Forem: Jenny Met

Unified Billing for Claude and GPT: One Invoice, One Dashboard, One API Key

Jenny Met — Thu, 16 Apr 2026 08:31:07 +0000

Unified Billing for Claude and GPT: One Invoice, One Dashboard, One API Key

If your team uses more than one AI model provider, you already know the pain:

Separate accounts for OpenAI, Anthropic, Google, DeepSeek
Separate billing dashboards
Separate usage reports
Separate API keys to manage and rotate
No single view of total AI spend

For a solo developer, this is annoying. For a team of 10+, it is a real operational problem.

The cost of fragmented billing

It is not just about convenience. Fragmented billing creates real issues:

Budget visibility — When spend is split across 4 providers, nobody knows the real total until month-end reconciliation.

Cost allocation — Which project used how much? Which team member? You cannot answer this if usage is scattered.

Procurement complexity — Each provider means a separate vendor relationship, separate payment method, separate compliance review.

Overspending — Without a unified view, teams routinely overspend because they cannot see the full picture.

How an API gateway solves this

An AI API gateway like Crazyrouter consolidates everything:

from openai import OpenAI

# One client, one key, all models
client = OpenAI(
    api_key="your-crazyrouter-key",
    base_url="https://crazyrouter.com/v1"
)

# Claude
client.chat.completions.create(model="claude-sonnet-4.6", messages=[...])

# GPT
client.chat.completions.create(model="gpt-5.2", messages=[...])

# DeepSeek
client.chat.completions.create(model="deepseek-chat", messages=[...])

Same code structure. Same API key. Same billing dashboard.

What unified billing actually looks like

Per-model breakdown

Model	Requests	Spend
claude-sonnet-4.6	1,200	$18.50
gpt-5-mini	8,500	$4.20
gpt-5-nano	15,000	$1.80
deepseek-chat	3,200	$2.10
Total	27,900	$26.60

Per-member breakdown

Create separate API keys for each team member. Usage is automatically isolated:

Member	Spend	Primary model
Alice	$12.30	Claude Sonnet
Bob	$8.20	GPT-5 Mini
Carol	$6.10	DeepSeek

Usage export

Export detailed CSV records with:

Timestamp
Model used
Input/output tokens
Cost per request

Perfect for expense reports, project accounting, and budget reviews.

Cost savings on top of consolidation

Beyond the operational savings, gateway pricing is typically 40-55% below direct provider pricing:

Model	Direct pricing	Gateway pricing	Savings
Claude Sonnet 4.6	$3/$15 per 1M	~$1.65/$8.25	45%
GPT-5	$1.25/$10 per 1M	~$0.69/$5.50	45%
GPT-5 Nano	$0.05/$0.40	~$0.03/$0.22	45%

Pay-as-you-go. No monthly fee. No minimum spend.

Getting started

Sign up at crazyrouter.com
Create API keys for your team
Change base_url to https://crazyrouter.com/v1
Monitor usage from one dashboard

For enterprise procurement inquiries: support@crazyrouter.com

Как использовать Claude API дешевле в 2026: экономия 40%+ без подписки

Jenny Met — Thu, 16 Apr 2026 03:44:09 +0000

Claude API стоит дорого? Вы не одиноки.

Claude Opus 4 — $15 за вход / $75 за выход (на 1M токенов), Sonnet 4.6 — $3/$15. При десятках тысяч запросов в месяц счёт растёт быстро.

В этой статье — 5 способов использовать Claude API максимально дёшево, без подписки.

Официальные цены Claude API в 2026

Модель	Вход (1M токенов)	Выход (1M токенов)	Кэш	Для чего
Claude Opus 4	$15.00	$75.00	$1.50	Сложный анализ
Claude Sonnet 4.6	$3.00	$15.00	$0.30	Лучшее соотношение цена/качество
Claude Haiku 3.5	$0.80	$4.00	$0.08	Быстрые дешёвые задачи

Способ 1: AI API Gateway — скидка 40-55%

Самый простой способ сэкономить — использовать AI API gateway. Шлюзы договариваются об оптовых скидках и передают их пользователям.

Crazyrouter предоставляет 627+ моделей примерно за 55% от официальной цены:

Модель	Anthropic напрямую	Crazyrouter	Экономия
Claude Opus 4	$15/$75	~$8.25/$41.25	45%
Claude Sonnet 4.6	$3/$15	~$1.65/$8.25	45%
Claude Haiku 3.5	$0.80/$4.00	~$0.44/$2.20	45%

Код: меняем 2 строки

from openai import OpenAI

# Доступ к Claude API через Crazyrouter
client = OpenAI(
    base_url="https://crazyrouter.com/v1",
    api_key="your-crazyrouter-key"
)

response = client.chat.completions.create(
    model="claude-sonnet-4.6",
    messages=[{"role": "user", "content": "Напиши сортировку слиянием на Python"}]
)
print(response.choices[0].message.content)

Формат совместим с OpenAI — просто замените base_url и api_key. Без абонентской платы, платите только за использование.

Способ 2: Кэширование промптов — минус 90% на входных токенах

Повторяющиеся системные промпты можно кэшировать со скидкой 90%:

messages = [
    {
        "role": "system",
        "content": "Ты опытный Python-разработчик..."  # Кэшируется
    },
    {
        "role": "user",
        "content": "Найди баг в этом коде"  # Меняется каждый раз
    }
]

Системный промпт в 1000 токенов: со второго запроса $0.30 → $0.03.

Способ 3: Выбор модели под задачу

Не каждый запрос требует Claude Opus 4:

Задача	Рекомендуемая модель	Стоимость (выход 1M)
Простые вопросы	Claude Haiku 3.5	$4.00
Программирование	Claude Sonnet 4.6	$15.00
Глубокий анализ	Claude Opus 4	$75.00
Классификация	DeepSeek Chat	$0.28

С Crazyrouter одним API-ключом можно переключаться между моделями:

# Простые задачи — DeepSeek (ещё дешевле)
response = client.chat.completions.create(
    model="deepseek-chat",
    messages=[{"role": "user", "content": "Кратко перескажи этот текст"}]
)

# Код — Claude Sonnet
response = client.chat.completions.create(
    model="claude-sonnet-4.6",
    messages=[{"role": "user", "content": "Напиши HTTP-сервер на Go"}]
)

Способ 4: Пакетная обработка — скидка 50%

Если ответ не нужен в реальном времени, используйте Batch API — обработка за 24 часа со скидкой 50%:

Массовая классификация текстов
Разметка датасетов
Пакетный перевод

Способ 5: Контроль длины ответа

Установите max_tokens, чтобы не тратить токены на лишний вывод:

response = client.chat.completions.create(
    model="claude-sonnet-4.6",
    messages=[{"role": "user", "content": "Да или Нет: Москва — столица России?"}],
    max_tokens=10
)

Выходные токены в 5 раз дороже входных — экономия здесь ощутима.

Сравнение затрат

100 000 запросов в месяц (в среднем 500 входных + 500 выходных токенов):

Метод	Claude Sonnet 4.6 в месяц
Anthropic напрямую	~$900
Через Crazyrouter	~$495
+ Выбор модели	~$200-300
+ Кэширование	~$100-200

Комбинация методов позволяет сэкономить более 80% от прямой оплаты.

Итого

AI API Gateway (Crazyrouter) снижает базовую цену на 40-55%
Кэширование промптов — минус 90% на повторяющемся контексте
Выбор модели — не переплачивайте за простые задачи
Пакетная обработка — 50% скидка на фоновые задачи
max_tokens — убирает лишние выходные токены

Claude API дорогой — но при правильном подходе можно платить в разы меньше.

👉 Crazyrouter — 627+ моделей по лучшей цене

ChatGPT 6 Release Date: What Is Official, What Is Rumor, and What to Watch

Jenny Met — Thu, 16 Apr 2026 03:44:08 +0000

ChatGPT 6 Release Date: What Is Official, What Is Rumor, and What to Watch

A lot of pages ranking for "ChatGPT 6 release date" are really doing one thing: mixing rumors with official information.

The clean answer is much shorter.

There is no officially announced public ChatGPT 6 release date right now.

That does not stop the rumor cycle, but it does change how developers and content teams should approach this keyword.

ChatGPT 6 release date vs GPT-6 release date

These phrases overlap, but they are not exactly the same:

Phrase	What people usually mean
GPT-6 release date	next major OpenAI model release
ChatGPT 6 release date	next major ChatGPT experience or default model rollout

Why this matters:

API access and ChatGPT rollout do not always happen together
launch can be staged by plan, region, or account type
naming can shift without a clean "ChatGPT 6" label everywhere

Quick public verification

I checked OpenAI's public ChatGPT release notes page for obvious GPT-6 references.

import requests

url = "https://help.openai.com/en/articles/6825453-chatgpt-release-notes"
html = requests.get(url, headers={"User-Agent": "Mozilla/5.0"}, timeout=30).text
print("gpt-6" in html.lower())

Observed output:

False

That does not predict the future. It just shows that public official confirmation is not there yet.

What signals matter more than rumors?

The best sources are:

official release notes
API docs
pricing pages
dashboard or playground references
product wording changes

If a page gives an exact launch date without an official OpenAI source, treat it as speculation.

What should developers do now?

Do not wait for certainty. Prepare your stack:

keep model names in config
add fallback models
track token cost and latency
use one OpenAI-compatible layer

from openai import OpenAI

client = OpenAI(
    api_key="your-crazyrouter-key",
    base_url="https://crazyrouter.com/v1"
)

MODEL = "gpt-5.2"  # change later if GPT-6 becomes available

That makes launch-week testing much easier.

