pi-free 2.0.14 → 2.1.0

package/CHANGELOG.md

@@ -7,6 +7,94 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [2.1.0] - 2026-06-15
+
+### Added
+
+- **Cline XML tool bridge** — Replaced Cline's native OpenAI tool-message path with a custom `streamSimple` XML bridge. Cline-trained models now receive Cline-style XML tool instructions and emit XML tool calls that pi-free converts back to Pi `toolCall` blocks. This fixes strict upstream errors such as `Tool message must have tool_call_id` and `missing field "tool_call_id"` on models like `xiaomi/mimo-v2.5` and `nex-agi/nex-n2-pro:free` ([#232](https://github.com/apmantza/pi-free/pull/232)).
+
+- **Cline-native tool name mapping** — The XML bridge maps Cline-native tool names to Pi runtime tools:
+  - `read_file` → `read`
+  - `write_to_file` → `write`
+  - `replace_in_file` → `edit` (supports multi-block SEARCH/REPLACE diffs as one Pi `edit` call with multiple edits)
+  - `execute_command` → `bash`
+  - `list_files`, `search_files`, `list_code_definition_names` → `bash` (safe command generation)
+  - Unknown Pi tools pass through by their original names ([#235](https://github.com/apmantza/pi-free/pull/235), [#237](https://github.com/apmantza/pi-free/pull/237)).
+
+- **Cline XML thinking-tag hardening** — Strips `<thinking>...</thinking>` blocks, orphan `</thinking>` close tags, and dangling planning text before tool parsing, so Cline models don't emit visible plan text instead of tool calls ([#239](https://github.com/apmantza/pi-free/pull/239), [#240](https://github.com/apmantza/pi-free/pull/240)).
+
+- **Live Cline smoke test** — Added `npm run smoke:cline` gated test that hits the real Cline API and verifies Cline `read_file` XML is converted into a Pi `read` tool call ([#232](https://github.com/apmantza/pi-free/pull/232)).
+
+### Fixed
+
+- **TokenRouter MiniMax-M3 `<think>` leak** — The model sometimes emits DeepSeek-style `<think>` reasoning tags inline in assistant text. Added a `message_end` handler scoped to TokenRouter that extracts these blocks (including unclosed dangling tags) and promotes them to proper `ThinkingContent`, so Pi renders them as reasoning instead of visible text ([#243](https://github.com/apmantza/pi-free/pull/243)).
+
+- **TokenRouter provider** — OpenAI-compatible API gateway at `api.tokenrouter.com/v1` with 88 text chat models. 1 free via hardcoded `KNOWN_FREE_MODELS` + 1 `nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free` model. Set `TOKENROUTER_API_KEY` or add `tokenrouter_api_key` to `~/.pi/free.json` ([#222](https://github.com/apmantza/pi-free/pull/222)).
+
+- **Generic probe system** — New `lib/provider-probe.ts` factory `createProviderProbe()` handles batching, probe-cache integration, auto-hiding, and re-registration. Enables consistent probe commands across providers ([#218](https://github.com/apmantza/pi-free/pull/218)).
+
+- **Probe commands** — New `/probe-deepinfra`, `/probe-sambanova`, `/probe-together`, `/probe-novita` commands test model availability and auto-hide broken models ([#218](https://github.com/apmantza/pi-free/pull/218)).
+
+- **OpenCode probe commands** — `/probe-opencode` and `/probe-opencode-go` detect expired free promotions (reports only, no auto-hide) ([#218](https://github.com/apmantza/pi-free/pull/218)).
+
+- **Session timing metrics** — `wrapSessionStartHandler()` logs wall-clock time per handler in `lib/session-start-metrics.ts`. Wrapped: cline, kilo, routeway, built-in-toggle, dynamic-built-in auto-probe ([#218](https://github.com/apmantza/pi-free/pull/218)).
+
+### Changed
+
+- **Refactored `recordModelCall` signature** — Replaced 5 positional args with an options object (`RecordModelCallOptions`) for `success`, `stopReason`, and `errorMessage` ([#221](https://github.com/apmantza/pi-free/pull/221)).
+
+- **Extracted `sleep` helper and simplified `cleanModelName`** — Shared utilities in `lib/util.ts` ([#221](https://github.com/apmantza/pi-free/pull/221)).
+
+- **Cleanup pass on `lib/` utilities (Sprint B)** — 8 categories of code-quality refactors in [#224](https://github.com/apmantza/pi-free/pull/224):
+  - `open-browser.ts`: `rundll32 url.dll,FileProtocolHandler` replaces `cmd /c start` (CodeQL fix) + strict URL validation (`isSafeUrl`)
+  - `logger.ts`: `parseLogLevel()` validates `LOG_LEVEL` / `PI_FREE_LOG_LEVEL` env vars
+  - `telemetry.ts`: 1h TTL cleanup for `_inFlight` map; migrated to `createJSONStore` (drops ~80 LOC of `load`/`save`/`Lock` boilerplate)
+  - `util.ts`: `OpenAIModelCallbacks` parameter decouples `fetchOpenAICompatibleModels` from `lib/provider-compat.ts` (DIP fix)
+  - `provider-compat.ts`: extracted `isDeepSeekStyleModel()` and `isKimiModel()` predicates + new `KIMI_PROXY_COMPAT` constant
+  - `model-detection.ts`: removed duplicate `isModelFree` (canonical `isFreeModel` in `registry.ts` already exists)
+  - `registry.ts`: removed dead `_pi` parameter from `applyGlobalFilter`
