencyMs += e.latencyMs;
+		totalCost += e.cost;
+		if (e.success) {
+			successCalls++;
+			totalTokensFromSuccessful += e.totalTokens;
+			totalLatencyFromSuccessful += e.latencyMs;
+		}
+	}
+
 	const totalCalls = entries.length;
-	const successCalls = entries.filter((e) => e.success).length;
-	const errorCalls = totalCalls - successCalls;
-
-	const stats = entries.reduce(
-		(acc, e) => {
-			acc.totalTokens += e.totalTokens;
-			acc.totalPromptTokens += e.promptTokens;
-			acc.totalCompletionTokens += e.completionTokens;
-			acc.totalLatencyMs += e.latencyMs;
-			acc.totalCost += e.cost;
-			return acc;
-		},
-		{
-			totalTokens: 0,
-			totalPromptTokens: 0,
-			totalCompletionTokens: 0,
-			totalLatencyMs: 0,
-			totalCost: 0,
-		},
-	);
-
-	const totalSuccessEntries = entries.filter((e) => e.success);
-	const totalTokensFromSuccessful = totalSuccessEntries.reduce(
-		(s, e) => s + e.totalTokens,
-		0,
-	);
-	const totalLatencyFromSuccessful = totalSuccessEntries.reduce(
-		(s, e) => s + e.latencyMs,
-		0,
-	);
 
 	return {
 		totalCalls,
 		successCalls,
-		errorCalls,
-		totalTokens: stats.totalTokens,
-		totalPromptTokens: stats.totalPromptTokens,
-		totalCompletionTokens: stats.totalCompletionTokens,
-		totalLatencyMs: stats.totalLatencyMs,
-		totalCost: stats.totalCost,
+		errorCalls: totalCalls - successCalls,
+		totalTokens,
+		totalPromptTokens,
+		totalCompletionTokens,
+		totalLatencyMs,
+		totalCost,
 		avgLatencyMs:
-			totalSuccessEntries.length > 0
-				? Math.round(totalLatencyFromSuccessful / totalSuccessEntries.length)
+			successCalls > 0
+				? Math.round(totalLatencyFromSuccessful / successCalls)
 				: 0,
 		avgTokensPerSecond:
 			totalLatencyFromSuccessful > 0
-				? parseFloat(
+				? Number.parseFloat(
 						(
 							totalTokensFromSuccessful /
 							(totalLatencyFromSuccessful / 1000)
 						).toFixed(1),
-					)
+				)
 				: 0,
 		successRate:
 			totalCalls > 0
-				? parseFloat(((successCalls / totalCalls) * 100).toFixed(1))
+				? Number.parseFloat(((successCalls / totalCalls) * 100).toFixed(1))
 				: 0,
 		recentCalls: recent,
 	};
@@ -186,7 +178,7 @@ async function addEntry(entry: TelemetryEntry): Promise<void> {
 			...store,
 			models: {
 				...store.models,
-				[modelKey]: deriveModelTelemetry(modelKey, pruned),
+				[modelKey]: deriveModelTelemetry(pruned),
 			},
 			lastUpdated: Date.now(),
 		};
@@ -310,7 +302,7 @@ export async function recordModelCall(
 	const totalTokens = usage.totalTokens || usage.input + usage.output;
 	const tokensPerSecond =
 		latencyMs > 0
-			? parseFloat((totalTokens / (latencyMs / 1000)).toFixed(1))
+			? Number.parseFloat((totalTokens / (latencyMs / 1000)).toFixed(1))
 			: 0;
 
 	const entry: TelemetryEntry = {

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "pi-free",
-	"version": "2.2.2",
+	"version": "2.2.4",
 	"type": "module",
 	"description": "AI model providers for Pi with free model filtering and dynamic model fetching",
 	"keywords": [
@@ -52,7 +52,8 @@
 		"test": "vitest",
 		"test:ui": "vitest --ui",
 		"test:run": "vitest run",
-		"smoke:cline": "tsx scripts/smoke-cline-xml-bridge.ts"
+		"smoke:cline": "tsx scripts/smoke-cline-xml-bridge.ts",
+		"smoke:openmodel": "tsx scripts/smoke-openmodel-wire-format.ts"
 	},
 	"peerDependencies": {
 		"@earendil-works/pi-ai": "^0.79.8",