Short answer

no official public ChatGPT 6 release date has been announced
rumors are common, but official OpenAI sources matter more
the smartest move is to prepare for a staged rollout rather than one perfect launch day

Read the full version here:

ChatGPT 6 Release Date: Rumors, Official Signals, and What Users Should Actually Watch in 2026

2026 Yılında Claude API En Ucuz Nasıl Kullanılır: Abonelik Olmadan %40+ Tasarruf

Jenny Met — Wed, 15 Apr 2026 18:44:46 +0000

Claude API pahalı mı geliyor? Yalnız değilsiniz.

Claude Opus 4 giriş $15/çıkış $75 (1M token başına), Sonnet 4.6 bile $3/$15. Ayda on binlerce istek gönderiyorsanız fatura hızla şişer.

Bu yazıda abonelik olmadan Claude API'yi en ucuza kullanmanın 5 yolunu anlatıyorum.

Claude API 2026 Resmi Fiyatları

Model	Giriş (1M token)	Çıkış (1M token)	Önbellek	Kullanım
Claude Opus 4	$15.00	$75.00	$1.50	Karmaşık analiz
Claude Sonnet 4.6	$3.00	$15.00	$0.30	En iyi fiyat/performans
Claude Haiku 3.5	$0.80	$4.00	$0.08	Hızlı, ucuz işler

Yöntem 1: AI API Gateway ile %40-55 İndirim

En basit tasarruf yöntemi: AI API gateway üzerinden erişim. Gateway'ler toplu indirim anlaşmaları yapar ve bu indirimi kullanıcılara yansıtır.

Crazyrouter 627+ modeli resmi fiyatın yaklaşık %55'ine sunan bir gateway:

Model	Anthropic Direkt	Crazyrouter	Tasarruf
Claude Opus 4	$15/$75	~$8.25/$41.25	%45
Claude Sonnet 4.6	$3/$15	~$1.65/$8.25	%45
Claude Haiku 3.5	$0.80/$4.00	~$0.44/$2.20	%45

Kod: Sadece 2 Satır Değişiklik

from openai import OpenAI

# Crazyrouter üzerinden Claude API'ye erişim
client = OpenAI(
    base_url="https://crazyrouter.com/v1",
    api_key="your-crazyrouter-key"
)

response = client.chat.completions.create(
    model="claude-sonnet-4.6",
    messages=[{"role": "user", "content": "Python'da binary search yaz"}]
)
print(response.choices[0].message.content)

OpenAI uyumlu format — mevcut kodunuzda sadece base_url ve api_key değiştirin. Aylık ücret yok, kullandığınız kadar ödeyin.

Yöntem 2: Prompt Önbelleği ile Giriş Maliyetini %90 Düşürün

Tekrarlanan sistem promptlarının maliyetini %90 azaltabilirsiniz:

messages = [
    {
        "role": "system",
        "content": "Sen deneyimli bir yazılım mühendisisin..."  # Önbelleğe alınır
    },
    {
        "role": "user",
        "content": "Bu koddaki hatayı bul"  # Her seferinde değişir
    }
]

1.000 tokenlik sistem promptu: ikinci istekten itibaren $0.30 → $0.03.

Yöntem 3: Göreve Göre Model Seçimi

Her istek için Claude Opus 4 kullanmanıza gerek yok:

Görev	Önerilen Model	Maliyet (Çıkış 1M)
Basit soru-cevap	Claude Haiku 3.5	$4.00
Kodlama	Claude Sonnet 4.6	$15.00
Derin analiz	Claude Opus 4	$75.00
Basit sınıflandırma	DeepSeek Chat	$0.28

Crazyrouter ile aynı API anahtarıyla tüm modellere erişebilirsiniz:

# Basit görevler için DeepSeek (çok daha ucuz)
response = client.chat.completions.create(
    model="deepseek-chat",
    messages=[{"role": "user", "content": "Bu metni özetle"}]
)

# Kodlama için Claude Sonnet
response = client.chat.completions.create(
    model="claude-sonnet-4.6",
    messages=[{"role": "user", "content": "Rust'ta HTTP sunucu yaz"}]
)

Yöntem 4: Toplu İşleme ile %50 İndirim

Gerçek zamanlı yanıt gerekmiyorsa, Batch API kullanın — 24 saat içinde işlenir, %50 indirimli:

Büyük metin sınıflandırma işleri
Veri seti etiketleme
Toplu çeviri

Yöntem 5: Yanıt Uzunluğunu Kontrol Edin

max_tokens ayarlayarak gereksiz çıkış tokenlarını azaltın:

response = client.chat.completions.create(
    model="claude-sonnet-4.6",
    messages=[{"role": "user", "content": "Evet veya Hayır: İstanbul Türkiye'nin en büyük şehri mi?"}],
    max_tokens=10
)

Çıkış tokenları girişin 5 katı pahalı — burayı kısmak büyük fark yaratır.

Maliyet Karşılaştırması

Ayda 100.000 istek (ortalama 500 giriş + 500 çıkış token):

Yöntem	Claude Sonnet 4.6 Aylık
Anthropic direkt	~$900
Crazyrouter	~$495
+ Model seçimi	~$200-300
+ Önbellek	~$100-200

Kombinasyon ile direkt fiyatın %80'inden fazla tasarruf mümkün.

Özet

AI API Gateway (Crazyrouter) ile temel fiyatı %40-55 düşürün
Prompt önbelleği ile tekrar maliyetini %90 azaltın
Model seçimi ile gereksiz harcamayı önleyin
Toplu işleme ile acil olmayan işleri %50 ucuza yapın
max_tokens ile çıkış israfını kesin

Claude API pahalı — ama doğru kullanırsanız çok daha ucuza gelir.

👉 Crazyrouter — 627+ modeli en ucuza kullanın

How to Use Hermes Agent with Crazyrouter — 600+ Models, Lower Cost

Jenny Met — Wed, 15 Apr 2026 18:44:44 +0000

How to Use Hermes Agent with Crazyrouter — 600+ Models, Lower Cost

Hermes Agent by NousResearch has quickly become one of the most capable open-source AI agents available. With 57,000+ GitHub stars, built-in terminal access, browser automation, MCP support, and persistent memory, it's a serious tool for developers who want an agent that actually gets things done.

By default, Hermes routes all LLM calls through OpenRouter. But here's something most people don't realize: you can swap the provider to any OpenRouter-compatible gateway without touching a single line of Hermes source code.

This guide shows you how to set up Hermes Agent with Crazyrouter — an AI API gateway that gives you access to 600+ models at prices typically 30-50% lower than going direct.

Why Switch from OpenRouter to Crazyrouter?

Both OpenRouter and Crazyrouter use the same OpenAI-compatible API format. The difference is what you get on the other side:

	OpenRouter	Crazyrouter
Models available	300+	600+
Pricing	Standard markup	30-50% cheaper on most models
API format	OpenAI-compatible	OpenAI-compatible
Free tier	✅	✅
Video/Audio models	❌	✅ (Sora, Whisper, TTS)
Image generation	Limited	✅ (DALL-E, Flux, Midjourney)

The key point: Crazyrouter is a drop-in replacement. Same API format, same model naming convention, different base URL.

Quick Setup (One Command)

If you already have Hermes installed, run this:

curl -fsSL https://raw.githubusercontent.com/xujfcn/crazyrouter-hermes/main/setup-crazyrouter.sh | bash

The script will:

Ask for your Crazyrouter API key (get one free at crazyrouter.com)
Update ~/.hermes/.env with your key
Update ~/.hermes/config.yaml to point to Crazyrouter's endpoint

That's it. Run hermes and you're using Crazyrouter.

You can also pass the key as an environment variable:

CRAZYROUTER_API_KEY=sk-your-key curl -fsSL https://raw.githubusercontent.com/xujfcn/crazyrouter-hermes/main/setup-crazyrouter.sh | bash

Manual Setup (Step by Step)

Step 1: Get a Crazyrouter API Key

Step 2: Edit the Config File

Open ~/.hermes/config.yaml and update the model section:

model:
  provider: "openrouter"
  base_url: "https://crazyrouter.com/v1"
  default: "anthropic/claude-opus-4.6"

The provider: "openrouter" setting tells Hermes to use the OpenRouter-compatible code path. The base_url override redirects all requests to Crazyrouter instead.

Step 3: Set the API Key

Edit ~/.hermes/.env:

OPENROUTER_API_KEY=sk-your-crazyrouter-api-key

Hermes reads OPENROUTER_API_KEY when the provider is set to openrouter. Since Crazyrouter uses the same API format, your Crazyrouter key works here directly.

Step 4: Verify

hermes chat "What model are you using?"

How It Works Under the Hood

Hermes Agent's architecture makes this swap trivial. Here's why:

Hermes uses the OpenAI Python SDK internally
When provider: "openrouter" is set, it reads OPENROUTER_API_KEY and uses the configured base_url
Crazyrouter exposes an OpenAI-compatible API at https://crazyrouter.com/v1
All endpoints work: /chat/completions, /audio/transcriptions, /images/generations, etc.

No monkey-patching. No forks. Just a config change.

Recommended Models

Once connected to Crazyrouter, you can switch models with hermes model:

Model	Best for	Input cost
`anthropic/claude-opus-4.6`	Complex reasoning, coding	~$15/M tokens
`anthropic/claude-sonnet-4-5`	Balanced speed and quality	~$3/M tokens
`openai/gpt-5.2`	General purpose	~$12/M tokens
`google/gemini-3-pro`	Multimodal, long context (2M)	~$7/M tokens
`deepseek/deepseek-r2`	Budget reasoning	~$0.5/M tokens
`google/gemini-3-flash`	Fast tasks, cheap	~$0.1/M tokens

Advanced: Smart Model Routing

Hermes supports smart routing — using a cheap model for simple turns and your main model for complex ones:

smart_model_routing:
  enabled: true
  max_simple_chars: 160
  max_simple_words: 28
  cheap_model:
    provider: openrouter
    model: google/gemini-3-flash

Simple responses go to Gemini Flash (~$0.10/M tokens), while complex reasoning stays on Claude Opus. This can cut your costs significantly without sacrificing quality.