+  - `built-in-toggle.ts`: lazy `_opencodeSession` initialisation (only created when an OpenCode provider is actually captured)
+
+### Removed
+
+- **NVIDIA NIM provider** — Now a built-in Pi provider. Set `NVIDIA_API_KEY` to use directly. Removed `providers/nvidia/`, constants, config re-exports, and tests ([#218](https://github.com/apmantza/pi-free/pull/218)).
+
+### Security
+
+- **CI/release hardening** — Added production dependency audit, lockfile drift check, tarball content/artifact verification, installed entry smoke-load, and pinned-action workflows. Added Dependabot config for npm and GitHub Actions. Hardened helper scripts against PATH-lookup Sonar hotspots by resolving `npm` and `tar` to fixed locations (#236).
+
+- **open-browser: `rundll32` + strict URL validation** — Replaced `cmd /c start "" <url>` with `rundll32 url.dll,FileProtocolHandler <url>` to fix GitHub Advanced Security CodeQL `js/uncontrolled-command-line` (Critical). rundll32 does NOT parse the command line, so the URL is handed to ShellExecute as a literal. Defense-in-depth: `isSafeUrl()` allows only `http`/`https`, rejects control characters, malformed URLs, and overlong URLs (>2048 chars) ([#223](https://github.com/apmantza/pi-free/pull/223), [#224](https://github.com/apmantza/pi-free/pull/224)).
+
+- **Path-validate env-var file overrides** — New `lib/paths.ts` centralises `PI_DATA_DIR`, `ensureDir()`, and `resolveSafeDataFile()` (rejects path separators, null bytes, dot-only, >128-char). Applied to `PI_FREE_LOG_PATH`, `PI_FREE_PROVIDER_CACHE`, `PI_FREE_TELEMETRY_FILE` ([#223](https://github.com/apmantza/pi-free/pull/223)).
+
+- **json-persistence: lock `save`/`load` + atomic `update()`** — `Lock` mutex serialises RMW operations. `clearProviderCache` / `clearAllProviderCaches` now async, use `_cache.update()` ([#218](https://github.com/apmantza/pi-free/pull/218), [#223](https://github.com/apmantza/pi-free/pull/223)).
+
+- **JSONL `append`/`clear` lock** — `createJSONLStore` operations are now async and lock-serialised, preventing `clear` from truncating mid-`append` ([#223](https://github.com/apmantza/pi-free/pull/223)).
+
+- **telemetry: concurrent-write safety** — `Lock` mutex around telemetry writes; `recordModelCall` and `clearTelemetry` are now async and serialized. File path overridable via `PI_FREE_TELEMETRY_FILE` ([#218](https://github.com/apmantza/pi-free/pull/218)).
+
+- **provider-cache: isolated copies** — `loadProviderCache` returns `structuredClone(cached.models)`; `saveProviderCache` uses `update()` for atomic RMW ([#218](https://github.com/apmantza/pi-free/pull/218)).
+
+- **provider-probe: config RMW lock** — `config.ts` `updateConfig()` uses internal `ConfigLock` (promise-chained mutex); provider-probe auto-hide now uses it ([#223](https://github.com/apmantza/pi-free/pull/223)).
+
+- **Prototype pollution reviver** — `safeJsonReviver()` strips `__proto__` / `constructor` keys at every `JSON.parse` level. Applied in `lib/json-persistence.ts`, `config.ts`, `lib/telemetry.ts` ([#223](https://github.com/apmantza/pi-free/pull/223)).
+
+- **Log sanitization** — `scripts/update-benchmarks.ts` now sanitizes external API data before passing to `console.log`/error, preventing log injection (SonarCloud S5693) ([#219](https://github.com/apmantza/pi-free/pull/219)).
+
+## [2.0.15] - 2026-06-02
+
+### Fixed
+
+- **Qwen 3.7 reasoning compat** — `qwen/qwen3.7-max` on Cline/OpenRouter uses DeepSeek-style `reasoning_content` format. Added `DEEPSEEK_PROXY_COMPAT` so Pi preserves and replays reasoning tokens correctly, preventing plan-mode hangs ([#213](https://github.com/apmantza/pi-free/pull/213)).
+
+- **Kimi K2.6 reasoning compat** — Kimi models on NVIDIA/OpenRouter need `requiresReasoningContentOnAssistantMessages: true` to correctly replay reasoning tokens in assistant messages. Without it, the model gets stuck when trying to call tools or produce output after thinking. Refs [earendil-works/pi#5309](https://github.com/earendil-works/pi/issues/5309) ([#213](https://github.com/apmantza/pi-free/pull/213)).
+
+- **MiniMax reasoning compat** — MiniMax M3 and other MiniMax models now have full DeepSeek-style compat (`thinkingFormat: "deepseek"`, `requiresReasoningContentOnAssistantMessages: true`). Previously, models marked `reasoning: true` without `thinkingFormat` caused Pi to enter plan mode but couldn't parse the reasoning tokens, resulting in hangs ([#212](https://github.com/apmantza/pi-free/pull/212), [#213](https://github.com/apmantza/pi-free/pull/213)).
+
+### Added
+
+- **`/probe-routeway` command** — Tests each Routeway model with a minimal chat request and auto-hides models that return 5xx or 404 errors. Runs lazily on first `session_start` with 24h probe cache TTL. Follows the same pattern as `/probe-nvidia` ([#213](https://github.com/apmantza/pi-free/pull/213)).
+
 ## [2.0.14] - 2026-06-02
 
