PHPackages andrecorugda/ai-openrouter-gateway - PHPackages - PHPackages [Skip to content](#main-content)[PHPackages](/)[Directory](/)[Categories](/categories)[Trending](/trending)[Leaderboard](/leaderboard)[Changelog](/changelog)[Analyze](/analyze)[Collections](/collections)[Log in](/login)[Sign up](/register) 1. [Directory](/) 2. / 3. [API Development](/categories/api) 4. / 5. andrecorugda/ai-openrouter-gateway ActiveLibrary[API Development](/categories/api) andrecorugda/ai-openrouter-gateway ================================== A self-hostable, OpenRouter-backed AI gateway for Laravel — named integrations, versioned prompts, telemetry, rate & cost limiting, a Sanctum HTTP API, and a Filament admin UI with an AI-assisted prompt builder. v2.0.2(today)043↑2900%MITPHPPHP ^8.2CI passing Since Jun 27Pushed todayCompare [ Source](https://github.com/andrecorugda/ai-openrouter-gateway)[ Packagist](https://packagist.org/packages/andrecorugda/ai-openrouter-gateway)[ Docs](https://github.com/andrecorugda/ai-openrouter-gateway)[ RSS](/packages/andrecorugda-ai-openrouter-gateway/feed)WikiDiscussions develop Synced today READMEChangelog (5)Dependencies (12)Versions (27)Used By (0) AI OpenRouter Gateway for Laravel ================================= [](#ai-openrouter-gateway-for-laravel) [![AI OpenRouter Gateway — one OpenRouter key for every model, managed in Filament](art/cover.jpg)](art/cover.jpg) [![Latest Version on Packagist](https://camo.githubusercontent.com/bd513fba7a6da5372d1eac076dc685040bc48f17d21dc6e74073331413fd3219/68747470733a2f2f696d672e736869656c64732e696f2f7061636b61676973742f762f616e647265636f72756764612f61692d6f70656e726f757465722d676174657761792e7376673f7374796c653d666c61742d737175617265)](https://packagist.org/packages/andrecorugda/ai-openrouter-gateway)[![Total Downloads](https://camo.githubusercontent.com/57ce94fb1c786cc8ce66cce0f3182c6a75d5f6cac9a21c1566f91de57e6c82ec/68747470733a2f2f696d672e736869656c64732e696f2f7061636b61676973742f64742f616e647265636f72756764612f61692d6f70656e726f757465722d676174657761792e7376673f7374796c653d666c61742d737175617265)](https://packagist.org/packages/andrecorugda/ai-openrouter-gateway)[![Tests](https://camo.githubusercontent.com/2b1b108415f2e4f9847bef26fe5fcd5e8d8e21aa3e645dce659264fb83cce715/68747470733a2f2f696d672e736869656c64732e696f2f6769746875622f616374696f6e732f776f726b666c6f772f7374617475732f616e647265636f72756764612f61692d6f70656e726f757465722d676174657761792f74657374732e796d6c3f6272616e63683d6d61696e266c6162656c3d7465737473267374796c653d666c61742d737175617265)](https://github.com/andrecorugda/ai-openrouter-gateway/actions)[![PHP Version](https://camo.githubusercontent.com/efc059a31d092806ca3ab5cff89b4e1079b10065f9a73063228a76ff09aad731/68747470733a2f2f696d672e736869656c64732e696f2f7061636b61676973742f7068702d762f616e647265636f72756764612f61692d6f70656e726f757465722d676174657761792e7376673f7374796c653d666c61742d737175617265)](https://packagist.org/packages/andrecorugda/ai-openrouter-gateway)[![License](https://camo.githubusercontent.com/40e5f0136aadf9ec84ec0c1b9b0282429720f39a25af91cd1a237a912a34e2cc/68747470733a2f2f696d672e736869656c64732e696f2f7061636b61676973742f6c2f616e647265636f72756764612f61692d6f70656e726f757465722d676174657761792e7376673f7374796c653d666c61742d737175617265)](LICENSE) **Manage every AI feature in your Laravel app as a versioned, runtime-tunable integration — one OpenRouter key for every model, behind one service, one audit log, and one cost view.** Stop hardcoding prompts and model ids. Define a use case once, then tune its prompt, model, and parameters from a bundled **Filament** admin UI — no code change, no redeploy. Call it from PHP or over an authenticated HTTP API, with multi-turn conversations, rate limits, daily cost caps, and per-call telemetry built in. ``` use Andre\AiGateway\Facades\AiGateway; $result = AiGateway::invoke('lead_summary', ['company' => 'Acme Corp']); $result->text; // the model's reply $result->model_used; // 'anthropic/claude-sonnet-4' $result->cost_usd; // 0.0004 ``` Swap `anthropic/claude-sonnet-4` for `openai/gpt-4o` or `google/gemini-2.5-pro` from a dropdown — every model OpenRouter offers, no code change. > Your prompts, customer data, and OpenRouter key never leave your app — there's no third-party SaaS in the trust boundary. --- Why? ---- [](#why) - **No hardcoded prompts.