Advanced: Compression Model

Hermes compresses long conversations to stay within context limits. Route the compression model through Crazyrouter too:

compression:
  enabled: true
  threshold: 0.50
  summary_model: "google/gemini-3-flash"

Switching Back

If you ever want to switch back to OpenRouter:

model:
  provider: "openrouter"
  base_url: "https://openrouter.ai/api/v1"

Repository

The setup script and full documentation are on GitHub:

github.com/xujfcn/crazyrouter-hermes

Conclusion

Hermes Agent is one of the best open-source AI agents available today. By pointing it at Crazyrouter instead of the default provider, you get access to more models at lower cost — without changing any source code.

The setup takes about 30 seconds. One config change, one API key, and you're running Hermes with 600+ models behind it.

Get your free API key at crazyrouter.com and try it out.

Crazyrouter is an AI API gateway that provides unified access to 600+ AI models through a single API key. OpenAI-compatible format, no vendor lock-in.

Claude Code Complete Tutorial: From Installation to Production (2026)

Jenny Met — Sat, 11 Apr 2026 16:35:40 +0000

What is Claude Code?

Claude Code is Anthropic's official AI coding assistant that runs directly in your terminal. Unlike Cursor or GitHub Copilot, it understands your entire codebase and can edit multiple files simultaneously.

Key capabilities:

🔍 Understands full project context (not just the open file)
✏️ Edits multiple files in a single command
🖥️ Executes shell commands (tests, builds, deploys)
🗣️ Natural language interaction — describe what you want, AI delivers

Best for: Backend development, scripting, automation, CI/CD integration, code review.

Installation

Requirements

Node.js 18.0+
macOS, Linux, or Windows 10+
Terminal with ANSI color support

Install

# npm (all platforms)
npm install -g @anthropic-ai/claude-code

# Verify
claude-code --version

Configuration

Option 1: Official Anthropic API

export ANTHROPIC_API_KEY="sk-ant-api03-xxxxx"

Pricing: Claude Opus 4.6 — $15/M input, $75/M output tokens.

Option 2: API Gateway (Recommended — 45% Cheaper)

Why use an API gateway?

45% cheaper than official pricing
600+ models with one key (Claude + GPT + Gemini + DeepSeek)
Better international latency
No credit card required (Alipay/WeChat supported)

export ANTHROPIC_API_KEY="sk-xxxxxx"  # Your gateway key
export ANTHROPIC_BASE_URL="https://crazyrouter.com/v1"

Method	Input Price	Output Price	Savings
Official API	$15/M	$75/M	—
API Gateway	$8.25/M	$41.25/M	45%

Get your API key →

Core Usage

# Start in your project
cd /path/to/project
claude-code

Essential Commands

Command	Description
`/add <path>`	Add files/dirs to context
`/clear`	Clear conversation history
`/model <name>`	Switch model
`/cost`	Show session cost
`/exit`	Quit

Real-World Examples

Create a REST API

> Create a Flask REST API with:
> 1. GET /users - list all users
> 2. POST /users - create user with validation
> 3. SQLite database
> 4. Error handling and logging

Claude Code creates app.py, models.py, requirements.txt, init_db.py — all tested and runnable.

Refactor Entire Codebase

> Convert all sync functions in src/ to async/await

Scans all Python files → identifies sync functions → converts to async → updates all call sites → adds asyncio imports.

Security Audit

> Review src/auth.py for security vulnerabilities

Output:

🔴 Security Issues:
  - Line 45: Passwords stored unencrypted
  - Line 78: SQL injection risk (string concatenation)

🟡 Performance:
  - Line 120: N+1 query in loop
  - Line 156: Missing database index

✅ Recommendations:
  1. Use bcrypt for password hashing
  2. Switch to parameterized queries
  3. Add composite index on (user_id, created_at)

Generate Tests

> Generate unit tests for all functions in src/api.py

Creates tests/test_api.py with happy path, edge cases, error handling, and mocked dependencies.

Advanced: CI/CD Integration

# .github/workflows/ai-review.yml
name: AI Code Review
on: [pull_request]
jobs:
  review:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v3
      - uses: actions/setup-node@v3
        with:
          node-version: '18'
      - run: npm install -g @anthropic-ai/claude-code
      - name: AI Review
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
          ANTHROPIC_BASE_URL: "https://crazyrouter.com/v1"
        run: |
          claude-code --read-only --non-interactive \
            "Review this PR for bugs, security issues, and style" \
            > review.txt
          cat review.txt >> $GITHUB_STEP_SUMMARY

Claude Code vs Cursor vs Copilot

Feature	Claude Code	Cursor	Copilot
Interface	Terminal	GUI (VS Code)	IDE Plugin
Full codebase context	✅	⚠️ Limited	❌
Multi-file editing	✅	✅	❌
Custom API endpoint	✅	❌	❌
CI/CD integration	✅	❌	❌
Price	Pay-per-token	$20/mo	$10/mo
Model flexibility	✅ Any model	Claude only	GPT only

Choose Claude Code if: you do backend work, need CI/CD integration, want model flexibility, or prefer terminal workflows.

Choose Cursor if: you prefer visual editing and work primarily on frontend.

Cost Optimization

Use Sonnet 4.5 for simple tasks — 5x cheaper than Opus 4.6
Clear context with /clear when switching topics
Use an API gateway — save 45% on every API call
Monitor spending with /cost command
Batch similar tasks instead of making many small requests

FAQ

Q: Does Claude Code upload my entire codebase?
Only files you add to context via /add or that Claude requests to read. Use --read-only for extra safety.

Q: Which languages are supported?
All major languages: Python, JavaScript/TypeScript, Go, Rust, Ruby, PHP, Swift, Kotlin, C/C++, SQL, and more.

Q: Can I use non-Claude models?
Yes! With an API gateway:

/model gpt-4o
/model gemini-2.5-pro
/model deepseek-v3

Q: How do I handle rate limits?
Use an API gateway with automatic fallback — if one provider is rate-limited, it routes to another model automatically.

Q: Is my code secure?
Code is sent to the API provider (Anthropic or your gateway). For sensitive codebases, consider self-hosting or using --read-only mode.

Get Started

Install: npm install -g @anthropic-ai/claude-code
Get an API key: crazyrouter.com (45% cheaper, 600+ models)
Configure: export ANTHROPIC_BASE_URL="https://crazyrouter.com/v1"
Start coding: claude-code

Try Crazyrouter — one API key for 600+ AI models, 45% cheaper than going direct. crazyrouter.com

Cheapest Way to Access GPT-5 and Claude API in 2026: Pricing Breakdown

Jenny Met — Fri, 10 Apr 2026 17:23:38 +0000

GPT-5 and Claude are the two most popular AI APIs in 2026, but their pricing adds up fast at scale. If you're making thousands of API calls per day, even small per-token savings compound into significant monthly differences.

I compared every way to access these models — direct API, batch processing, cached inputs, and third-party gateways — to find the actual cheapest options.

GPT-5 Pricing: Official Rates (OpenAI Direct)

OpenAI offers multiple GPT-5 variants at different price points:

Model	Input (per 1M tokens)	Output (per 1M tokens)	Cached Input	Best For
GPT-5	$1.25	$10.00	$0.125	General tasks
GPT-5 Mini	$0.25	$2.00	$0.025	Cost-sensitive workloads
GPT-5 Nano	$0.05	$0.40	$0.005	High-volume, simple tasks

Cheapest direct option: GPT-5 Nano at $0.05/$0.40 per million tokens. For most use cases, GPT-5 Mini at $0.25/$2.00 offers the best quality-to-cost ratio.

Batch API Discount

OpenAI's Batch API gives you 50% off standard pricing for non-time-sensitive workloads (processed within 24 hours):

Model	Batch Input	Batch Output
GPT-5	$0.625	$5.00
GPT-5 Mini	$0.125	$1.00
GPT-5 Nano	$0.025	$0.20

If your workload can tolerate 24-hour turnaround, batch processing is the single biggest cost reduction available.

Claude Pricing: Official Rates (Anthropic Direct)

Anthropic's Claude model lineup in 2026:

Model	Input (per 1M tokens)	Output (per 1M tokens)	Cached Input	Best For
Claude Opus 4	$15.00	$75.00	$1.50	Complex reasoning
Claude Sonnet 4.6	$3.00	$15.00	$0.30	Best quality/cost balance
Claude Haiku 3.5	$0.80	$4.00	$0.08	Fast, cheap tasks

Cheapest direct option: Claude Haiku 3.5 at $0.80/$4.00. For coding and complex tasks, Claude Sonnet 4.6 at $3.00/$15.00 is the sweet spot.

5 Ways to Reduce GPT-5 and Claude API Costs

1. Use Smaller Model Variants

The most obvious savings: use GPT-5 Nano or Claude Haiku instead of the flagship models when quality requirements allow it.

Savings: 75-95% compared to flagship models.

2. Enable Prompt Caching

Both OpenAI and Anthropic offer cached input pricing at ~90% discount. If your prompts share common system messages or context:

# Prompt caching reduces repeated context costs by ~90%
# Structure your prompts with static system messages first
messages = [
    {"role": "system", "content": "Your long system prompt here..."},  # Cached
    {"role": "user", "content": "User's specific question"}  # Not cached
]

Savings: Up to 90% on input tokens for repeated context.

3. Use Batch API for Non-Urgent Workloads

OpenAI's Batch API processes requests within 24 hours at 50% off:

# Submit batch requests for background processing
import openai

client = openai.OpenAI()
batch = client.batches.create(
    input_file_id="file-abc123",
    endpoint="/v1/chat/completions",
    completion_window="24h"
)

Savings: 50% on all token costs.

