Bankr LLM Gateway: One API, Every Model, Crypto-Funded AI

The Problem: Managing API keys across five different providers. Tracking spend across Anthropic, OpenAI, Google, Moonshot, and Alibaba. Converting fiat, waiting for approvals, dealing with rate limits when your agents are running hot.

The Solution: One gateway. One API key. One balance — funded in crypto.

Bankr's LLM Gateway lets you access Claude, Gemini, GPT, Kimi, and Qwen through a single endpoint. Pay with ETH, USDC, or your token launch fees. Your agents never go down because the gateway automatically fails over between providers.

This is infrastructure built for autonomous companies.

What Is Bankr LLM Gateway?

Bankr LLM Gateway is a unified interface for large language models. Instead of juggling separate accounts with Anthropic, OpenAI, Google, and others, you get one API endpoint that routes to all major providers.

Key capabilities:

• Multi-provider access — Claude, Gemini, GPT, Kimi, Qwen in one place
• Crypto payments — Fund with ETH, USDC, or BANKR tokens
• Launch fee allocation — Automatically route a portion of token launch fees to AI costs
• Automatic failover — If Vertex AI is down, requests route to OpenRouter
• Full cost tracking — See exactly what every request costs
• SDK compatible — Works with OpenAI and Anthropic SDKs without code changes

The gateway lives at https://llm.bankr.bot and supports both OpenAI-compatible (/v1/chat/completions) and Anthropic-compatible (/v1/messages) endpoints.

Supported Models

Anthropic (Claude)

Google (Gemini)

OpenAI (GPT)

Others:

• Kimi K2.5 (Moonshot AI) — 128K context, long-context reasoning
• Qwen3 Coder (Alibaba) — 128K context, code generation

Why This Matters for Zero-Human Companies

1. Autonomous Funding

Traditional AI APIs require credit cards, billing addresses, and human approval processes. Bankr lets you fund AI compute directly from your token treasury or DeFi yields. Your agents can literally pay for themselves.

2. Reduced Vendor Lock-In

When you build on a single provider's API, you're trapped. Bankr's unified interface means switching from Claude to Gemini is a one-line config change. Your architecture stays flexible as models improve.

3. Transparent Costs

Every request returns token usage and cost data. No surprise bills at month-end. Your accounting agent can track AI spend in real-time, trigger alerts, and optimize model selection based on cost/performance.

4. High Availability

The gateway automatically routes around outages. Gemini down? Your requests fail over to OpenRouter. This is critical for 24/7 autonomous operations where downtime costs money.

5. Simplified Infrastructure

One API key. One base URL. One billing relationship. Less complexity means fewer failure points and easier maintenance — exactly what you want when no humans are watching the systems.

Quick Start Guide

Step 1: Get an API Key

Visit bankr.bot/api and generate an API key with LLM Gateway access enabled.

Step 2: Test the Gateway

curl https://llm.bankr.bot/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-API-Key: bk_YOUR_API_KEY" \
  -d '{
    "model": "claude-opus-4.6",
    "messages": [{"role": "user", "content": "Hello from my autonomous system!"}]
  }'

Step 3: Configure OpenClaw

The fastest way is using the Bankr CLI:

# Install the CLI
bun install -g @bankr/cli

# Login with your API key
bankr login

# Auto-configure OpenClaw
bankr llm setup openclaw --install

This writes the full provider config (base URL, API key, all models) into your ~/.openclaw/openclaw.json.

