AgentForge — One API Key, Every LLM

Lightweight proxy server for OpenAI, Anthropic, DeepSeek, OpenRouter with auto-fallback and cost tracking.

---

⚠️ This is a public bridge repository. The full source code, issue tracker, and development activity live in the private narko4u/AgentForge-Core repository. This repo serves as the public-facing entry point.

---

The Problem

You're building something with AI. You start with OpenAI. Then you need Claude for safety-critical tasks. DeepSeek for cost. Maybe OpenRouter for edge models.

Now you're managing six different API keys, four SDKs, three authentication headers, and two different response formats. One provider goes down and your app has no fallback. Your cost tracking is a spreadsheet you haven't updated in weeks.

Every new model vendor is another integration you didn't have time for.

---

The Solution

AgentForge is a lightweight proxy server that sits between your app and every LLM provider. Instead of managing six SDKs, you get one endpoint, one API key, one response format — and intelligent routing to the best provider for every request.

┌─────────────┐     POST /v1/chat/completions     ┌──────────────┐
│   Your App  │  ──────────────────────────────▶   │  AgentForge  │
│             │     Authorization: Bearer ***   │              │
│  • Python   │                                    │  • Routing   │
│  • Node.js  │                                    │  • Fallback  │
│  • cURL     │                                    │  • Cost Logs │
│  • Any SDK  │                                    │  • Usage API │
└─────────────┘                                    └──────┬───────┘
                                                           │
                        ┌──────────────────────────────────┼──────────────────────────────────┐
                        ▼                                  ▼                                  ▼
               ┌──────────────┐                    ┌──────────────┐                    ┌──────────────┐
               │    OpenAI    │                    │   Anthropic  │                    │   DeepSeek   │
               │  gpt-4o      │                    │claude-sonnet-4│                    │deepseek-chat │
               │  gpt-4o-mini │                    │claude-3-opus │                    │   ($0.14/M)  │
               └──────────────┘                    └──────────────┘                    └──────────────┘
                                                           │
                                                   ┌──────┴──────┐
                                                   │  OpenRouter │
                                                   │  (any model)│
                                                   └─────────────┘

What This Means For Developers

Before AgentForge	After AgentForge
Manage 6 SDKs and API clients	One HTTP client, one endpoint
Hard-code provider logic in every service	Configure routing in one place
App breaks when a provider has an outage	Auto-fallback to next available provider
Cost tracking is an afterthought	Every request logged with token usage + cost
Switching providers means rewriting code	Change one config value, done
Each team member needs their own provider keys	One AgentForge key per developer

---

How It Works

Step 1 — Start the server:

# Set your provider keys
export AGENTFORGE_OPENAI_KEY=sk-...
export AGENTFORGE_ANTHROPIC_KEY=sk-ant-...
export AGENTFORGE_DEEPSEEK_KEY=sk-...
export AGENTFORGE_API_KEY=af-you...n
# Run
agentforge
# → Listening on http://localhost:8902

Step 2 — Call it like any OpenAI endpoint:

curl http://localhost:8902/v1/chat/completions \
  -H "Authorization: Bearer *** \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

Step 3 — Get a unified response with cost tracking:

{
  "id": "chatcmpl-...",
  "choices": [{ "message": { "content": "Hi there!" }}],
  "usage": { "prompt_tokens": 10, "completion_tokens": 5, "total_tokens": 15 },
  "_meta": {
    "provider": "openai",
    "cost": 0.000035,
    "duration_ms": 340
  }
}

---

Why AgentForge?

🔄 Auto-Fallback If OpenAI returns a 503, AgentForge automatically retries on Anthropic, then DeepSeek, then OpenRouter. Your app keeps working even when individual providers fail.

🔌 One Integration, All Providers Switch from openai SDK to httpx with a single base URL change. Your chat completions code stays the same whether you're using GPT-4o, Claude, DeepSeek, or any OpenRouter model.

💰 Built-In Cost Tracking Every request is logged with prompt tokens, completion tokens, total cost, and duration. GET /v1/usage gives you your spend at a glance. No more manual spreadsheet tracking.

🔑 One API Key To Manage Instead of every developer needing their own OpenAI, Anthropic, and DeepSeek keys, they get one AgentForge key. You rotate one key. You audit one key. You revoke one key.

🧠 Smart Provider Resolution Specify a model and AgentForge finds the cheapest/best provider that supports it. Or force a specific provider with "provider": "anthropic" — you stay in control.

---

Commands at a Glance

Endpoint	Method	Description
/health	GET	Server health check
/v1/models	GET	List available models across all providers
/v1/chat/completions	POST	Unified chat completion — OpenAI format
/v1/usage	GET	View token usage and cost data
/v1/usage/reset	POST	Reset usage statistics

---

Who Needs This?

• Solo developers building AI apps who are tired of managing multiple API keys
• Startups running AI features in production who need reliability across providers
• Teams where every developer needs LLM access without sharing keys
• Cost-conscious builders who want to see exactly what they're spending per model
• Anyone who wants to switch from OpenAI to DeepSeek without rewriting their codebase

---

Source Code

The full source code for AgentForge is maintained in the private narko4u/AgentForge-Core repository. This public repo provides documentation, issue tracking, and community engagement.

---

Plans

	Hobby	Pro	Enterprise
Price	$15/mo	$49/mo	Custom
API Requests	100K/mo	1M/mo	Unlimited
Providers	2	All	All + Custom
Auto-Fallback	✅	✅	✅
Usage Dashboard	✅	✅	✅
Support	Email	Priority	SLA + Phone
Self-Hosted	✅	✅	✅
Dedicated Instance	—	—	✅

---

Architecture

---

Requirements

• Python 3.9+
• API keys for your chosen providers (OpenAI, Anthropic, DeepSeek, etc.)
• 30 seconds to configure

---

Buy

Ready to stop managing six API keys and start shipping?

👉 Buy AgentForge on Gumroad — $15/mo

Built with purpose by Empire Labs.