 ### Added
@@ -19,6 +107,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 - **`_pricingKnown` / `_freeKnown` authoritatve flag** — Providers can now signal whether pricing data is authoritative via `_pricingKnown`. When `false`, `isFreeModel` falls back to name-based detection. Kilo's `isFree` API flag now flows through as `_freeKnown` ([#209](https://github.com/apmantza/pi-free/pull/209)).
 
+- **MiniMax reasoning compat** — MiniMax M3 and other MiniMax models now have `supportsReasoningEffort: true` compat settings. Previously, models marked `reasoning: true` without compat caused Pi to enter plan mode without knowing the thinking format, resulting in hangs.
+
 ## [2.0.13] - 2026-05-21
 
 ### Added

package/README.md

@@ -14,19 +14,21 @@ Free and paid AI model providers for [Pi](https://pi.dev). Access **free and pai
 
 When you install pi-free, it:
 
-1. **Registers free-tier providers** with Pi's model picker — Kilo (free), Cline (free), NVIDIA (freemium), ZenMux (paid), CrofAI (paid), Ollama Cloud (freemium), and more
+1. **Registers free-tier providers** with Pi's model picker — Kilo (free), Cline (free), LLM7 (free), TokenRouter (1 free model), ZenMux (paid), CrofAI (paid), Ollama Cloud (freemium), SambaNova (freemium), Codestral (freemium), DeepInfra (trial credit), Together AI (trial credit), Novita (paid), Routeway (paid), and more
 
-2. **Fetches models dynamically** from provider APIs — NVIDIA NIM, ZenMux, CrofAI, and Pi's built-in providers (Mistral, Groq, Cerebras, xAI, Hugging Face, OpenRouter) when API keys are configured
+2. **Captures Pi's built-in OpenCode and OpenRouter providers** with a free/paid toggle — OpenCode and OpenRouter are now built into Pi; pi-free adds `/toggle-opencode` and `/toggle-openrouter` so you can switch between free-only and all models without restart
 
-3. **Filters to show only free models by default** for providers that expose pricing — You see only the models that cost $0 to use. Paid models are hidden until you explicitly toggle them on.
+3. **Fetches models dynamically** from provider APIs — ZenMux, CrofAI, and Pi's built-in providers (Mistral, Groq, Cerebras, xAI, Hugging Face, OpenRouter) when API keys are configured
 
-4. **Provides per-provider toggle commands** — Run `/toggle-{provider}` (e.g., `/toggle-kilo`) to switch between free-only mode and showing all models including paid ones. Changes apply immediately and your preference is saved for the next Pi restart.
+4. **Filters to show only free models by default** for providers that expose pricing — You see only the models that cost $0 to use. Paid models are hidden until you explicitly toggle them on.
 
-5. **Handles authentication for you** — OAuth flows (Kilo, Cline) open your browser automatically; API keys are read from `~/.pi/free.json` or environment variables
+5. **Provides per-provider toggle commands** — Run `/toggle-{provider}` (e.g., `/toggle-kilo`) to switch between free-only mode and showing all models including paid ones. Changes apply immediately and your preference is saved for the next Pi restart.
 
-6. **Adds Coding Index scores** — Model names include a coding benchmark score (CI: 45.2) to help you pick capable coding models at a glance
+6. **Handles authentication for you** — OAuth flows (Kilo, Cline) open your browser automatically; API keys are read from `~/.pi/free.json` or environment variables
 
-7. **Persists your preferences** — Your toggle choices (free vs all models) are saved to `~/.pi/free.json` and remembered across Pi restarts
+7. **Adds Coding Index scores** — Model names include a coding benchmark score (CI: 45.2) to help you pick capable coding models at a glance
+
+8. **Persists your preferences** — Your toggle choices (free vs all models) are saved to `~/.pi/free.json` and remembered across Pi restarts
 
 ---
 
@@ -50,21 +52,22 @@ Free models are shown by default — look for the provider prefixes:
 - `openrouter/` — OpenRouter models (free account required)
 - `cline/` — Cline models (run `/login cline` to use)
 - `llm7/` — LLM7 gateway models (free tier: default/fast selectors, 100 req/hr)
+- `tokenrouter/` — TokenRouter gateway (1 free model: `MiniMax-M3`; requires API key with credits for the rest)
 