package/provider-failover/benchmark-lookup.ts

@@ -551,23 +551,35 @@ function tryDirectSubstringMatch(
 	modelId: string,
 	modelName: string,
 ): HardcodedBenchmark | null {
+	// Collect ALL substring matches, then return the LONGEST key. This
+	// prevents short general keys (e.g. "mistral-medium-3") from shadowing
+	// longer specific keys (e.g. "mistral-medium-3.5") when a provider
+	// uses a different separator convention in the model ID.
+	let bestKey: string | null = null;
+	let bestData: HardcodedBenchmark | null = null;
 	for (const [key, data] of Object.entries(HARDCODED_BENCHMARKS) as [
 		string,
 		HardcodedBenchmark,
 	][]) {
 		if (search.includes(key.toLowerCase())) {
-			logDebug({
-				provider,
-				modelId,
-				modelName,
-				action: "match",
-				strategy: "direct-substring",
-				matchKey: key,
-				codingIndex: data.codingIndex,
-			});
-			return data;
+			if (bestKey === null || key.length > bestKey.length) {
+				bestKey = key;
+				bestData = data;
+			}
 		}
 	}
+	if (bestKey !== null && bestData !== null) {
+		logDebug({
+			provider,
+			modelId,
+			modelName,
+			action: "match",
+			strategy: "direct-substring",
+			matchKey: bestKey,
+			codingIndex: bestData.codingIndex,
+		});
+		return bestData;
+	}
 	return null;
 }
 
@@ -685,14 +697,17 @@ export function findHardcodedBenchmark(
 
 	logDebug({ provider, modelId, modelName, action: "attempt" });
 
-	// 1. Direct substring match
-	const direct = tryDirectSubstringMatch(search, provider, modelId, modelName);
-	if (direct) return direct;
-
-	// 2. Variant alias matching
+	// 1. Variant alias matching (human-curated, runs first so deliberate
+	// aliases for separator/suffix mismatches can override generic substring
+	// matches).
 	const variant = tryVariantAliasMatch(search, provider, modelId, modelName);
 	if (variant) return variant;
 
+	// 2. Direct substring match (longest-key wins, so "minimax-m2.5" beats
+	// "minimax-m2" when both could match).
+	const direct = tryDirectSubstringMatch(search, provider, modelId, modelName);
+	if (direct) return direct;
+
 	// 3. Provider-specific normalization
 	const { result: normalizedResult, normalized } = tryProviderNormalizedMatch(
 		modelId,

package/provider-helper.ts

@@ -13,14 +13,11 @@ import type {
 } from "@earendil-works/pi-coding-agent";
 import { saveConfig } from "./config.ts";
 import { createLogger } from "./lib/logger.ts";
+import type { ModelsDevEnrichedMetadata } from "./lib/types.ts";
 import { enhanceModelNameWithCodingIndex } from "./provider-failover/benchmark-lookup.ts";
 
 const _logger = createLogger("provider-helper");
 
-type ModelsDevEnrichedMetadata = {
-	modelsDev?: Parameters<typeof enhanceModelNameWithCodingIndex>[3];
-};
-
 // =============================================================================
 // Types
 // =============================================================================
@@ -66,6 +63,14 @@ export interface OpenAICompatibleConfig {
 	baseUrl: string;
 	/** Environment variable name for the API key */
 	apiKey: string;
+	/**
+	 * Wire API to use. Defaults to `"openai-completions"` for backward
+	 * compatibility with the 17 existing providers that pass through
+	 * this helper without setting it. Set to `"anthropic-messages"`
+	 * for Anthropic-protocol gateways (e.g. OpenModel). The pi-ai
+	 * runtime dispatches to the right client based on this value.
+	 */
+	api?: "openai-completions" | "anthropic-messages";
 	/** Additional headers to include */
 	headers?: Record<string, string>;
 	/** OAuth configuration (optional) */
@@ -105,12 +110,19 @@ export function registerOpenAICompatible(
 	config: OpenAICompatibleConfig,
 	models: ProviderModelConfig[],
 ): void {
-	const { providerId, baseUrl, apiKey, headers, oauth } = config;
+	const {
+		providerId,
+		baseUrl,
+		apiKey,
+		api = "openai-completions",
+		headers,
+		oauth,
+	} = config;
 
 	pi.registerProvider(providerId, {
 		baseUrl,
 		apiKey,
-		api: "openai-completions" as const,
+		api,
 		headers: {
 			"User-Agent": "pi-free-providers",
 			...headers,
@@ -143,13 +155,20 @@ export function createCtxReRegister(
 	},
 	config: OpenAICompatibleConfig,
 ): (models: ProviderModelConfig[]) => void {
-	const { providerId, baseUrl, apiKey, headers, oauth } = config;
+	const {
+		providerId,
+		baseUrl,
+		apiKey,
+		api = "openai-completions",
+		headers,
+		oauth,
+	} = config;
 
 	return (models: ProviderModelConfig[]) => {
 		ctx.modelRegistry.registerProvider(providerId, {
 			baseUrl,
 			apiKey,
-			api: "openai-completions" as const,
+			api,
 			headers: {
 				"User-Agent": "pi-free-providers",
 				...headers,