pi-free 1.0.3 → 1.0.5

package/CHANGELOG.md

@@ -7,10 +7,47 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [1.0.5] - 2025-04-03
+
+### Fixed
+- **NVIDIA provider non-chat model filtering** (comment/implementation mismatch)
+  - Added modalities-based filtering to exclude embedding, speech-to-text, OCR, and image-gen models
+  - Filters models where `output` is not `["text"]` (e.g., image generation like `black-forest-labs/flux.1-dev`)
+  - Filters models where `input` lacks `"text"` (e.g., OCR like `nvidia/nemoretriever-ocr-v1`, speech-to-text like `openai/whisper-large-v3`)
+  - Updated file comment to accurately describe the filtering behavior
+  - Added 8 comprehensive unit tests for model filtering logic
+
+## [1.0.4] - 2025-04-03
+
+### Fixed
+- **All tests now passing** (127/127)
+  - Fixed mock paths in kilo.test.ts, zen.test.ts, ollama.test.ts
+  - Fixed createCtxReRegister mocks in zen.test.ts and openrouter.test.ts
+  - Fixed cline.test.ts to test actual provider re-registration behavior
+  - Added missing DEFAULT_MIN_SIZE_B constant to openrouter mock
+
+### Changed
+- **Code quality improvements**
+  - Refactored usage modules to break circular dependency (limits.ts ↔ formatters.ts)
+  - Created usage/types.ts with shared interfaces (FreeTierLimit, FreeTierUsage)
+  - Bumped version to 1.0.4
+
+## [1.0.3] - 2025-04-03
+
+### Changed
+- Updated package.json metadata (name, description, keywords, repository URL)
+- Updated .npmignore for cleaner publishes
+
+## [1.0.0] - 2024-03-28
+
 ### Added
-- Structured logging with namespaced logger (lib/logger.ts)
-- LICENSE file (MIT)
-- CHANGELOG.md
+- Initial release with 6 providers: Kilo, Zen, OpenRouter, NVIDIA, Cline, Fireworks
+- Free tier usage tracking across all sessions
+- Provider failover with model hopping
+- Autocompact integration for rate limit recovery
+- Usage widget with glimpseui
+- Command toggles for free/all model filtering
+- Hardcoded benchmark data from Artificial Analysis
 
 ### Changed
 - **Major refactoring**: Split free-tier-limits.ts into usage/* modules
@@ -46,14 +83,3 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - fetchWithRetry() now properly throws after exhausting retries
 - Auth error pattern matching now handles more message variants
 - Test isolation for free-tier-limits tests
-
-## [1.0.0] - 2024-03-28
-
-### Added
-- Initial release with 6 providers: Kilo, Zen, OpenRouter, NVIDIA, Cline, Fireworks
-- Free tier usage tracking across all sessions
-- Provider failover with model hopping
-- Autocompact integration for rate limit recovery
-- Usage widget with glimpseui
-- Command toggles for free/all model filtering
-- Hardcoded benchmark data from Artificial Analysis

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "pi-free",
-	"version": "1.0.3",
+	"version": "1.0.5",
 	"type": "module",
 	"description": "AIO Free AI models for Pi - Access free models from Kilo, Zen, OpenRouter, NVIDIA, Cline, Mistral, and Ollama",
 	"keywords": [

package/providers/nvidia.ts

@@ -5,8 +5,10 @@
  * All models use NVIDIA's free credit system — requires NVIDIA_API_KEY.
  * Get a free key at: https://build.nvidia.com
  *
- * Small models (< 70B), embedding, speech, OCR, and image-gen models are
- * filtered out to keep the list focused on useful chat/coding models.
+ * Small models (< 70B) are filtered out to keep the list focused on useful
+ * chat/coding models. Non-chat models (embedding, speech-to-text, OCR,
+ * image-gen) are filtered by their modalities (output must be ["text"],
+ * input must include "text").
  *
  * Set NVIDIA_SHOW_PAID=true to show paid-tier models (same key, uses credits).
  */
@@ -52,6 +54,20 @@ async function fetchNvidiaModels(): Promise<ProviderModelConfig[]> {
 	const result = applyHidden(
 		Object.values(provider.models)
 			.filter((m) => isUsableModel(m.id, NVIDIA_MIN_SIZE_B))
+			.filter((m) => {
+				// Filter non-chat models by modalities
+				// Embedding, speech-to-text, OCR, and image-gen models are excluded
+				const modalities = m.modalities;
+				if (modalities) {
+					const output = modalities.output ?? [];
+					const input = modalities.input ?? [];
+					// Exclude models that don't output text (e.g., image generation)
+					if (!output.includes("text")) return false;
+					// Exclude models that don't accept text input (e.g., pure OCR, speech-to-text)
+					if (!input.includes("text")) return false;
+				}
+				return true;
+			})
 			.filter((m) => {
 				// All NVIDIA models are credit-based (no hard cost.input = 0 distinction).
 				// Respect NVIDIA_SHOW_PAID: without the flag, only expose models marked free.