4. Use an AI API Gateway with Below-Official Pricing

Some AI API gateways offer access to GPT-5 and Claude at prices lower than going direct to OpenAI/Anthropic. This works because gateways negotiate volume discounts and pass savings to users.

Crazyrouter is one gateway that consistently prices below official rates:

from openai import OpenAI

client = OpenAI(
    base_url="https://crazyrouter.com/v1",
    api_key="sk-your-crazyrouter-key"
)

# Access GPT-5 at below-official pricing
response = client.chat.completions.create(
    model="gpt-5",
    messages=[{"role": "user", "content": "Hello"}]
)

# Same key works for Claude too
response = client.chat.completions.create(
    model="claude-sonnet-4.6",
    messages=[{"role": "user", "content": "Hello"}]
)

Savings: 10-40% below official pricing depending on the model. Check crazyrouter.com/pricing for current rates.

Bonus: You also get access to 627+ other models (DeepSeek, Gemini, Qwen, Grok, etc.) through the same API key, so you can easily switch to cheaper alternatives when appropriate.

5. Route to the Cheapest Capable Model Dynamically

Not every request needs GPT-5. Use a routing strategy:

def smart_route(task_complexity, messages):
    """Route to the cheapest model that can handle the task"""
    client = OpenAI(
        base_url="https://crazyrouter.com/v1",
        api_key="sk-your-key"
    )

    if task_complexity == "simple":
        model = "gpt-5-nano"       # $0.05/$0.40 per 1M
    elif task_complexity == "medium":
        model = "deepseek-chat"     # Even cheaper for many tasks
    elif task_complexity == "coding":
        model = "claude-sonnet-4.6" # Best for code
    else:
        model = "gpt-5"            # Full power when needed

    return client.chat.completions.create(
        model=model,
        messages=messages
    )

Savings: 60-90% by routing simple tasks to cheaper models.

Cost Comparison: Direct vs. Gateway

Here's what 1 million API calls (average 500 input + 500 output tokens each) costs across different access methods:

Method	GPT-5 Cost	Claude Sonnet 4.6 Cost
Direct API	$5.63	$9.00
Direct + Batch API	$2.81	N/A
Direct + Caching	~$1.00-3.00	~$2.00-5.00
Crazyrouter	Below direct	Below direct
OpenRouter	~$6.19 (+10%)	~$9.90 (+10%)

Estimates based on 500 input + 500 output tokens per request. Actual costs vary by usage pattern.

The Cheapest Possible Setup

For maximum cost savings, combine multiple strategies:

Use Crazyrouter for below-official base pricing
Route simple tasks to GPT-5 Nano or DeepSeek
Use prompt caching for repeated context
Batch non-urgent work through OpenAI's Batch API
Monitor costs with a tool like Helicone

This combination can reduce your AI API costs by 70-90% compared to naively calling GPT-5 for everything.

Quick Start: Cheapest GPT-5 + Claude Access

from openai import OpenAI

# One key, access to both GPT-5 and Claude at below-official prices
client = OpenAI(
    base_url="https://crazyrouter.com/v1",
    api_key="sk-your-crazyrouter-key"
)

# Cheap GPT-5 access
response = client.chat.completions.create(
    model="gpt-5-nano",  # Cheapest GPT-5 variant
    messages=[{"role": "user", "content": "Summarize this text..."}]
)

# Cheap Claude access
response = client.chat.completions.create(
    model="claude-haiku-3.5",  # Cheapest Claude variant
    messages=[{"role": "user", "content": "Fix this bug..."}]
)

# Or use DeepSeek for even cheaper inference
response = client.chat.completions.create(
    model="deepseek-chat",
    messages=[{"role": "user", "content": "Simple question..."}]
)

Last updated: April 2026. Pricing changes frequently — verify current rates on each provider's website.

How to Access All AI Models with a Single API Key in 2026

Jenny Met — Thu, 02 Apr 2026 08:03:36 +0000

You want to use GPT-5 for general tasks, Claude for coding, Gemini for long documents, and DeepSeek for cheap inference. That means four API keys, four billing accounts, four different SDKs, and four sets of rate limits to manage.

There's a better way. Unified AI API gateways let you access all of these models — and hundreds more — through a single API key and endpoint.

This guide shows you exactly how to set it up in under 5 minutes.

The Problem with Multiple API Keys

If you're calling AI models directly, your setup looks something like this:

# The painful way — managing multiple clients
import openai
import anthropic
import google.generativeai as genai

openai_client = openai.OpenAI(api_key="sk-openai-...")
anthropic_client = anthropic.Anthropic(api_key="sk-ant-...")
genai.configure(api_key="AIza...")

# Different APIs, different formats, different error handling for each

This creates real problems:

Key management overhead — rotating, securing, and tracking 5+ API keys
Billing fragmentation — separate invoices from each provider
Code complexity — different SDKs and response formats per provider
No failover — if OpenAI goes down, your app goes down

The Solution: Unified AI API Gateways

A unified gateway gives you one endpoint that routes to all providers:

from openai import OpenAI

# One client, one key, access to everything
client = OpenAI(
    base_url="https://crazyrouter.com/v1",
    api_key="sk-your-single-key"
)

# Use any model — just change the model name
gpt_response = client.chat.completions.create(
    model="gpt-5",
    messages=[{"role": "user", "content": "Explain quantum computing"}]
)

claude_response = client.chat.completions.create(
    model="claude-sonnet-4.6",
    messages=[{"role": "user", "content": "Review this code..."}]
)

gemini_response = client.chat.completions.create(
    model="gemini-2.5-pro",
    messages=[{"role": "user", "content": "Summarize this 500-page PDF..."}]
)

Same client. Same format. Same API key. Just change the model parameter.

Step-by-Step Setup

Option 1: Crazyrouter (627+ Models, Broadest Coverage)

Crazyrouter provides access to 627+ models from 20+ providers including OpenAI, Anthropic, Google, DeepSeek, ByteDance, Alibaba, and xAI.

Step 1: Sign up at crazyrouter.com (you get $0.20 free credit)

Step 2: Copy your API key from the dashboard

Step 3: Use it with the OpenAI SDK:

from openai import OpenAI

client = OpenAI(
    base_url="https://crazyrouter.com/v1",
    api_key="sk-your-crazyrouter-key"
)

# Now you can access any of 627+ models
response = client.chat.completions.create(
    model="gpt-5",  # or claude-sonnet-4.6, gemini-2.5-pro, deepseek-r1, etc.
    messages=[{"role": "user", "content": "Hello!"}]
)
print(response.choices[0].message.content)

Step 4 (optional): Use environment variables for production:

export OPENAI_BASE_URL=https://crazyrouter.com/v1
export OPENAI_API_KEY=sk-your-crazyrouter-key

Now any tool that uses the OpenAI SDK (Cursor, Continue, LangChain, etc.) will automatically route through Crazyrouter.

Bonus: Crazyrouter also supports native Anthropic and Gemini API formats, so you don't have to use the OpenAI compatibility layer if you prefer the native SDKs.

Option 2: OpenRouter (300+ Models, Free Tier)

OpenRouter is the most popular option with a free tier for select models.

Step 1: Sign up at openrouter.ai

Step 2: Get your API key

Step 3: Use it:

from openai import OpenAI

client = OpenAI(
    base_url="https://openrouter.ai/api/v1",
    api_key="sk-or-your-key"
)

response = client.chat.completions.create(
    model="openai/gpt-5",  # OpenRouter uses provider/model format
    messages=[{"role": "user", "content": "Hello!"}]
)

Note: OpenRouter adds a 5% fee on BYOK (bring your own key) usage and has markup on pay-per-token pricing.

Option 3: LiteLLM (Open-Source, Self-Hosted)

If you need to self-host, LiteLLM is the go-to open-source option.

Step 1: Install LiteLLM:

pip install litellm

Step 2: Use it as a Python library:

from litellm import completion

# You still need individual provider keys, but LiteLLM unifies the interface
response = completion(
    model="gpt-5",
    messages=[{"role": "user", "content": "Hello!"}],
    api_key="sk-openai-key"
)

response = completion(
    model="claude-sonnet-4.6",
    messages=[{"role": "user", "content": "Hello!"}],
    api_key="sk-ant-key"
)

Note: LiteLLM unifies the API format but you still manage individual provider keys unless you run the proxy server.

Which Gateway Should You Choose?

Need	Best Option
Most models (627+)	Crazyrouter
Free tier / experimentation	OpenRouter
Self-hosted / open-source	LiteLLM
Enterprise observability	Portkey
Multimodal beyond text	AIMLAPI or Eden AI

Using Your Single API Key with Popular Tools

Once you have a gateway API key, you can use it with most AI-powered tools:

Cursor IDE

// .cursor/config.json
{
  "openai.baseUrl": "https://crazyrouter.com/v1",
  "openai.apiKey": "sk-your-key"
}

LangChain

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    base_url="https://crazyrouter.com/v1",
    api_key="sk-your-key",
    model="gpt-5"
)

cURL

curl https://crazyrouter.com/v1/chat/completions \
  -H "Authorization: Bearer sk-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4.6",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

Node.js

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'https://crazyrouter.com/v1',
  apiKey: 'sk-your-key'
});

const response = await client.chat.completions.create({
  model: 'gemini-2.5-pro',
  messages: [{ role: 'user', content: 'Hello!' }]
});

Cost Comparison: Direct vs. Gateway

Using a gateway doesn't have to cost more. Some gateways actually offer lower prices than going direct:

Model	Official Price (per 1M tokens)	Crazyrouter	OpenRouter
GPT-5	$1.25 / $10.00	Below official	~$1.38 / $11.00
Claude Sonnet 4.6	$3.00 / $15.00	Below official	~$3.30 / $16.50
Gemini 2.5 Pro	$1.25 / $10.00	Below official	~$1.38 / $11.00
DeepSeek R1	$0.55 / $2.19	Below official	~$0.60 / $2.41

Prices are approximate and change frequently. Check each platform for current rates.

