# pi-requesty-provider [![npm version](https://img.shields.io/npm/v/pi-requesty-provider?color=blue&logo=npm)](https://www.npmjs.com/package/pi-requesty-provider) [![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](https://opensource.org/licenses/MIT) [![pi Extension](https://img.shields.io/badge/pi-extension-purple.svg)](https://pi.dev) > **Unlock 300+ frontier & open-source LLMs in [pi](https://pi.dev) with native tool calling, smart reasoning, real-time pricing, and built-in web search.** `pi-requesty-provider` seamlessly bridges [Requesty](https://requesty.ai) — the next-generation LLM router and AI gateway — directly into the **pi** coding agent. Experience Claude Fable 5, OpenAI GPT-5.6 Sol, xAI Grok 4.6, Gemini 3.7 Flash, Kimi K3, Qwen3.8 Max, DeepSeek V4 Pro, GLM-5, and hundreds more under a single unified provider. --- ## ⚡ Superpowers & Features - 🌐 **Native Web Search Integration** — Browse the live web during agent execution without external tools. Toggle dynamically with `/requesty search native on/off` (powered by Requesty's server-side web grounding). - 🛡️ **100% Agent-Ready (Smart Tool Filtering)** — Automatically filters out models lacking function/tool calling support. Every model in your picker is guaranteed to work reliably with `pi` tools. - 🧠 **Deep Reasoning & Vision Support** — Automatic detection of thinking budgets, reasoning effort, and multimodal vision capabilities (`["text", "image"]`). - 💰 **Real-Time Cost & Cache Tracking** — Precise per-million-token cost mapping, including tiered pricing, prompt caching read/write rates, and context limits. - ⌨️ **Extensible CLI & Multi-Level Autocomplete** — Unified `/requesty` command router with dynamic categorized help, interactive nested tab-completion, and status monitoring. - 🔐 **Native Auth & Multi-Region** — Standard `/login requesty` flow storing credentials securely in `~/.pi/agent/auth.json`, supporting US and EU endpoints. --- ## 🚀 Quick Installation ### From npm (Recommended) ```bash pi install npm:pi-requesty-provider ``` ### From GitHub ```bash pi install git:github.com/xinaps-dev/pi-requesty-provider ``` ### Local Development ```bash # Clone and build git clone https://github.com/xinaps-dev/pi-requesty-provider.git cd pi-requesty-provider pnpm install pnpm build # Run locally in pi pi -e ./ ``` --- ## 🎯 Getting Started ### 1. Authenticate Run the native login command inside `pi`: ``` /login requesty ``` 1. Enter your API key (get one at [app.requesty.ai/api-keys](https://app.requesty.ai/api-keys)). 2. Select your regional endpoint: - **US (Global)**: `https://router.requesty.ai/v1` (Default) - **EU (Frankfurt)**: `https://router.eu.requesty.ai/v1` (GDPR compliant endpoint) ### 2. Pick Any Model Open the model selector: ``` /model ``` All models are listed as `requesty/` (e.g. `requesty/anthropic/claude-3-7-sonnet`, `requesty/deepseek/deepseek-r1`). Type to filter with fuzzy search. ### 3. Enable Native Web Search (Optional) Give your models real-time web access: ``` /requesty search native on ``` --- ## 🛠️ Command Reference The `/requesty` command comes with full autocomplete and categorized subcommands: | Command | Category | Description | |---|---|---| | `/requesty` | General | Show interactive help and subcommand index | | `/requesty sync` | Provider | Fetch and refresh the active model catalog from Requesty API | | `/requesty status` | Provider | Display connection status, active model, endpoints, and search support | | `/requesty search native on` | Search | Enable Requesty native web search grounding (persisted) | | `/requesty search native off` | Search | Disable Requesty