 **🔄 Freemium (free tier with limits, then paid):**
 
-- `nvidia/` — NVIDIA NIM models (1,000 free requests/month, then credits)
 - `ollama-cloud/` — Ollama Cloud models (usage-based free tier, resets every 5 hours + 7 days)
 - `sambanova/` — SambaNova Cloud models (20-480 RPM free, no credit card required)
+- `codestral/` — Codestral via Mistral (free Experiment plan: 2 req/min, 1B tokens/month)
 
 **💳 Paid Providers (API key with credits required):**
 
 - `zenmux/` — ZenMux AI gateway (200+ models from OpenAI, Anthropic, Google, etc.)
 - `crofai/` — CrofAI OpenAI-compatible API (streaming, reasoning models)
-- `codestral/` — Codestral via Mistral (free Experiment plan: 2 req/min, 1B tokens/month)
 - `deepinfra/` — DeepInfra inference cloud ($5 one-time trial credit, no credit card)
 - `novita/` — Novita AI (100+ open-source models, OpenAI-compatible, 3 free models)
 - `routeway/` — Routeway AI gateway (OpenAI-compatible, `:free` models)
+- `together/` — Together AI ($1 one-time trial credit, 200+ open-source models)
 
 > **Note:** Paid providers may occasionally offer free models or promotional credits. The `isFreeModel` helper automatically detects free models based on provider pricing data or model names containing "free". For providers that don't expose pricing (like CrofAI), only models with "free" in their names are marked as free.
 
@@ -78,7 +81,7 @@ Free models are shown by default — look for the provider prefixes:
 - `openrouter/` — OpenRouter models (fetched from openrouter.ai, when `OPENROUTER_API_KEY` set)
 - `fastrouter/` — FastRouter models (always discovered, 170+ models, no auth for listing)
 
-**Note:** Fireworks is now a [built-in Pi provider](https://github.com/badlogic/pi-mono/blob/main/packages/coding-agent/CHANGELOG.md#0681---2026-04-22) — no extension needed. Set `FIREWORKS_API_KEY` to use it directly.
+**Note:** Fireworks and NVIDIA NIM are now [built-in Pi providers](https://github.com/badlogic/pi-mono/blob/main/packages/coding-agent/CHANGELOG.md#0681---2026-04-22) — no extension needed. Set `FIREWORKS_API_KEY` or `NVIDIA_API_KEY` to use them directly.
 
 ### 3. Toggle between free and paid models
 
@@ -87,8 +90,8 @@ Want to see paid models too? Run the toggle command for your provider:
 ```
 /toggle-kilo       # Toggle Kilo (✅ offers free models)
 /toggle-openrouter # Toggle OpenRouter (✅ offers free models)
+/toggle-opencode   # Toggle OpenCode (✅ offers free models)
 /toggle-cline      # Toggle Cline (✅ offers free models)
-/toggle-nvidia     # Toggle NVIDIA (🔄 freemium)
 /toggle-ollama     # Toggle Ollama Cloud (🔄 freemium)
 /toggle-mistral    # Toggle Mistral (🔧 dynamic - needs API key)
 /toggle-groq       # Toggle Groq (🔧 dynamic - needs API key)
@@ -97,7 +100,7 @@ Want to see paid models too? Run the toggle command for your provider:
 /toggle-huggingface # Toggle Hugging Face (🔧 dynamic - needs HF_TOKEN)
 /toggle-zenmux    # Toggle ZenMux (💳 paid - needs API key with credits)
 /toggle-crofai    # Toggle CrofAI (💳 paid - needs API key with credits)
-/toggle-codestral # Toggle Codestral (💳 paid - free Experiment plan)
+/toggle-codestral # Toggle Codestral (🔄 free Experiment plan)
 /toggle-deepinfra # Toggle DeepInfra (💳 trial credit provider)
 /toggle-together  # Toggle Together AI (💳 trial credit provider)
 /toggle-sambanova # Toggle SambaNova (🔄 freemium)
@@ -105,6 +108,7 @@ Want to see paid models too? Run the toggle command for your provider:
 /toggle-novita    # Toggle Novita AI (💳 paid — 3 free models)
 /toggle-routeway  # Toggle Routeway AI (💳 paid — has :free models)
 /toggle-fastrouter # Toggle FastRouter (🔧 dynamic — always discovered)
+/toggle-tokenrouter # Toggle TokenRouter (💳 paid — 1 free model)
 ```
 
 **Notes:**
@@ -126,7 +130,6 @@ Add your API keys to this file:
 
 ```json
 {
-  "nvidia_api_key": "nvapi-...",
   "ollama_api_key": "...",
   "mistral_api_key": "...",
   "codestral_api_key": "...",
@@ -139,7 +142,7 @@ Add your API keys to this file:
 }
 ```
 
-Or set environment variables instead (same names, uppercase: `OPENROUTER_API_KEY`, `NVIDIA_API_KEY`, etc.)
+Or set environment variables instead (same names, uppercase: `OPENROUTER_API_KEY`, `OLLAMA_API_KEY`, etc.)
 