FAQ

Q: Is there latency overhead?
A: Minimal. Most gateways add 10-100ms depending on your location. Crazyrouter's 7 global edge nodes keep overhead under 50ms for most regions.

Q: What happens if a provider goes down?
A: Good gateways have automatic failover. Your request gets rerouted to an alternative provider transparently.

Q: Can I use my own provider API keys?
A: Some gateways support BYOK (bring your own key). OpenRouter charges 5% on BYOK usage. Crazyrouter uses its own pooled keys with below-official pricing.

Q: Is it secure?
A: Your requests are proxied through the gateway. Choose a gateway with a clear privacy policy and data handling practices.

Summary

Accessing all AI models with a single API key is straightforward in 2026:

Pick a gateway (Crazyrouter for broadest coverage, OpenRouter for free tier, LiteLLM for self-hosting)
Sign up and get your API key
Replace your base URL and API key in your existing code
Access 300-627+ models through one endpoint

The migration takes under 5 minutes and eliminates multi-vendor complexity entirely.

Last updated: April 2026.

Top 10 AI API Aggregator Platforms in 2026: One Key, Hundreds of Models

Jenny Met — Thu, 02 Apr 2026 07:18:44 +0000

Managing separate API keys for OpenAI, Anthropic, Google, and DeepSeek is one of the biggest pain points for AI developers in 2026. AI API aggregator platforms solve this by providing a single endpoint and API key to access hundreds of models.

I tested 10 platforms over the past month across three criteria: model coverage, pricing transparency, and developer experience. Here's the ranking.

What Is an AI API Aggregator?

An AI API aggregator (also called an AI API gateway or unified AI API) sits between your application and multiple AI providers. You get:

One API key for all models
One billing account instead of managing 5-10 provider accounts
Unified API format (usually OpenAI-compatible)
Automatic failover when a provider goes down

The 10 Best AI API Aggregator Platforms (Ranked)

1. Crazyrouter

Website: crazyrouter.com

Crazyrouter provides access to 627+ models across 102 model families from 20+ providers. It stands out for having the broadest model coverage of any aggregator I tested, including hard-to-find models from ByteDance (Doubao, Seedance), Alibaba (Qwen3), and xAI (Grok).

Feature	Detail
Models	627+ across 102 families
API Format	OpenAI, Anthropic, and Gemini compatible
Pricing	Pay-as-you-go, below official rates
Free Credit	$0.20 on signup
Global Nodes	7 regions (US, Japan, Korea, UK, HK, PH, RU)
Capabilities	Chat, vision, image gen, video gen, music, TTS, STT, embeddings

Why it ranks #1: Broadest model coverage, multi-format API support (not just OpenAI-compatible), and consistently lower pricing than going direct. The 7 global edge nodes also give it the best latency profile for international teams.

Quick start:

from openai import OpenAI
client = OpenAI(
    base_url="https://crazyrouter.com/v1",
    api_key="sk-your-crazyrouter-key"
)
response = client.chat.completions.create(
    model="gpt-5",
    messages=[{"role": "user", "content": "Hello"}]
)

2. OpenRouter

Website: openrouter.ai

The most well-known aggregator with 300+ models. Strong community, good documentation, and OAuth support for user-facing apps.

Feature	Detail
Models	300+
API Format	OpenAI-compatible
Pricing	Pay-per-token + 5% BYOK fee
Free Models	Yes (limited selection)

Strengths: Large community, free tier, well-documented.
Weaknesses: 10-30% markup on popular models, US-only infrastructure, no native Anthropic/Gemini format support.

3. AIMLAPI

Website: aimlapi.com

Offers 400+ models covering chat, image, video, audio, and 3D generation. Good for teams that need multimodal capabilities beyond text.

Feature	Detail
Models	400+
API Format	OpenAI-compatible
Pricing	Pay-as-you-go
Capabilities	Chat, image, video, audio, voice, 3D

Strengths: Broad multimodal coverage, competitive pricing.
Weaknesses: Less established community, documentation could be better.

4. Portkey

Website: portkey.ai

Enterprise-focused gateway with strong observability, prompt management, and compliance features. Best for teams that need production-grade monitoring.

Feature	Detail
Models	100+ via provider integrations
API Format	OpenAI-compatible
Pricing	Free tier + paid plans
Key Feature	Built-in observability and prompt versioning

Strengths: Enterprise governance, detailed analytics, prompt management.
Weaknesses: Smaller model catalog than pure aggregators, enterprise pricing can be steep.

5. LiteLLM

Website: litellm.ai

Open-source Python library that provides a unified interface to 100+ LLMs. Can be self-hosted or used as a managed proxy.

Feature	Detail
Models	100+
API Format	OpenAI-compatible
Pricing	Free (open-source) or managed plans
Key Feature	Self-hosting option

Strengths: Open-source, self-hostable, no vendor lock-in.
Weaknesses: Requires infrastructure management if self-hosted, higher latency overhead than compiled alternatives.

6. Eden AI

Website: edenai.co

Specializes in multimodal AI APIs beyond just LLMs — OCR, document parsing, image analysis, translation, and more.

Feature	Detail
Models	500+ (including specialized AI services)
API Format	Custom unified API
Pricing	Pay-as-you-go with automatic cost optimization

Strengths: Broadest multimodal coverage including non-LLM services.
Weaknesses: Not OpenAI-compatible format, steeper learning curve.

7. Helicone

Website: helicone.ai

Primarily an observability platform that also functions as an AI gateway. Best for teams that prioritize logging and cost tracking.

Feature	Detail
Models	Routes to major providers
API Format	OpenAI-compatible proxy
Pricing	Free tier available
Key Feature	Request logging and cost analytics

Strengths: Free observability, easy integration as a proxy layer.
Weaknesses: Not a full aggregator — you still need provider keys.

8. Bifrost (by Maxim AI)

Website: getmaxim.ai

High-performance open-source gateway built in Go. Achieves extremely low latency overhead (11 microseconds at 5,000 RPS).

Feature	Detail
Models	15+ providers
API Format	OpenAI-compatible
Pricing	Open-source (free)
Key Feature	Ultra-low latency, automatic failover

Strengths: Best raw performance, open-source, self-hostable.
Weaknesses: Smaller model catalog, requires infrastructure setup.

9. SiliconFlow

Website: siliconflow.com

All-in-one AI cloud platform combining model hosting, inference optimization, and API aggregation. Strong in the Asian market.

Feature	Detail
Models	Major LLMs + open-source models
API Format	OpenAI-compatible
Pricing	Competitive, especially for Asian providers

Strengths: Optimized inference, good Asian model coverage.
Weaknesses: Less established in Western markets.

10. Unified.to

Website: unified.to

Provides unified APIs across multiple categories (not just AI), including CRM, HRIS, and accounting integrations alongside generative AI.

Feature	Detail
Models	Major LLM providers
API Format	Custom unified API
Pricing	Subscription-based

Strengths: Broad integration coverage beyond AI.
Weaknesses: AI is one of many features, not the primary focus.

Comparison Table

Platform	Models	OpenAI Compatible	Self-Host	Free Tier	Best For
Crazyrouter	627+	✅ + Anthropic + Gemini	❌	$0.20 credit	Broadest model access, international teams
OpenRouter	300+	✅	❌	✅	Community, free models
AIMLAPI	400+	✅	❌	✅	Multimodal (image/video/audio)
Portkey	100+	✅	❌	✅	Enterprise observability
LiteLLM	100+	✅	✅	✅ (OSS)	Self-hosted deployments
Eden AI	500+	❌	❌	✅	Non-LLM AI services
Helicone	Proxy	✅	✅	✅	Logging and cost tracking
Bifrost	15+ providers	✅	✅	✅ (OSS)	Ultra-low latency
SiliconFlow	Major LLMs	✅	❌	✅	Asian market, inference optimization
Unified.to	Major LLMs	❌	❌	❌	Multi-category integrations

How to Choose

Maximum model coverage: Crazyrouter (627+ models) or AIMLAPI (400+)
Free experimentation: OpenRouter (free tier with select models)
Enterprise compliance: Portkey (governance + observability)
Self-hosting required: LiteLLM or Bifrost
Beyond text (OCR, translation): Eden AI
Best performance: Bifrost (lowest latency) or Crazyrouter (global edge nodes)
Cost tracking: Helicone (free observability layer)

Getting Started

Most aggregators support the OpenAI SDK format. Here's the general pattern — just swap the base URL and API key:

from openai import OpenAI

# Works with Crazyrouter, OpenRouter, AIMLAPI, Portkey, LiteLLM, etc.
client = OpenAI(
    base_url="https://<aggregator-endpoint>/v1",
    api_key="your-aggregator-key"
)

response = client.chat.completions.create(
    model="gpt-5",  # or claude-sonnet-4.6, gemini-2.5-pro, etc.
    messages=[{"role": "user", "content": "Hello!"}]
)
print(response.choices[0].message.content)

The migration from direct provider APIs typically takes under 5 minutes.

Last updated: April 2026. Pricing and model counts change frequently — check each platform's website for current details.

2026'da En İyi 10 AI API Proxy: Geliştiriciler İçin Kapsamlı Karşılaştırma

Jenny Met — Mon, 30 Mar 2026 07:45:06 +0000

2026'da En İyi 10 AI API Proxy: Geliştiriciler İçin Kapsamlı Karşılaştırma

Ticari, açık kaynak ve kendi sunucunuzda barındırabileceğiniz 10 AI API proxy'sini gerçek fiyatlarla, çalışan kod örnekleriyle ve özellik matrisiyle karşılaştırdık.