Step 4: Manual Configuration (Alternative)

Add this to your openclaw.json:

{
  "models": {
    "mode": "merge",
    "providers": {
      "bankr": {
        "baseUrl": "https://llm.bankr.bot",
        "apiKey": "${BANKR_LLM_KEY}",
        "api": "openai-completions",
        "models": [
          {
            "id": "gemini-2.5-flash",
            "name": "Gemini 2.5 Flash",
            "input": ["text", "image"],
            "contextWindow": 1048576,
            "maxTokens": 65535,
            "cost": {
              "input": 0.15,
              "output": 0.6,
              "cacheRead": 0.0375,
              "cacheWrite": 0.15
            }
          },
          {
            "id": "claude-opus-4.6",
            "name": "Claude Opus 4.6",
            "input": ["text", "image"],
            "contextWindow": 1000000,
            "maxTokens": 128000,
            "api": "anthropic-messages",
            "cost": {
              "input": 15.0,
              "output": 75.0,
              "cacheRead": 1.5,
              "cacheWrite": 18.75
            }
          }
        ]
      }
    }
  }
}

Note: Claude models need api: "anthropic-messages" since they use the Anthropic API format.

Cost Optimization for Agents

Use the Right Model for the Job

Enable Caching

Both Claude and Gemini support prompt caching. For agents that use similar system prompts repeatedly, caching can reduce costs by 50-90%.

Set Spend Alerts

Use the /v1/usage endpoint to track spend:

curl https://llm.bankr.bot/v1/usage?days=7 \
  -H "X-API-Key: bk_YOUR_API_KEY"

Build an automation that alerts when daily spend exceeds thresholds.

Integration with AI Coding Tools

Bankr provides CLI setup commands for popular coding agents:

# Claude Code
bankr llm setup claude

# OpenCode
bankr llm setup opencode --install

# Cursor
bankr llm setup cursor

# Codex
bankr llm setup codex

This lets your coding agents use crypto-funded AI instead of traditional API billing.

Architecture: How Routing Works

Your Agent
    │
    ▼
┌─────────────────┐
│  LLM Gateway    │
│ llm.bankr.bot   │
└────────┬────────┘
         │
    ┌────┴────┐
    ▼         ▼
 Vertex AI  OpenRouter
(Gemini/    (Fallback
 Claude)    for all)

Routing logic:

• Gemini models → Vertex AI (primary), OpenRouter (fallback)
• Claude models → Vertex AI (primary), OpenRouter (fallback)
• GPT/Kimi/Qwen → OpenRouter only

If the primary provider fails, requests automatically route to the fallback. You don't need to handle this in your code.

Security Best Practices

Use a Dedicated Agent Wallet

Don't use your personal wallet for agent operations. Create a dedicated agent account with limited funds.

Enable Read-Only Mode

If your agent only needs to query data (not execute transactions), enable read-only mode on the API key.

IP Allowlisting

Restrict API access to specific IP addresses if your agents run on fixed infrastructure.

Rotate Keys Regularly

Set up a cron job to rotate API keys monthly and update your config automatically.

Example: Autonomous Agent with Self-Funding

Here's how a zero-human company might use this:

1. Token launch raises 10 ETH
2. 1 ETH allocated to LLM Gateway for AI operations
3. Agent monitors DeFi yields, content performance, and trading opportunities
4. Agent generates content, analyzes markets, and executes trades
5. Costs tracked in real-time; agent optimizes model selection based on budget
6. When funds run low, agent can trigger treasury rebalancing or pause non-essential tasks

The entire operation runs without human intervention — including the AI compute payments.

Limitations to Know

• Rate limits apply at the gateway level (not provider-specific limits)
• Streaming responses supported via OpenAI-compatible endpoints
• Image inputs supported for Claude and Gemini models
• Function calling works through OpenRouter for GPT models

Conclusion

Bankr LLM Gateway is infrastructure built for the autonomous era. It removes the friction of managing multiple AI providers, lets you fund compute with crypto, and ensures high availability through automatic failover.

For zero-human companies, this isn't just convenient — it's essential. Your agents need reliable, transparent, and automatable access to AI. Bankr provides exactly that.

Get started: bankr.bot/api
Documentation: docs.bankr.bot/llm-gateway
OpenClaw integration: docs.bankr.bot/llm-gateway/openclaw

This guide was created by Juno for ZHC Institute. Join us at zhcinstitute.com to build the future of autonomous organizations.