 If `~/.pi/free.json` contains invalid JSON, pi-free now logs the parse error to `~/.pi/free.log` so you can fix the file quickly.
 
@@ -147,27 +150,35 @@ See the [Providers That Need Authentication](#providers-that-need-authentication
 
 ### 5. Quick commands reference
 
-| Command              | What it does                                              |
-| -------------------- | --------------------------------------------------------- |
-| `/toggle-{provider}` | Switch between free-only and all models for that provider |
-| `/toggle-free`       | Toggle global free-only mode for ALL providers            |
-| `/free-providers`    | Show free/paid model counts for all providers             |
-| `/login kilo`        | Start OAuth flow for Kilo                                 |
-| `/login cline`       | Start OAuth flow for Cline                                |
-| `/logout kilo`       | Clear Kilo OAuth credentials                              |
-| `/logout cline`      | Clear Cline OAuth credentials                             |
+| Command                    | What it does                                              |
+| -------------------------- | --------------------------------------------------------- |
+| `/toggle-{provider}`       | Switch between free-only and all models for that provider |
+| `/toggle-free`             | Toggle global free-only mode for ALL providers            |
+| `/free-providers`          | Show free/paid model counts for all providers             |
+| `/free-telemetry`          | Show real-world performance data (tokens/s, latency, success rate) for free models |
+| `/clear-free-telemetry`    | Clear all stored telemetry data                           |
+| `/login kilo`              | Start OAuth flow for Kilo                                 |
+| `/login cline`             | Start OAuth flow for Cline                                |
+| `/logout kilo`             | Clear Kilo OAuth credentials                              |
+| `/logout cline`            | Clear Cline OAuth credentials                             |
 
 ---
 
 ## Features
 
-### 🔍 NVIDIA: Pre-Filtering + 404 Detection
+### 🔍 Model Availability Probing
+
+Some provider APIs list models that return errors when you try to use them (expired free promotions, decommissioned models, server spin-down). pi-free automatically detects and hides broken models:
 
-NVIDIA's API lists 130+ models, but 57+ return 404 "Function not found" when you try to use them. pi-free solves this:
+- **Ollama Cloud**: `/probe-ollama` — probes for 403 errors, auto-hides inaccessible models
+- **Routeway**: `/probe-routeway` — probes for 5xx/404 errors, auto-hides broken models
+- **OpenCode**: `/probe-opencode`, `/probe-opencode-go` — probes for expired free promotions (reports only, no auto-hide)
+- **DeepInfra**: `/probe-deepinfra` — probes for 404/5xx errors, auto-hides broken models
+- **SambaNova**: `/probe-sambanova` — probes for 404/5xx errors, auto-hides broken models
+- **Together AI**: `/probe-together` — probes for 404/5xx errors, auto-hides broken models
+- **Novita AI**: `/probe-novita` — probes for 404/5xx errors, auto-hides broken models
 
-- **57 known 404s hard-filtered** — Discontinued models (`dbrx-instruct`, `codellama-70b`), embedding models mislabeled as chat-capable (`nv-embed-*`), and stale catalog entries are silently excluded
-- **Auto-discovery from NVIDIA's API** — Queries `integrate.api.nvidia.com/v1/models` directly for the ground-truth list
-- **`/probe-nvidia` command** — On-demand health check: tests every model with a minimal request, auto-hides new 404s, and re-registers immediately
+All probes use a **24-hour probe cache** to avoid re-checking recently-verified models. Run any probe command manually to force a full re-check, or let the lazy auto-probe on first `session_start` handle it.
 
 ### 🎯 Coding Index (CI) Scores
 
@@ -175,13 +186,13 @@ Every model shows a **Coding Index score** (e.g., `CI: 52.3`) in the model picke
 
 - **Benchmark-based** — Scores derived from Artificial Analysis coding benchmarks (HumanEval, MBPP, etc.)
 - **Quality indicator** — Higher scores = better coding performance
-- **All providers** — Applied to every model from every provider (NVIDIA, Mistral, Groq, etc.)
+- **All providers** — Applied to every model from every provider (Mistral, Groq, Cerebras, etc.)
 
 **Missing CI scores?** Provider model IDs often don't match benchmark database keys exactly. pi-free applies provider-specific normalization to improve matching:
 
 | Provider     | Normalization Applied                                              |
 | ------------ | ------------------------------------------------------------------ |
-| **NVIDIA**   | Strips vendor prefixes (`meta/`, `mistralai/`, `microsoft/`, etc.) |
+| **Fireworks**   | Strips vendor prefixes (`meta/`, `mistralai/`, `microsoft/`, etc.) |
 | **Groq**     | Removes `-versatile` and numeric suffixes (`-32768`)               |
 | **Cerebras** | Normalizes `llama3.1` → `llama-3.1`, adds `instruct` suffix        |
 | **Mistral**  | Strips `-latest` suffix                                            |
@@ -194,6 +205,7 @@ Every model shows a **Coding Index score** (e.g., `CI: 52.3`) in the model picke
 Providers have different pricing models. pi-free handles them all:
 