package/usage/formatters.ts

@@ -3,25 +3,9 @@
  */
 
 import type { CumulativeUsageReport } from "./cumulative.ts";
-import {
-	type FreeTierLimit,
-	getFreeTierUsage,
-	getLimitWarning,
-} from "./limits.ts";
+import { getFreeTierUsage, getLimitWarning } from "./limits.ts";
 import type { SessionUsageReport } from "./tracking.ts";
-
-export interface FreeTierUsage {
-	provider: string;
-	requestsToday: number;
-	requestsThisHour: number;
-	requestsThisMonth?: number;
-	limit: FreeTierLimit;
-	remainingToday?: number;
-	remainingThisHour?: number;
-	remainingThisMonth?: number;
-	percentUsed: number;
-	status: "ok" | "warning" | "critical" | "unknown";
-}
+// Types imported via limits.ts (which re-exports from types.ts)
 
 export function formatSessionUsage(report: SessionUsageReport): string {
 	if (report.providers.length === 0) {

package/usage/limits.ts

@@ -9,6 +9,7 @@
 
 import { createLogger } from "../lib/logger.ts";
 import { getDailyRequestCount } from "./metrics.ts";
+import type { FreeTierLimit, FreeTierUsage } from "./types.ts";
 
 export {
 	type CumulativeUsageReport,
@@ -30,6 +31,8 @@ export {
 	resetUsageStats,
 	type SessionUsageReport,
 } from "./tracking.ts";
+// Re-export types for consumers
+export type { FreeTierLimit, FreeTierUsage } from "./types.ts";
 
 const _logger = createLogger("free-tier");
 
@@ -37,14 +40,6 @@ const _logger = createLogger("free-tier");
 // Free Tier Limits Configuration
 // =============================================================================
 
-export interface FreeTierLimit {
-	provider: string;
-	requestsPerDay?: number;
-	requestsPerHour?: number;
-	requestsPerMonth?: number;
-	description: string;
-}
-
 export const FREE_TIER_LIMITS: Record<string, FreeTierLimit> = {
 	kilo: {
 		provider: "kilo",
@@ -80,19 +75,6 @@ export const FREE_TIER_LIMITS: Record<string, FreeTierLimit> = {
 // Usage Status and Warnings
 // =============================================================================
 
-export interface FreeTierUsage {
-	provider: string;
-	requestsToday: number;
-	requestsThisHour: number;
-	requestsThisMonth?: number;
-	limit: FreeTierLimit;
-	remainingToday?: number;
-	remainingThisHour?: number;
-	remainingThisMonth?: number;
-	percentUsed: number;
-	status: "ok" | "warning" | "critical" | "unknown";
-}
-
 export function getFreeTierUsage(provider: string): FreeTierUsage {
 	const limit = FREE_TIER_LIMITS[provider];
 	if (!limit) {

package/usage/types.ts

@@ -0,0 +1,26 @@
+/**
+ * Shared types for usage tracking modules
+ *
+ * Extracted to break circular dependency between limits.ts and formatters.ts
+ */
+
+export interface FreeTierLimit {
+	provider: string;
+	requestsPerDay?: number;
+	requestsPerHour?: number;
+	requestsPerMonth?: number;
+	description: string;
+}
+
+export interface FreeTierUsage {
+	provider: string;
+	requestsToday: number;
+	requestsThisHour: number;
+	requestsThisMonth?: number;
+	limit: FreeTierLimit;
+	remainingToday?: number;
+	remainingThisHour?: number;
+	remainingThisMonth?: number;
+	percentUsed: number;
+	status: "ok" | "warning" | "critical" | "unknown";
+}