AI API Proxy Nedir?

AI API proxy, uygulamanız ile AI sağlayıcıları (OpenAI, Anthropic, Google vb.) arasında yer alan bir ara katmandır. Kimlik doğrulama, yönlendirme, hız sınırlama, önbellekleme ve yedekleme (failover) işlemlerini yönetir. Her sağlayıcı için ayrı API anahtarı ve SDK yönetmek yerine tek bir endpoint kullanırsınız.

Neden ihtiyacınız var:

Tek anahtar, tüm modeller — 5 farklı sağlayıcı hesabıyla uğraşmayın
Maliyet optimizasyonu — Basit görevler için ucuz modellere yönlendirin, tekrar eden sorguları önbellekleyin
Güvenilirlik — Sağlayıcı çöktüğünde otomatik geçiş (düşündüğünüzden sık oluyor)
Gözlemlenebilirlik — Tüm sağlayıcılardaki maliyet, gecikme ve token kullanımını takip edin

Hızlı Karşılaştırma

Proxy	Tür	Model Sayısı	Fiyat	Kendi Sunucu	En İyi Kullanım
Crazyrouter	Ticari	627+	Resmi fiyatın ~%55'i	❌	En ucuz çok modlu erişim
LiteLLM	Açık Kaynak	100+	Ücretsiz	✅	Kendi sunucunuzda proxy
OpenRouter	Ticari	300+	Resmi + %10-30	❌	Hızlı prototipleme
Portkey	Ticari	1.600+ (BYOK)	Ücretsiz–$49/ay	✅	Kurumsal yönetişim
One API	Açık Kaynak	40+	Ücretsiz	✅	Yönetim paneli
Helicone	Ticari	BYOK	Ücretsiz–$20/ay	✅	Gözlemlenebilirlik
NGINX AI Proxy	Açık Kaynak	Yapılandırma	Ücretsiz	✅	NGINX kullanıcıları
Cloudflare AI GW	Ticari	BYOK	Ücretsiz	❌	Edge önbellekleme
Kong AI Gateway	Ticari	Eklenti bazlı	Kurumsal	✅	Kong kullanıcıları
Unify AI	Ticari	80+	Kullandıkça öde	❌	Akıllı yönlendirme

Ticari Proxy'ler

1. Crazyrouter — En Ucuz Çok Modlu API Proxy

Web sitesi: crazyrouter.com

Crazyrouter, resmi API fiyatlarının yaklaşık %55'ine 627+ model sunuyor. Onu benzersiz kılan şey: LLM + görüntü + video + müzik üretimini tek bir endpoint'ten sunan tek gateway olması.

Desteklenen modeller:

LLM: GPT-5/5.2, Claude Opus 4.6/Sonnet 4.6, Gemini 3 Pro, DeepSeek V3.2, Grok 4, Qwen 3
Görüntü: DALL-E 3, Midjourney, Flux Pro, Stable Diffusion 3.5
Video: Sora 2, Kling V2.6, Veo 3, Runway Gen4
Müzik: Suno V4

Fiyat karşılaştırması (Türk Lirası'yla düşünüldüğünde fark çok büyük):

Model	Resmi Fiyat	Crazyrouter	Tasarruf
GPT-5.2	$3 / $12 (1M token)	~$1,65 / $6,60	%45
Claude Opus 4.6	$15 / $75	~$8,25 / $41,25	%45
Claude Sonnet 4.6	$3 / $15	~$1,65 / $8,25	%45
Gemini 3 Pro	$1,25 / $10	~$0,69 / $5,50	%45

İki satır kod değişikliğiyle geçiş:

from openai import OpenAI

client = OpenAI(
    base_url="https://crazyrouter.com/v1",  # ← Sadece burayı değiştirin
    api_key="your-crazyrouter-key"           # ← Ve burayı
)

response = client.chat.completions.create(
    model="gpt-5-mini",
    messages=[{"role": "user", "content": "AI API proxy nedir? Bir cümleyle açıkla."}]
)
print(response.choices[0].message.content)

Gerçek test sonucu (Mart 2026):

{
  "model": "gpt-5-mini",
  "usage": {"prompt_tokens": 17, "completion_tokens": 37, "total_tokens": 54}
}

✅ En ucuz fiyat (~%55), 627+ model (görüntü/video/müzik dahil), OpenAI + Anthropic + Gemini uyumlu
❌ Kendi sunucunuzda barındırma yok

2. OpenRouter — En Popüler Varsayılan Seçenek

300+ model ve bazı modeller için ücretsiz katman. Ancak resmi fiyatların üzerine %10-30 ek ücret alıyor. Prototipleme için önemli değil, ölçekte ise ciddi fark yaratıyor.

✅ En büyük topluluk, bazı ücretsiz modeller
❌ %10-30 ek ücret, sadece LLM (görüntü/video yok)

3. Portkey — Kurumsal Kontrol Paneli

SOC 2 uyumluluğu, RBAC, denetim günlükleri ve koruma bariyerleri (PII algılama, içerik filtreleme) gereken ekipler için.

✅ En kapsamlı yönetişim, SOC 2
❌ BYOK (token indirimi yok), karmaşık kurulum

4. Helicone — Gözlemlenebilirlik Odaklı

AI API çağrılarınız için Datadog. Tek satır entegrasyon, 100K istek/ay ücretsiz.

✅ En iyi AI gözlemlenebilirliği
❌ Model toplayıcı değil (BYOK)

5-7. Diğer Ticari Seçenekler

Proxy	Odak	Kimin İçin
Cloudflare AI GW	Ücretsiz edge önbellekleme	Cloudflare kullanıcıları
Unify AI	Benchmark bazlı otomatik model seçimi	Deney yapan ekipler
Kong AI GW	Kong API Gateway'e AI eklentisi	Zaten Kong kullananlar

Açık Kaynak / Kendi Sunucunuzda

8. LiteLLM — Açık Kaynak Birincisi (18K+ Yıldız)

GitHub: github.com/BerriAI/litellm

En popüler açık kaynak AI API proxy. Python tabanlı, 100+ sağlayıcı, OpenAI uyumlu endpoint.

pip install litellm && litellm --config config.yaml

✅ MIT lisans, verileriniz sunucunuzdan çıkmaz
❌ Altyapıyı siz yönetiyorsunuz, BYOK

9. One API — En İyi Yönetim Paneli (20K+ Yıldız)

GitHub: github.com/songquanpeng/one-api

Web yönetim paneli, çoklu kiracı desteği, token kotası, kanal dengeleme. Docker ile tek tıkla kurulum.

✅ En iyi yönetim arayüzü, çoklu kiracı
❌ İngilizce topluluğu küçük

10. NGINX AI Proxy — NGINX Deneyimlilere

NGINX artık AI'ya özel proxy yapılandırmalarını destekliyor — SSE akışı, istek dönüşümü, yük dengeleme.

✅ Mevcut NGINX deneyimini kullanır
❌ Manuel yapılandırma, maliyet takibi yok

Özellik Matrisi

Özellik	Crazyrouter	LiteLLM	OpenRouter	Portkey	One API	Helicone
Model	627+	100+	300+	1.600+	40+	BYOK
Fiyat indirimi	~%45	Yok	-%10-30	Yok	Yok	Yok
Görüntü/Video/Müzik	✅	❌	❌	❌	❌	❌
Kendi sunucu	❌	✅	❌	✅	✅	✅
Yönetim paneli	Var	✅	❌	✅	✅	✅
Akıllı yönlendirme	❌	Temel	❌	✅	❌	❌
OpenAI uyumlu	✅	✅	✅	✅	✅	✅

Nasıl Seçmeli?

3 soru ile karar verin:

S1: En çok neye önem veriyorsunuz?

Maliyet tasarrufu → Crazyrouter (~%45 indirim, TL ile düşünüldüğünde büyük fark)
Veri güvenliği → LiteLLM veya One API (kendi sunucunuzda)
Kurumsal uyumluluk → Portkey (SOC 2)

S2: Görüntü/video/müzik üretimi de gerekiyor mu?

Evet → Crazyrouter (tek çok modlu gateway)
Hayır → Herhangi biri

S3: Yönetim paneli istiyor musunuz?

Evet → One API (en iyi arayüz) veya Portkey
Hayır → LiteLLM (en hafif)

SSS

AI API proxy ile AI API gateway arasındaki fark nedir?

Pratikte birbirinin yerine kullanılıyor. Teknik olarak "proxy" istekleri iletir, "gateway" ise yönetim özellikleri (kimlik doğrulama, hız sınırlama, analitik) ekler. Bu listedeki tüm araçlar her ikisini de yapıyor.

Birden fazla AI proxy birlikte kullanılabilir mi?

Evet. Yaygın kalıp: model erişimi için Crazyrouter (en ucuz tokenler) + gözlemlenebilirlik için Helicone. Veya kendi sunucunuzda LiteLLM çalıştırıp, maliyet tasarrufu için Crazyrouter'a yönlendirme.

AI proxy'ler gecikme ekler mi?

Minimum — ticari proxy'ler için genellikle 5-20ms. Kendi sunucunuzdaki proxy'ler (LiteLLM, One API) aynı bölgede konuşlandırılırsa neredeyse sıfır gecikme ekler.

Üçüncü taraf proxy üzerinden API anahtarı göndermek güvenli mi?

Ticari proxy'ler (Crazyrouter, OpenRouter, Portkey) anahtarları sizin adınıza yönetir — sağlayıcı anahtarlarınızı onlar üzerinden göndermezsiniz. Maksimum güvenlik için LiteLLM veya One API ile kendi sunucunuzda barındırın.

Son güncelleme: Mart 2026. Fiyatlar ve model sayıları sık değişir — güncel bilgi için her platformun web sitesini kontrol edin.