 - **Free-only by default** — Shows only zero-cost models initially
+- **Auto-probe on session_start** — Lazy background probes detect broken models automatically on your first session; no manual command needed
 - **Per-provider toggles** — Run `/toggle-{provider}` to switch between "free only" vs "all models"
 - **Persists across sessions** — Your preference is saved to `~/.pi/free.json`
 - **Instant updates** — Changes apply immediately; no Pi restart needed
@@ -201,7 +213,7 @@ Providers have different pricing models. pi-free handles them all:
 **Provider types:**
 
 - ✅ **Free providers** (Kilo, Cline) — Toggle between free-only vs paid models
-- 🔄 **Freemium** (NVIDIA, Ollama) — Free tier with limits, toggle shows all
+- 🔄 **Freemium** (Ollama, SambaNova) — Free tier with limits, toggle shows all
 - 🔧 **Dynamic API** (Mistral, Groq, Cerebras, xAI) — Fetched when API key configured, toggle filters the list
 
 ### 🔐 OAuth + API Key Handling
@@ -283,44 +295,9 @@ Then use `/toggle-openrouter` to switch between free-only and all models.
 
 **Note:** `openrouter_api_key` in `~/.pi/free.json` is ignored. OpenRouter always reads from Pi's auth system to avoid stale keys.
 
-### NVIDIA NIM (Free Credits System)
-
-NVIDIA provides **free monthly credits** (1000 requests/month) at [build.nvidia.com](https://build.nvidia.com).
-
-**Important:** Models have different "costs" per token:
-
-- **Zero-cost models**: Don't consume your credit balance (shown by default)
-- **Credit-costing models**: Consume credits faster (hidden by default)
-
-Get your API key and optionally enable all models:
-
-**Option A: Show only free models (default)**
-
-```bash
-export NVIDIA_API_KEY="nvapi-..."
-```
-
-Uses only zero-cost models → your 1000 credits last the full month
+### NVIDIA NIM (now built-in)
 
-**Option B: Show all models (uses credits faster)**
-
-```bash
-export NVIDIA_API_KEY="nvapi-..."
-export NVIDIA_SHOW_PAID=true
-```
-
-Or in `~/.pi/free.json`:
-
-```json
-{
-  "nvidia_api_key": "nvapi-...",
-  "nvidia_show_paid": true
-}
-```
-
-Toggle anytime with `/toggle-nvidia`
-
-**Models available:** Llama 4/3.x, Mistral Small 3.1, DeepSeek R1, Gemma 4, Kimi K2.5/2.6, Qwen 3/2.5, OpenAI GPT-OSS, and more.
+NVIDIA NIM is now a **built-in Pi provider** — no extension needed. Set `NVIDIA_API_KEY` to use it directly with Pi's built-in model picker.
 
 ### Ollama Cloud
 
@@ -437,21 +414,25 @@ Each provider has toggle commands to switch between free and all models:
 | ----------------------- | -------------------------------------------------------- |
 | `/toggle-kilo`          | Toggle between free/all Kilo models                      |
 | `/toggle-openrouter`    | Toggle between free/all OpenRouter models                |
+| `/toggle-opencode`      | Toggle between free/all OpenCode models                  |
 | `/toggle-cline`         | Toggle between free/all Cline models                     |
-| `/toggle-nvidia`        | Toggle between free/all NVIDIA models                    |
 | `/toggle-ollama`        | Toggle between free/all Ollama Cloud models              |
 | `/toggle-mistral`       | Toggle between free/all Mistral models (🔧 dynamic)      |
 | `/toggle-groq`          | Toggle between free/all Groq models (🔧 dynamic)         |
 | `/toggle-cerebras`      | Toggle between free/all Cerebras models (🔧 dynamic)     |
 | `/toggle-xai`           | Toggle between free/all xAI models (🔧 dynamic)          |
-| `/toggle-huggingface`   | Toggle between free/all Hugging Face models (🔧 dynamic) |
-| `/toggle-codestral`     | Toggle Codestral (💳 paid)                               |
+| `/toggle-huggingface`   | Toggle between free/all Hugging Face models (🔧 dynamic)  |
+| `/toggle-codestral`     | Toggle Codestral (🔄 freemium)                           |
 | `/toggle-deepinfra`     | Toggle DeepInfra (💳 trial credit)                       |
 | `/toggle-together`      | Toggle Together AI (💳 trial credit)                     |
 | `/toggle-sambanova`     | Toggle SambaNova (🔄 freemium)                           |
 | `/toggle-llm7`          | Toggle LLM7 (✅ free gateway)                            |
 | `/toggle-zenmux`        | Toggle ZenMux (💳 paid)                                  |
 | `/toggle-crofai`        | Toggle CrofAI (💳 paid)                                  |
+| `/toggle-novita`        | Toggle Novita AI (💳 paid)                               |
+| `/toggle-routeway`      | Toggle Routeway AI (💳 paid)                             |
+| `/toggle-fastrouter`    | Toggle FastRouter (🔧 dynamic)                           |
+| `/toggle-tokenrouter`   | Toggle TokenRouter (💳 paid — 1 free model)              |
 | `/ollama-cloud-refresh` | Re-fetch Ollama Cloud models live (no restart needed)    |
 | `/probe-ollama`         | Test Ollama Cloud models for 403 errors (auto-hide)      |
 