** Prompt, variables, model, and params live in the database and the admin UI — not your source tree. - **Tune in production, instantly.** Edit and save; the next call uses it. Every save mints a new **version** you can roll back to. - **Test across models in seconds.** Pick any model from the live catalog, hit **Test**, and compare output, tokens, latency, and cost — then promote the better or cheaper one. - **One use case, every caller.** Invoke it from any PHP service *and* any external app or language over HTTPS — one source of truth, reused across platforms. - **Spend you can see and cap.** Per-call cost / token / latency telemetry, plus per-integration rate limits and daily budgets. --- How it works ------------ [](#how-it-works) ### Who uses it (use cases) [](#who-uses-it-use-cases) [![Use case diagram](screenshots/usecases.png)](screenshots/usecases.png) ### Architecture [](#architecture) [![Architecture diagram](screenshots/architecture.png)](screenshots/architecture.png) **Request flow (one call):** resolve the integration's active version (cached) → render the prompt template with the caller's args → enforce rate + daily-cost limits → compose the OpenRouter payload (model\[s\], params, server tools, optional cache markers) → call OpenRouter → write an `ai_invocations` telemetry row (cost / tokens / latency / status) → return a typed `AiResult`. Conversational calls additionally load and persist thread turns via `ConversationStore`. --- Features -------- [](#features) - 🔌 **One key, every model** — OpenRouter under the hood; switch models per-integration with no code change. - 🧩 **Named integrations** — register a use case once, invoke it by slug everywhere. - 🗂️ **Versioned prompts** — every edit mints a new version; activate/roll back without losing history. - 🧮 **Telemetry built in** — tokens, cost, latency, model, and status for every call in `ai_invocations`. - 🚦 **Rate limiting** — per-integration, per-caller, per-minute. - 💰 **Cost limiting** — per-integration daily USD budget, enforced before each call. - 🌐 **HTTP API** — `POST /api/ai/{integration}/chat`, Sanctum-authenticated, toggleable at runtime. - 💬 **Conversation threads** — opt-in multi-turn memory with `/start` + `/converse`, per-caller ownership, TTL expiry, and a prune command. - 📖 **Live API docs** — an OpenAPI 3 spec + interactive Scalar "try it" page generated from your integrations. - 🔑 **API token management** — mint and revoke scoped tokens from the admin UI. - 🔎 **OpenRouter server tools** — per-version `web_search` / `web_fetch`. - ✨ **AI prompt builder** — describe what you want; a fast Haiku drafts the template + variables. - 🗂️ **Live model catalog** — searchable model picker from OpenRouter's `/models`, with per-model generation params and caching eligibility. - 📊 **Invocations browser** — read-only telemetry with status/caller/date filters, cost + token Σ summaries, and per-call detail. - 🎛️ **Filament admin UI** — integration CRUD, versions (load-into-form), a live test panel, an interactive prompt editor, settings. - ⚙️ **Fully configurable** — connection, table names, route prefix/middleware, cache store, limits, models. Requirements ------------ [](#requirements) - PHP 8.2+ - Laravel 11, 12, or 13 - An [OpenRouter](https://openrouter.ai) API key - **Filament 4 or 5** for the admin UI *(optional)* — see version compatibility below Version compatibility --------------------- [](#version-compatibility) PackageFilament`^2.0`**4.x / 5.x**`^1.0`**3.x**Composer installs the right line for your Filament version automatically — `composer require andrecorugda/ai-openrouter-gateway`. Installation ------------ [](#installation) ``` composer require andrecorugda/ai-openrouter-gateway ``` Publish and run the migrations: ``` php artisan vendor:publish --tag="ai-openrouter-gateway-migrations" php artisan migrate ``` Optionally publish the config: ``` php artisan vendor:publish --tag="ai-openrouter-gateway-config" ``` Add your key to `.env`: ``` OPENROUTER_API_KEY=sk-or-v1-... ``` That's