Top 10 AI API Proxies & Reverse Proxies for Developers in 2026

Jenny Met — Mon, 30 Mar 2026 07:45:04 +0000

Top 10 AI API Proxies & Reverse Proxies for Developers in 2026

A hands-on comparison of 10 AI API proxies — commercial gateways, open-source solutions, and self-hosted options — with real pricing data, working code examples, and a feature matrix.

What Is an AI API Proxy?

An AI API proxy sits between your application and AI providers (OpenAI, Anthropic, Google, etc.), handling authentication, routing, rate limiting, caching, and failover. Instead of managing separate API keys and SDKs for each provider, you hit one endpoint.

Think of it as NGINX for AI models — but purpose-built for the unique challenges of LLM traffic: streaming responses, token-based billing, model-specific quirks, and multi-provider failover.

Why you need one:

One API key, many models — Stop juggling 5 different provider accounts
Cost optimization — Route to cheaper models for simple tasks, cache repeated queries
Reliability — Automatic failover when a provider goes down (it happens more than you think)
Observability — Track costs, latency, and token usage across all providers
Security — Centralize API key management instead of scattering keys across services

Quick Comparison

Proxy	Type	Models	Pricing	Self-Host	Best For
Crazyrouter	Commercial	627+	~55% of official	❌	Cheapest multi-modal access
LiteLLM	Open Source	100+ providers	Free	✅	Self-hosted LLM proxy
OpenRouter	Commercial	300+	Official + 10-30%	❌	Quick prototyping
Portkey	Commercial	1,600+ (BYOK)	Free–$49/mo	✅	Enterprise governance
One API	Open Source	40+ providers	Free	✅	Chinese dev community
Helicone	Commercial	BYOK	Free–$20/mo	✅	Observability layer
NGINX AI Proxy	Open Source	Config-based	Free	✅	Existing NGINX users
Cloudflare AI GW	Commercial	BYOK	Free	❌	Edge caching
Kong AI Gateway	Commercial	Plugin-based	Enterprise	✅	Existing Kong users
Unify AI	Commercial	80+	Pay-per-token	❌	Benchmark-driven routing

Commercial Proxies

1. Crazyrouter — Cheapest Multi-Modal API Proxy

Website: crazyrouter.com

Crazyrouter is the most aggressive on pricing: roughly 55% of official API prices across 627+ models. But what makes it unique among proxies is multi-modal coverage — it's the only gateway that handles LLM, image generation, video generation, and music generation through a single endpoint.

Models covered:

LLMs: GPT-5/5.2, Claude Opus 4.6/Sonnet 4.6, Gemini 3 Pro, DeepSeek V3.2/R1, Grok 4, Qwen 3
Image: DALL-E 3, Midjourney, Flux Pro, Stable Diffusion 3.5
Video: Sora 2, Kling V2.6, Veo 3, Runway Gen4
Music: Suno V4

Pricing comparison:

Model	Official	Crazyrouter	Savings
GPT-5.2	$3.00 / $12.00 per 1M tokens	~$1.65 / $6.60	45%
Claude Opus 4.6	$15.00 / $75.00	~$8.25 / $41.25	45%
Claude Sonnet 4.6	$3.00 / $15.00	~$1.65 / $8.25	45%
Gemini 3 Pro	$1.25 / $10.00	~$0.69 / $5.50	45%

Drop-in replacement — change two lines:

from openai import OpenAI

client = OpenAI(
    base_url="https://crazyrouter.com/v1",
    api_key="your-crazyrouter-key"
)

response = client.chat.completions.create(
    model="gpt-5-mini",
    messages=[{"role": "user", "content": "What is an AI API proxy? One sentence."}]
)
print(response.choices[0].message.content)

Tested response (March 2026):

{
  "model": "gpt-5-mini",
  "choices": [{
    "message": {
      "content": "An AI API proxy is an intermediary service that routes and transforms requests and responses between applications and AI providers while handling authentication, security, rate limiting, caching, logging, and policy enforcement."
    },
    "finish_reason": "stop"
  }],
  "usage": {"prompt_tokens": 17, "completion_tokens": 37, "total_tokens": 54}
}

Also natively supports Anthropic SDK format and Google Gemini format — no forced conversion to OpenAI format.

✅ Cheapest pricing (~55% of official), 627+ models including image/video/music, OpenAI + Anthropic + Gemini compatible, 7 global regions
❌ No self-hosting, smaller community than OpenRouter

2. OpenRouter — The Popular Default

Website: openrouter.ai

OpenRouter is the most widely known AI API proxy. 300+ models, a free tier for some models, and a large community. It's the "safe default" choice.

The catch: 10-30% markup on top of official prices. For prototyping this doesn't matter. At scale, it adds up fast — a $10,000/month API bill becomes $11,000-$13,000.

✅ Largest community, free tier for some models, easy to start
❌ 10-30% markup, LLM only (no image/video), no self-hosting

3. Portkey — Enterprise Control Plane

Website: portkey.ai

Portkey positions itself as the "control plane for AI apps." If your team needs SOC 2 compliance, RBAC, audit logs, and guardrails (PII detection, content filtering), Portkey is the enterprise answer.

Key features:

1,600+ LLMs (BYOK — Bring Your Own Key)
Guardrails: PII detection, input/output validation
Distributed tracing, cost dashboards, latency monitoring
Prompt management with A/B testing
Automatic failover between providers

Pricing: Free (10K requests/month) → Pro $49/month → Enterprise custom

✅ Most comprehensive governance, SOC 2, open-source core
❌ BYOK (no token cost savings), complex setup, overkill for simple projects

4. Helicone — Observability-First Proxy

Website: helicone.ai

Helicone isn't a model aggregator — it's Datadog for AI API calls. One-line integration (change your base URL), and you get full request/response logging, cost tracking, semantic caching, and latency monitoring.

Pricing: Free (100K requests/month) → Pro $20/month

✅ Best AI observability, 100K free requests/month, one-line setup
❌ Not a model aggregator (BYOK), adds a proxy hop

5. Cloudflare AI Gateway — Edge Caching

Website: developers.cloudflare.com/ai-gateway

Free proxy layer running on Cloudflare's edge network. Provides caching, rate limiting, and basic analytics. If you're already using Cloudflare, this is a no-brainer addition.

✅ Free, global edge network, edge caching reduces repeated query costs
❌ BYOK (no cost savings), basic analytics, no smart routing

6. Unify AI — Benchmark-Driven Routing

Website: unify.ai

Unify's unique angle: instead of you picking the model, it automatically routes to the optimal model based on benchmarks, cost, and latency.

✅ Intelligent routing, data-driven decisions
❌ Only 80+ models, benchmark scores may not match your use case

7. Kong AI Gateway — For Kong Shops

Website: konghq.com

AI plugins for the popular Kong API Gateway. Natural extension if your org already uses Kong for API management.

✅ Reuses existing Kong infrastructure, enterprise-grade
❌ Pointless without Kong, complex AI-specific configuration

Open-Source / Self-Hosted Proxies

8. LiteLLM — The Go-To Open Source Proxy

GitHub: github.com/BerriAI/litellm (18K+ stars)

LiteLLM is the most popular open-source AI API proxy. Python-based, supports 100+ providers, OpenAI-compatible endpoint. If you want full control over your AI infrastructure with data staying on your servers, LiteLLM is the standard choice.

pip install litellm
litellm --config config.yaml

model_list:
  - model_name: gpt-5
    litellm_params:
      model: openai/gpt-5
      api_key: sk-xxx
  - model_name: claude-opus
    litellm_params:
      model: anthropic/claude-opus-4-6
      api_key: sk-ant-xxx

Features: virtual keys, budget management, rate limiting, load balancing, spend tracking.

✅ MIT license, data never leaves your infra, active community, cost tracking
❌ You manage the infrastructure, BYOK (no token discounts), LLM only

9. One API — Popular in Chinese Dev Community

GitHub: github.com/songquanpeng/one-api (20K+ stars)

One API is the most widely used open-source AI proxy in Chinese-speaking developer communities. It provides a web-based admin panel for managing multiple API keys, channels, and token quotas. Think of it as LiteLLM + admin dashboard, with a focus on key management and reselling scenarios.

Key features:

Web admin panel (manage keys, channels, quotas)
Multi-tenant support (create sub-accounts with token limits)
Channel balancing (distribute requests across multiple keys)
40+ providers supported
Docker one-click deployment

✅ Best admin UI among open-source options, multi-tenant, 20K+ GitHub stars
❌ Less active English community, some providers lag behind LiteLLM

10. NGINX AI Proxy — For NGINX Veterans

Blog: blog.nginx.org/blog/using-nginx-as-an-ai-proxy

NGINX now supports AI-specific proxy configurations — streaming SSE responses, request/response transformation for different providers, and load balancing across AI backends. If your team already runs NGINX, adding AI proxy capabilities is a natural extension.

✅ Leverages existing NGINX expertise, no new dependencies, battle-tested infrastructure
❌ Manual configuration, no built-in model routing or cost tracking, steep learning curve for AI-specific features

Feature Matrix

Feature	Crazyrouter	LiteLLM	OpenRouter	Portkey	One API	Helicone	CF GW
Models	627+	100+	300+	1,600+(BYOK)	40+	BYOK	BYOK
Price discount	~45%	None	-10-30%	None	None	None	None
Image/Video/Music	✅	❌	❌	❌	❌	❌	❌
Self-host	❌	✅	❌	✅	✅	✅	❌
Admin UI	Dashboard	✅	❌	✅	✅	✅	✅
Guardrails	❌	❌	❌	✅	❌	❌	❌
Smart routing	❌	Basic	❌	✅	❌	❌	❌
Caching	❌	✅	❌	✅	❌	✅	✅
OpenAI compatible	✅	✅	✅	✅	✅	✅	✅

How to Choose

Decision tree:

"I need the cheapest token prices" → Crazyrouter (~45% off official prices, widest model coverage)
"I need data to stay on my servers" → LiteLLM (open source, MIT, most providers) or One API (if you want an admin panel)
"I need enterprise governance" → Portkey (SOC 2, guardrails, RBAC, audit logs)
"I need to understand my AI spending" → Helicone (best observability, 100K free requests)
"I also need image/video/music generation" → Crazyrouter (only gateway covering full multi-modal spectrum)
"I'm already using Cloudflare/Kong/NGINX" → Add their AI gateway plugins to your existing stack

FAQ

What's the difference between an AI API proxy and an AI API gateway?