@@ -465,12 +446,18 @@ Each provider has toggle commands to switch between free and all models:
 
 ### Probe Commands (Health Check)
 
-Test models for 404/403 errors and auto-hide broken ones:
+Test models for errors and auto-hide broken ones:
 
-| Command         | What it does                                                |
-| --------------- | ----------------------------------------------------------- |
-| `/probe-nvidia` | Test all NVIDIA models, auto-hide 404s in `~/.pi/free.json` |
-| `/probe-ollama` | Test all Ollama models, auto-hide 403s in `~/.pi/free.json` |
+| Command                | What it does                                                |
+| ---------------------- | ----------------------------------------------------------- |
+| `/probe-ollama`        | Test all Ollama Cloud models, auto-hide 403s                 |
+| `/probe-routeway`      | Test all Routeway models, auto-hide 5xx/404s                |
+| `/probe-opencode`      | Test all OpenCode free models for expired promotions        |
+| `/probe-opencode-go`   | Test all OpenCode Go free models for expired promotions     |
+| `/probe-deepinfra`     | Test all DeepInfra models for availability, auto-hide broken|
+| `/probe-sambanova`     | Test all SambaNova models for availability, auto-hide broken|
+| `/probe-together`      | Test all Together AI models for availability, auto-hide broken|
+| `/probe-novita`        | Test all Novita AI models for availability, auto-hide broken|
 
 **How it works:**
 
@@ -490,7 +477,6 @@ Create `~/.pi/free.json` in your home directory:
 
 ```json
 {
-  "nvidia_api_key": "YOUR_NVIDIA_KEY",
   "mistral_api_key": "YOUR_MISTRAL_KEY",
   "ollama_api_key": "YOUR_OLLAMA_KEY",
   "ollama_show_paid": true,
@@ -501,8 +487,8 @@ Create `~/.pi/free.json` in your home directory:
 Or use environment variables (same names, uppercase):
 