the only required variable — referer/title default to your `APP_URL` / `APP_NAME`. Quickstart ---------- [](#quickstart) ### 1. Create an integration [](#1-create-an-integration) Either through the Filament UI (recommended) or in code: ``` use Andre\AiGateway\Models\AiIntegration; use Andre\AiGateway\Services\AiIntegrationService; $integration = AiIntegration::create([ 'slug' => 'expense_extract', 'name' => 'Expense Extractor', 'visibility' => 'internal', ]); app(AiIntegrationService::class)->saveVersion($integration, [ 'system_prompt' => 'Extract the merchant, total, and date from this receipt:\n\n{{receipt_text}}', 'models' => ['anthropic/claude-sonnet-4', 'openai/gpt-4o'], // primary + fallback 'default_params' => ['max_tokens' => 512, 'temperature' => 0.1], 'prompt_args' => [ ['name' => 'receipt_text', 'type' => 'string', 'required' => true], ], ]); ``` ### 2. Invoke it from PHP [](#2-invoke-it-from-php) ``` use Andre\AiGateway\Facades\AiGateway; $result = AiGateway::invoke('expense_extract', [ 'receipt_text' => $ocrText, ]); $result->text; // the assistant's reply $result->model_used; // 'anthropic/claude-sonnet-4' $result->cost_usd; // 0.0021 $result->usage; // ['prompt_tokens' => ..., 'completion_tokens' => ...] ``` Multi-turn chat layers messages on top of the templated system prompt: ``` $result = AiGateway::invoke('support_assistant', args: ['kb_version' => 'v3'], messages: [ ['role' => 'user', 'content' => 'How do I reset my password?'], ], ); ``` ### 3. Or call it over HTTP [](#3-or-call-it-over-http) The HTTP API and the **API Tokens** admin page are authenticated with [Laravel Sanctum](https://laravel.com/docs/sanctum). One-time setup in your app: ``` composer require laravel/sanctum php artisan migrate # creates personal_access_tokens ``` Then add the `HasApiTokens` trait to your `User` model: ``` use Laravel\Sanctum\HasApiTokens; class User extends Authenticatable { use HasApiTokens; // ... } ``` Now call the endpoint: ``` curl -X POST https://your-app.test/api/ai/expense_extract/chat \ -H "Authorization: Bearer " \ -H "Content-Type: application/json" \ -d '{"args": {"receipt_text": "..."}, "options": {"max_tokens": 256}}' ``` Mint the token from the admin UI (**API Tokens** page — with an optional expiry of 7/30/90 days or 1 year) or in code: ``` // never expires $token = $user->createToken('integration-client', ['ai-gateway:invoke'])->plainTextToken; // expires in 30 days — Sanctum rejects it automatically after that $token = $user->createToken('integration-client', ['ai-gateway:invoke'], now()->addDays(30))->plainTextToken; ``` Sweep expired tokens from the table on a schedule with Sanctum's built-in command: ``` // routes/console.php Schedule::command('sanctum:prune-expired --hours=24')->daily(); ``` The admin UI (Filament) ----------------------- [](#the-admin-ui-filament) Register the plugin on your panel: ``` use Andre\AiGateway\Filament\AiGatewayPlugin; public function panel(Panel $panel): Panel { return $panel // ... ->plugin(AiGatewayPlugin::make()); } ``` You get: - **AI Integrations** — create/edit integrations with a **searchable model picker from the live OpenRouter catalog**, **generation params that auto-populate per model**, a **model-aware prompt-caching** control, an **interactive prompt editor** (click a declared variable to insert `{{name}}`), a **Versions** action that loads any past version back into the form, and a **Test** panel that runs it live and shows tokens/cost/latency. - **Draft with AI** — describe the use case in plain language; the prompt builder fills the template and variable schema for you. - **Invocations** — a read-only telemetry browser: filter by status / caller / integration / date, with cost + token Σ summaries and a per-call detail modal (usage, error, OpenRouter generation link). - **General settings** — toggle the HTTP API, toggle the prompt builder, and pick the helper model (also a catalog-backed Select). - **API Tokens** — mint (with an optional expiry) and revoke scoped invocation tokens; the one-time token has a one-click **Copy to clipboard** button. - **API docs** — the interactive OpenAPI (Scalar) reference embedded right in the panel; browse and