native web search grounding | | `/requesty search native status` | Search | Show web search configuration and model capability | --- ## ⚙️ Environment Variables Prefer configuring via environment variables? Simply export: | Variable | Description | Default | |---|---|---| | `REQUESTY_API_KEY` | Your Requesty API key | _(required if not logged in)_ | | `REQUESTY_BASE_URL` | Custom gateway endpoint | `https://router.requesty.ai/v1` | --- ## 🔍 Model Metadata Mapping The provider parses Requesty's model catalog into `pi` native capabilities: | Requesty Attribute | pi Model Feature | Impact | |---|---|---| | `supports_reasoning` | `reasoning: true`, `compat.supportsReasoningEffort` | Enables thinking/reasoning blocks | | `supports_vision` | `input: ["text", "image"]` | Allows image and multimodal inputs | | `supports_tools` / `supports_function_calling` | Catalog filter | Ensures 100% agent execution compatibility | | `supports_web_search` | Model capability flag | Controls native web search injection | | `context_window` & `max_output_tokens` | `contextWindow`, `maxTokens` | Optimizes context management and limits | | `pricing[]` / `input_price` / `caching_price` | `cost.input`, `cost.output`, `cost.tiers` | Accurate prompt and cache token cost tracking | --- ## 🏗️ Architecture ``` ┌────────────────────────────────────────────────────────────────────────┐ │ pi CLI / TUI Runtime │ │ │ │ /login requesty /requesty [sync|status|search] /model requesty │ │ │ │ │ │ │ ▼ ▼ ▼ │ │ ┌───────────────────────────────────────────────────────────────┐ │ │ │ pi-requesty-provider Extension │ │ │ │ │ │ │ │ ┌──────────────┐ ┌─────────────────┐ ┌──────────────┐ │ │ │ │ │ AuthEngine │ │ CommandRouter │ │ CatalogSync │ │ │ │ │ │ (ApiKeyAuth) │ │ (Multi-level AC)│ │(Model Parser)│ │ │ │ │ └──────────────┘ └─────────────────┘ └──────────────┘ │ │ │ │ │ │ │ │ │ ┌────────────────┴────────────────┐ │ │ │ │ │ before_provider_request Hook │ │ │ │ │ │ (Native Web Search Injection) │ │ │ │ │ └────────────────┬────────────────┘ │ │ │ └───────────────────────────────┼───────────────────────────────┘ │ │ │ │ └───────────────────────────────────┼────────────────────────────────────┘ ▼ ┌──────────────────────────────┐ │ Requesty Router Gateway │ │ https://router.requesty.ai │ ├──────────────────────────────┤ │ GET /v1/models │ │ POST /v1/chat/completions │ └──────────────────────────────┘ ▼ ~/.pi/agent/auth.json │ ~/.pi/agent/requesty.json ``` --- ## 💻 Development & Contributing ### Prerequisites - Node.js 20+ - pnpm 8+ ### Scripts ```bash pnpm install # Install dependencies pnpm typecheck # Run type checking pnpm test # Run Vitest test suite ``` ### Project Layout ``` pi-requesty-provider/ ├── index.ts # Extension entry point & lifecycle hooks ├── src/ │ ├── commands.ts # Unified hierarchical /requesty command router │ ├── types.ts # Global extension types & definitions │ ├── constants.ts # Extension-level constants │ ├── provider/ # Core Provider module │ │ ├── index.ts # Provider factory and model caching │ │ ├── auth.ts # ApiKeyAuth implementation │ │ ├── client.ts # Requesty REST client │ │ ├── models.ts # Model transformer & capability mapper │ │ ├── commands.ts # sync & status subcommand handlers │ │ └── types.ts # Requesty API & Model types │ └── search/ # Native Web Search module │ ├── index.ts # Search module exports │ ├── config.ts # Config store (~/.pi/agent/requesty.json) │ ├── interceptor.ts # before_provider_request payload interceptor │ └── commands.ts # /requesty search native subcommands └── tests/ # Unit and integration tests ``` --- ## 📄 License MIT © [xinaps](https://github.com/xinaps-dev)