In practice, they're used interchangeably. Technically, a "proxy" forwards requests to AI providers, while a "gateway" adds management features (auth, rate limiting, analytics, routing). Every tool in this list does both — the distinction is marketing, not technical.

Can I use multiple AI proxies together?

Yes. A common pattern: use Crazyrouter for model access (cheapest tokens) + Helicone for observability (track what's happening). Or self-host LiteLLM as your proxy layer, routing to Crazyrouter for cost savings. See our cost optimization guide for architecture patterns.

Do AI proxies add latency?

Minimal — typically 5-20ms for commercial proxies (Crazyrouter, OpenRouter). Self-hosted proxies (LiteLLM, One API) add almost no latency if deployed in the same region as your app. Cloudflare AI Gateway can actually reduce latency through edge caching. For a deep dive, see our latency optimization guide.

Which proxy is best for production use?

Depends on your constraints. For cost: Crazyrouter. For control: LiteLLM. For enterprise: Portkey. For observability: Helicone. Many production systems use 2-3 of these together. Check our load balancing guide for production architecture patterns.

Is it safe to route API keys through a third-party proxy?

Commercial proxies (Crazyrouter, OpenRouter, Portkey) manage keys on your behalf — you don't send your provider keys through them. For BYOK proxies (Helicone, Cloudflare), your keys pass through their infrastructure — check their security practices. For maximum security, self-host with LiteLLM or One API.

Last updated: March 2026. Prices and model counts change frequently — check each platform's website for the latest.

OpenClaw Tutorial: Build Your Own Private AI Assistant from Scratch

Jenny Met — Mon, 30 Mar 2026 07:41:46 +0000

OpenClaw Tutorial: The Complete Guide to Building Your Private AI Assistant

Ever wanted your own AI assistant — not a web-based ChatGPT wrapper, but something running on your server, always on, talking to you through Telegram? Something you actually control?

That's exactly what OpenClaw does. It's an open-source AI assistant framework you can deploy on a VPS, Raspberry Pi, or even your laptop. Connect it to Telegram, WhatsApp, Discord — whatever you use. It supports 627+ AI models, the config is flexible, and getting started is easier than you'd think.

Let's walk through it step by step.

Step 1: Installation & Environment Setup (Node.js 22+, One-Line Install)

System Requirements

Before we start, make sure your environment checks these boxes:

OS: Linux (Ubuntu 22.04+ recommended), macOS, or WSL2
Node.js: 22.0 or higher (hard requirement)
RAM: 512MB minimum, 1GB+ recommended
Network: Internet access (for AI model API calls)

If you don't have Node.js 22+ yet, grab it with nvm:

curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.3/install.sh | bash
source ~/.bashrc
nvm install 22
node -v  # Should output v22.x.x

Why Node.js 22+?

OpenClaw relies on Node.js 22 features like native WebSocket support and improved ESM modules. Older versions will just throw errors. Trust me on this one.

One-Line Install

Environment ready? Install OpenClaw with a single command:

curl -fsSL https://openclaw.ai/install.sh | bash

This script will:

Detect your system environment and Node.js version
Install OpenClaw globally via npm
Create the default config directory ~/.openclaw/
Generate initial configuration files

After installation, run the onboarding wizard:

openclaw onboard

This interactive guide walks you through basic setup — workspace directory, default model, etc. If you prefer manual config (like me), skip it and edit the config file directly.

Step 2: Configure Your First AI Model (API Keys & 627+ Models via Crazyrouter)

Understanding the Config File

All OpenClaw configuration lives in one file:

~/.openclaw/openclaw.json

Despite the .json extension, it actually supports JSON5 — meaning you can write comments, use trailing commas, and single quotes. Developer-friendly from the start.

Here's a minimal working config:

{
  // OpenClaw minimal config
  agents: {
    defaults: {
      // Working directory for SOUL.md and other persona files
      workspace: "~/.openclaw/workspace",

      // AI model config
      model: {
        primary: "openai/gpt-4o",  // Format: provider/model
      },
    },
  },
}

Workspace Files Explained

Your workspace directory contains several key files that define your AI assistant's "soul":

File	Purpose
`SOUL.md`	Core personality and behavioral guidelines
`AGENTS.md`	Workflow and session rules
`USER.md`	Information about you (the user)
`TOOLS.md`	Tool usage notes and API keys
`IDENTITY.md`	Assistant identity (name, personality, etc.)

Edit these freely to customize your assistant. Write "be concise, no fluff" in SOUL.md and your assistant will keep things short.

Access 627+ Models via Crazyrouter

OpenClaw has built-in providers: openai, anthropic, google, openrouter, deepseek. You can plug in API keys directly.

But if you don't want to juggle multiple API keys, there's a simpler approach — use Crazyrouter as a unified gateway. One API key, 627+ models including GPT-4o, Claude 4, Gemini 2.5, DeepSeek R1, and more.

{
  agents: {
    defaults: {
      workspace: "~/.openclaw/workspace",
      model: {
        // Use Claude via Crazyrouter
        primary: "openai/anthropic/claude-sonnet-4-20250514",
      },
      providers: {
        openai: {
          apiKey: "sk-your-crazyrouter-key",
          baseURL: "https://crazyrouter.com/v1",
        },
      },
    },
  },
}

Multi-Model Fallback Strategy

You can also configure fallback models for automatic switching when the primary is unavailable:

{
  agents: {
    defaults: {
      model: {
        primary: "openai/anthropic/claude-sonnet-4-20250514",
        fallback: "openai/gpt-4o",        // Backup when primary is down
        fast: "openai/gpt-4o-mini",        // Cheap model for simple tasks
      },
    },
  },
}

Your assistant will automatically pick the best model based on task complexity and availability.

Step 3: Connect a Chat Platform (Telegram, WhatsApp, Discord)

Model configured — now let's get your assistant online by connecting it to a chat platform. OpenClaw supports Telegram, WhatsApp, Discord, and more.

Telegram Bot Setup (Recommended)

Telegram is the most recommended platform for OpenClaw — simplest config, most features.

Step 1: Create a Telegram Bot

Find @BotFather on Telegram
Send /newbot and follow the prompts
Copy the Bot Token (looks like 123456:ABC-DEF...)

Step 2: Add to Config

{
  agents: {
    defaults: {
      workspace: "~/.openclaw/workspace",
      model: {
        primary: "openai/anthropic/claude-sonnet-4-20250514",
      },
      providers: {
        openai: {
          apiKey: "sk-your-crazyrouter-key",
          baseURL: "https://crazyrouter.com/v1",
        },
      },
    },
  },
  channels: {
    telegram: {
      token: "your-bot-token",
      allowFrom: ["your-telegram-username"],  // Whitelist — only these users can chat
    },
  },
}

Security Note: Always Set allowFrom

allowFrom is a whitelist. Only listed users can talk to your bot. Without it, anyone can use your bot — and your API credits. Learn from my mistake.

WhatsApp & Discord

WhatsApp uses the WhatsApp Business API:

{
  channels: {
    whatsapp: {
      allowFrom: ["+1234567890"],  // Allowed phone numbers
    },
  },
}

Discord requires an Application and Bot from the Discord Developer Portal:

{
  channels: {
    discord: {
      token: "your-discord-bot-token",
      allowFrom: ["your-discord-user-id"],
    },
  },
}

Same pattern everywhere: get the platform token, add it to config, set the whitelist.

Step 4: Send Your First Message (Gateway & Control UI)

Start the Gateway

Everything configured? Fire up the OpenClaw Gateway:

openclaw gateway start

It starts on port 18789 by default. Check status with:

openclaw gateway status

If you see running, you're good. Port conflict? Change it in the config file.

The Control UI

OpenClaw ships with a web management interface:

openclaw dashboard

Open http://localhost:18789 in your browser. You'll see:

Chat interface: Talk to your AI assistant directly in the browser
Config management: Visual config editor
Logs: Real-time conversation logs and error output
Model switching: One-click model changes

Send Your First Message

Now open Telegram, find your bot, and send:

Hey! Who are you?

If everything's configured correctly, your AI assistant will reply within seconds. Congrats — your private AI assistant is live!

No reply? Troubleshoot in this order:

openclaw gateway status — confirm the service is running
Check the Bot Token in your config
Verify your username is in allowFrom
Check terminal output from openclaw gateway

CLI Quick Reference

Commands you'll use often:

# Service management
openclaw gateway start      # Start the service
openclaw gateway stop       # Stop the service
openclaw gateway restart    # Restart the service
openclaw gateway status     # Check status

# Setup & management
openclaw onboard            # Interactive setup wizard
openclaw dashboard          # Open web management UI

Wrapping Up

That's the whole tutorial: install OpenClaw → fill in the config → start the service. Three steps to your own private AI assistant.

For AI model access, if you don't want to deal with multiple API keys, give Crazyrouter a try. One key for 627+ models — OpenAI, Anthropic, Google, DeepSeek, all the major providers. Transparent pricing, pay-as-you-go, and free credits to get started. For solo devs and small teams, it eliminates the hassle of managing multiple accounts and pairs perfectly with OpenClaw.

Questions? File an issue on OpenClaw GitHub or join the community.

Happy hacking! 🚀