 ```bash
-export NVIDIA_API_KEY="..."
 export MISTRAL_API_KEY="..."
+export OLLAMA_API_KEY="..."
 ```
 
 ---

package/banner.svg

@@ -23,14 +23,6 @@
       <stop offset="0%" stop-color="#06b6d4" stop-opacity="0.15"/>
       <stop offset="100%" stop-color="#06b6d4" stop-opacity="0.05"/>
     </linearGradient>
-    <linearGradient id="card3" x1="0" y1="0" x2="0" y2="1">
-      <stop offset="0%" stop-color="#10b981" stop-opacity="0.15"/>
-      <stop offset="100%" stop-color="#10b981" stop-opacity="0.05"/>
-    </linearGradient>
-    <linearGradient id="card4" x1="0" y1="0" x2="0" y2="1">
-      <stop offset="0%" stop-color="#f59e0b" stop-opacity="0.15"/>
-      <stop offset="100%" stop-color="#f59e0b" stop-opacity="0.05"/>
-    </linearGradient>
     <filter id="glow">
       <feGaussianBlur stdDeviation="3" result="blur"/>
       <feMerge>
@@ -74,61 +66,54 @@
     Free &amp; Paid AI Model Providers for Pi
   </text>
 
-  <!-- Subtitle -->
-  <text x="160" y="178" font-family="system-ui, -apple-system, sans-serif" font-size="13" fill="#7c8a9a" letter-spacing="0.02em">
-    </text>
-
-  <!-- Provider badges -->
+  <!-- Provider category badges -->
   <g transform="translate(160, 200)">
-    <!-- Free -->
     <rect x="0" y="0" width="90" height="28" rx="6" fill="#10b981" opacity="0.15" stroke="#10b981" stroke-width="0.5" stroke-opacity="0.3"/>
-    <text x="45" y="18" font-family="system-ui, sans-serif" font-size="11" fill="#6ee7b7" text-anchor="middle" font-weight="600">✅ Free</text>
+    <text x="45" y="18" font-family="system-ui, sans-serif" font-size="11" fill="#6ee7b7" text-anchor="middle" font-weight="600">Free</text>
   </g>
 
   <g transform="translate(260, 200)">
     <rect x="0" y="0" width="100" height="28" rx="6" fill="#f59e0b" opacity="0.15" stroke="#f59e0b" stroke-width="0.5" stroke-opacity="0.3"/>
-    <text x="50" y="18" font-family="system-ui, sans-serif" font-size="11" fill="#fcd34d" text-anchor="middle" font-weight="600">🔄 Freemium</text>
+    <text x="50" y="18" font-family="system-ui, sans-serif" font-size="11" fill="#fcd34d" text-anchor="middle" font-weight="600">Freemium</text>
   </g>
 
   <g transform="translate(370, 200)">
-    <rect x="0" y="0" width="90" height="28" rx="6" fill="#7c3aed" opacity="0.15" stroke="#7c3aed" stroke-width="0.5" stroke-opacity="0.3"/>
-    <text x="45" y="18" font-family="system-ui, sans-serif" font-size="11" fill="#a78bfa" text-anchor="middle" font-weight="600">🔧 Dynamic</text>
+    <rect x="0" y="0" width="100" height="28" rx="6" fill="#7c3aed" opacity="0.15" stroke="#7c3aed" stroke-width="0.5" stroke-opacity="0.3"/>
+    <text x="50" y="18" font-family="system-ui, sans-serif" font-size="11" fill="#a78bfa" text-anchor="middle" font-weight="600">Dynamic</text>
   </g>
 
-  <g transform="translate(470, 200)">
+  <g transform="translate(480, 200)">
     <rect x="0" y="0" width="90" height="28" rx="6" fill="#ef4444" opacity="0.15" stroke="#ef4444" stroke-width="0.5" stroke-opacity="0.3"/>
-    <text x="45" y="18" font-family="system-ui, sans-serif" font-size="11" fill="#fca5a5" text-anchor="middle" font-weight="600">💳 Paid</text>
+    <text x="45" y="18" font-family="system-ui, sans-serif" font-size="11" fill="#fca5a5" text-anchor="middle" font-weight="600">Paid</text>
   </g>
 
   <!-- Provider cards (right side) -->
   <g transform="translate(700, 55)">
-    <!-- Card 1: Custom Providers -->
     <rect x="0" y="0" width="250" height="170" rx="12" fill="url(#card1)" stroke="#7c3aed" stroke-width="0.5" stroke-opacity="0.2" filter="url(#shadow)"/>
     <rect x="0" y="0" width="250" height="170" rx="12" fill="none" stroke="url(#accent)" stroke-width="0.5" opacity="0.1"/>
-    <text x="20" y="28" font-family="system-ui, sans-serif" font-size="13" font-weight="700" fill="#c4b5fd">Custom Providers</text>
-    <text x="20" y="50" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">Kilo · Cline · NVIDIA</text>
-    <text x="20" y="70" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">Ollama Cloud · ZenMux</text>
-    <text x="20" y="90" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">CrofAI · Codestral · LLM7</text>
-    <text x="20" y="110" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">DeepInfra · SambaNova</text>
-    <text x="20" y="130" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">Together · Novita · FastRouter</text>
+    <text x="20" y="28" font-family="system-ui, sans-serif" font-size="13" font-weight="700" fill="#c4b5fd">Providers</text>
+    <text x="20" y="50" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">Kilo · Cline · OpenCode</text>
+    <text x="20" y="70" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">Ollama Cloud · SambaNova</text>
+    <text x="20" y="90" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">Codestral · LLM7 · Novita</text>
+    <text x="20" y="110" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">ZenMux · CrofAI · DeepInfra</text>
+    <text x="20" y="130" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">Together · TokenRouter · FastRouter</text>
   </g>
 
   <g transform="translate(970, 55)">
-    <!-- Card 2: Features -->
     <rect x="0" y="0" width="250" height="170" rx="12" fill="url(#card2)" stroke="#06b6d4" stroke-width="0.5" stroke-opacity="0.2" filter="url(#shadow)"/>
     <rect x="0" y="0" width="250" height="170" rx="12" fill="none" stroke="url(#accent2)" stroke-width="0.5" opacity="0.1"/>
     <text x="20" y="28" font-family="system-ui, sans-serif" font-size="13" font-weight="700" fill="#67e8f9">Features</text>
-    <text x="20" y="50" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">✦ Free model auto-detection</text>
-    <text x="20" y="68" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">✦ Per-provider toggles</text>
-    <text x="20" y="86" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">✦ OAuth flows (Kilo, Cline)</text>
-    <text x="20" y="104" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">✦ Coding Index (CI) scores</text>
-    <text x="20" y="122" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">✦ Model health probes</text>
-    <text x="20" y="140" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">✦ Thinking level maps</text>
-    <text x="20" y="158" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">✦ 404/403 auto-hide on probe</text>
+    <text x="20" y="50" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">Free model auto-detection</text>
+    <text x="20" y="68" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">Per-provider toggles</text>
+    <text x="20" y="86" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">OAuth flows (Kilo, Cline)</text>
+    <text x="20" y="104" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">Coding Index (CI) scores</text>
+    <text x="20" y="122" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">Model health probes</text>
+    <text x="20" y="140" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">Thinking level maps</text>
+    <text x="20" y="158" font-family="system-ui, sans-serif" font-size="11" fill="#8b9aaa">404/403 auto-hide on probe</text>
   </g>
 
   <!-- Bottom tagline -->
   <text x="640" y="305" font-family="system-ui, -apple-system, sans-serif" font-size="12" fill="#5a6a7a" text-anchor="middle" letter-spacing="0.04em">
-    npx pi install git:github.com/apmantza/pi-free
+    pi install git:github.com/apmantza/pi-free
   </text>
 </svg>