test every integration's endpoints without leaving Filament. ### Screenshots [](#screenshots) **Integration form** — catalog model picker, per-model params, caching, prompt editor[![Create integration](screenshots/integration-edit.png?v=2)](screenshots/integration-edit.png?v=2)**Integrations list**[![Integrations](screenshots/integrations-list.png?v=2)](screenshots/integrations-list.png?v=2)**Invocations** — telemetry with Σ summaries[![Invocations](screenshots/invocations.png?v=2)](screenshots/invocations.png?v=2)**Invocation detail** — per-call tokens, cost, latency, OpenRouter id[![Invocation detail](screenshots/invocation-detail.png?v=2)](screenshots/invocation-detail.png?v=2)**Versions** — load a past version into the form[![Versions](screenshots/modal-versions.png?v=2)](screenshots/modal-versions.png?v=2)**General settings**[![General settings](screenshots/general-settings.png?v=2)](screenshots/general-settings.png?v=2)**API tokens** — mint scoped tokens; one-time value with one-click copy[![API tokens](screenshots/api-tokens-created.png?v=2)](screenshots/api-tokens-created.png?v=2)Conversations (multi-turn threads) ---------------------------------- [](#conversations-multi-turn-threads) Flag an integration **conversational** (UI toggle, or `is_conversational` + `conversation_ttl_minutes`) to get server-side memory: the gateway persists each turn, so clients send only the next message — no replaying history. From PHP: ``` use Andre\AiGateway\Facades\AiGateway; $first = AiGateway::converse('support', null, 'My order is late'); // null → new thread $id = $first->conversation_id; // keep this $second = AiGateway::converse('support', $id, 'Order #4471'); // continues with full history ``` Over HTTP (two calls, à la a chatbot `/start` then `/chat`): ``` # 1) open a thread curl -X POST https://your-app.test/api/ai/support/start \ -H "Authorization: Bearer " # → { "data": { "conversation_id": "0779…", "expires_at": "…" } } # 2) send turns curl -X POST https://your-app.test/api/ai/support/converse \ -H "Authorization: Bearer " -H "Content-Type: application/json" \ -d '{"conversation_id": "0779…", "message": "Order #4471"}' ``` Threads are owned by their caller (a guessed id returns 404), expire after the TTL, and link each turn to its telemetry row. Prune expired threads on a schedule: ``` // routes/console.php Schedule::command('ai-gateway:prune-conversations')->daily(); ``` Interactive API docs -------------------- [](#interactive-api-docs) The package serves a **live OpenAPI 3 document built from your integrations**, plus an interactive **[Scalar](https://scalar.com)** docs page with a built-in request tester: - `GET {prefix}/docs` — the docs UI (paste a token, try any endpoint live) - `GET {prefix}/openapi.json` — the raw spec - **…and as an "API docs" page right inside the Filament panel** (embedded, so admins never leave the UI) Every API-visible integration becomes real endpoints: `POST /{slug}/chat` with a request body shaped from its **declared variables** (types + required flags) and the allow-listed `options`, plus `/{slug}/start` and `/{slug}/converse` when the integration is **conversational**. The model and prompt-caching mode appear in each endpoint's description. [![Interactive API docs inside Filament](screenshots/filament-api-docs.png?v=3)](screenshots/filament-api-docs.png?v=3) Gate or disable it via `config('ai-gateway.api.docs')` — add `middleware` (e.g. `['auth']`) to make it private, or override `script_src` to self-host the renderer instead of the CDN. Rate & cost limiting ------------------------ [](#rate--cost-limiting) Set ceilings per integration (UI → Limits, or the `rate_limit_per_minute` / `max_daily_cost_usd` columns). Blank falls back to the config default; a `null` default means unlimited. ``` // config/ai-gateway.php 'rate_limit' => [ 'enabled' => true, 'default_per_minute' => 60, // null = unlimited ], 'cost_limit' => [ 'enabled' => true, 'default_daily_usd' => 25.0, // null = uncapped 'window_hours' => 24, ], ``` When a caller exceeds a limit the gateway throws `RateLimitExceededException` (HTTP **429**) or `CostLimitExceededException` (HTTP **402**) before any spend occurs. Configuration highlights ------------------------ [](#configuration-highlights) Everything in `config/ai-gateway.php` is overridable. Common knobs: KeyPurpose`openrouter.api_key`Your OpenRouter key (`OPENROUTER_API_KEY`).`default_model`Model pre-filled on new integrations.`database.connection` / `database.tables`Relocate / rename the package's tables.`models.*`Swap any Eloquent model for an app subclass.`cache.store` / `cache.ttl_seconds`Integration-resolver cache.`api.enabled` / `api.prefix` / `api.middleware` / `api.token_ability`HTTP API surface.`prompt_builder.model`Helper model (defaults to `anthropic/claude-haiku-4.5`).`filament.navigation_group` / `filament.authorize`Admin UI placement & access gate.Observability ------------- [](#observability) Every call writes one row to `ai_invocations` (success and failure both): ``` use Andre\AiGateway\Models\AiInvocation; // Per-integration spend over the last 24h AiInvocation::where('ai_integration_id', $id) ->where('created_at', '>=', now()->subDay()) ->sum('cost_usd'); ``` Each row keeps OpenRouter's `openrouter_generation_id`, linked to `https://openrouter.ai/logs?transaction={id}` (and copyable) for full provider-side cost forensics — or fetch it via `GET /api/v1/generation?id=…`. Testing ------- [](#testing) ``` composer install vendor/bin/pest ``` Security -------- [](#security) The gateway never sends data to anyone but OpenRouter. Rotate your key by updating `OPENROUTER_API_KEY` and redeploying. If you discover a vulnerability, please email . Credits ------- [](#credits) Built by [Andre Corugda](https://github.com/andrecorugda). License ------- [](#license) The MIT License (MIT). See [LICENSE](LICENSE). ### Health Score 45 — FairBetter than 92% of packages Maintenance100 Actively maintained with recent releases Popularity11 Limited adoption so far Community2 Small or concentrated contributor base Maturity56 Maturing project, gaining track record How is this calculated?**Maintenance (25%)** — Last commit recency, latest release date, and issue-to-star ratio. Uses a 2-year decay window. **Popularity (30%)** — Total and monthly downloads, GitHub stars, and forks. Logarithmic scaling prevents top-heavy scores. **Community (15%)** — Contributors, dependents, forks, watchers, and maintainers. Measures real ecosystem engagement. **Maturity (30%)** — Project age, version count, PHP version support, and release stability. ### Release Activity Cadence Every ~0 days Total 25 Last Release 0d ago Major Versions v0.4.0 → v1.0.02026-06-27 v1.3.1 → v2.0.02026-06-27 ### Community Maintainers ![](https://avatars.githubusercontent.com/u/47452872?v=4)[Andre Corugda](/maintainers/andrecorugda)[@andrecorugda](https://github.com/andrecorugda) --- Tags aiai-gatewaychatbotchatgptclaudefilamentfilamentphpgptlaravellaravel-packagellmllmopsopenrouterphpprompt-managementsanctumlaravelaigatewayfilamentllmOpenRouterprompt-management ### Code Quality TestsPest Static AnalysisPHPStan Code StyleLaravel Pint ### Embed Badge ![Health badge](/badges/andrecorugda-ai-openrouter-gateway/health.svg) ``` [![Health](https://phpackages.com/badges/andrecorugda-ai-openrouter-gateway/health.svg)](https://phpackages.com/packages/andrecorugda-ai-openrouter-gateway) ``` ### Alternatives [psalm/plugin-laravel Psalm plugin for Laravel 3345.1M337](/packages/psalm-plugin-laravel)[larastan/larastan Larastan - Discover bugs in your code without running it. A phpstan/phpstan extension for Laravel 6.4k51.0M7.6k](/packages/larastan-larastan)[defstudio/telegraph A laravel facade to interact with Telegram Bots 815320.5k3](/packages/defstudio-telegraph)[simplestats-io/laravel-client Analytics for Laravel. Track visitors, registrations, and payments. Discover which channels actually drive revenue, not just traffic. Server-side, GDPR compliant, ad-blocker proof. 5019.3k](/packages/simplestats-io-laravel-client)[spatie/laravel-health Monitor the health of a Laravel application 87411.3M153](/packages/spatie-laravel-health)[api-platform/laravel API Platform support for Laravel 59156.3k11](/packages/api-platform-laravel) PHPackages © 2026 [Directory](/)[Categories](/categories)[Trending](/trending)[Changelog](/changelog)[Analyze](/analyze)