AI Gateway & Cloud Sandboxes.
Built for Autonomous Agents.
VERBAREX provides high-throughput, low-latency AI inference endpoints with 1,000,000 token context windows, alongside dedicated, hardware-isolated Linux MicroVM Cloud Sandboxes. Autonomous agents reason through frontier models and execute shell commands, Python scripts, and browser tools with hardware safety and zero cold starts.
https://verbarex.com/api/v1 with your VERBAREX API key.
Python SDK (OpenAI Compatible)
from openai import OpenAI
client = OpenAI(
base_url="https://verbarex.com/api/v1",
api_key="vx_live_YOUR_API_KEY"
)
response = client.chat.completions.create(
model="deepseek-v4.1-flash",
messages=[
{"role": "system", "content": "You are a senior software architect."},
{"role": "user", "content": "Explain microVM hardware isolation in 2 clear bullet points."}
],
temperature=0.3
)
print(response.choices[0].message.content)
cURL Inference Endpoint
curl -X POST https://verbarex.com/api/v1/chat/completions \
-H "Authorization: Bearer vx_live_YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek-v4.1-flash",
"messages": [
{"role": "user", "content": "List 3 advantages of Linux microVMs for autonomous agents."}
],
"temperature": 0.3
}'
Coding Agents & AI IDE Integration
VERBAREX AI Gateway provides native out-of-the-box API support for every major coding agent and AI-powered IDE in the world. The gateway supports both universal OpenAI endpoints (https://verbarex.com/v1/chat/completions) and native Anthropic Messages API (https://verbarex.com/v1/messages with x-api-key).
• OpenAI Compatible Base URL:
https://verbarex.com/v1 (or https://verbarex.com/api/v1)• Anthropic Messages Base URL:
https://verbarex.com/v1 (or https://verbarex.com/api/v1)• Supported Headers:
Authorization: Bearer vx_live_... or x-api-key: vx_live_...
Cursor
In Cursor → Settings → Models → OpenAI API Key, toggle Override OpenAI Base URL:
Base URL: https://verbarex.com/v1
API Key: vx_live_YOUR_API_KEY
Models to enable:
• deepseek-v4-1-flash (Default high-speed coder)
• claude-3-5-sonnet (Frontier architecture)
• gpt-4o (Omnimodal flagship)
• glm-5-3-flash (128K context reasoning)
• qwen3-8-flash (1M context code reasoning)
Claude Code (Anthropic CLI)
Anthropic's official claude terminal coding agent connects seamlessly via native Anthropic Messages API support:
export ANTHROPIC_BASE_URL="https://verbarex.com/v1"
export ANTHROPIC_API_KEY="vx_live_YOUR_API_KEY"
# Run Claude Code in your project directory
claude
Windsurf (Codeium)
In Windsurf Settings → AI Provider → Custom OpenAI Compatible:
Base URL: https://verbarex.com/v1
API Key: vx_live_YOUR_API_KEY
Model Name: deepseek-v4-1-flash
Aider
Run Aider in your terminal with your preferred model:
export OPENAI_BASE_URL="https://verbarex.com/v1"
export OPENAI_API_KEY="vx_live_YOUR_API_KEY"
# Start Aider
aider --model openai/deepseek-v4-1-flash
Cline & Roo Code (VS Code Extensions)
In the Cline / Roo Code settings panel, choose either OpenAI Compatible or Anthropic:
Provider: OpenAI Compatible (or Anthropic)
Base URL: https://verbarex.com/v1
API Key: vx_live_YOUR_API_KEY
Model ID: claude-3-5-sonnet (or deepseek-v4-1-flash)
Continue.dev (~/.continue/config.json)
{
"models": [
{
"title": "Verbarex DeepSeek V4.1 Flash",
"provider": "openai",
"model": "deepseek-v4-1-flash",
"apiBase": "https://verbarex.com/v1",
"apiKey": "vx_live_YOUR_API_KEY"
},
{
"title": "Verbarex Claude 3.5 Sonnet",
"provider": "anthropic",
"model": "claude-3-5-sonnet",
"apiBase": "https://verbarex.com/v1",
"apiKey": "vx_live_YOUR_API_KEY"
}
]
}
Goose CLI (Block)
goose configure
# Select provider: openai
# Host: https://verbarex.com/v1
# API Key: vx_live_YOUR_API_KEY
# Model: deepseek-v4-1-flash
Bearer API Keys
All API requests authenticate using HTTP Bearer authentication in the Authorization header:
Authorization: Bearer vx_live_...
You can create and manage API keys in the Console → API Keys section. Plaintext keys are shown only once upon creation, hashed securely at rest using salted SHA-256, and can be revoked instantly.
Key Scopes & Service Separation
/api/v1/chat/completions, /api/v1/models). Cannot execute sandbox commands.
/modelsActive Reasoning & Code Models
Returns the catalog of active model endpoints available to your organization, including context windows and token pricing:
curl https://verbarex.com/api/v1/models \
-H "Authorization: Bearer $VERBAREX_API_KEY"
/chat/completionsCreate Chat Completion
Executes model inference against frontier reasoning models. Fully compatible with OpenAI schema including model, messages, temperature, max_tokens, and stream.
{
"model": "deepseek-v4.1-flash",
"messages": [
{"role": "system", "content": "You are a helpful coding assistant."},
{"role": "user", "content": "Write a Python one-liner to fetch the public IP."}
],
"temperature": 0.2
}
Responses return standard OpenAI structure plus exact token metrics and execution latency:
{
"id": "chatcmpl-01M5...",
"object": "chat.completion",
"created": 1793110596,
"model": "deepseek-v4-1-flash",
"choices": [
{
"index": 0,
"message": {
"role": "assistant",
"content": "import urllib.request, json; print(json.loads(urllib.request.urlopen('https://httpbin.org/ip').read())['origin'])"
},
"finish_reason": "stop"
}
],
"usage": {
"prompt_tokens": 28,
"completion_tokens": 32,
"total_tokens": 60
},
"verbarex": {
"cost_microusd": 3,
"latency_ms": 780
}
}
Real-Time Token Streaming
Set "stream": true in your request payload to receive server-sent events (SSE) token-by-token with ultra-low time-to-first-token (TTFT). The stream terminates with data: [DONE].
from openai import OpenAI
client = OpenAI(
base_url="https://verbarex.com/api/v1",
api_key="vx_live_YOUR_API_KEY"
)
stream = client.chat.completions.create(
model="deepseek-v4.1-flash",
messages=[{"role": "user", "content": "Write a bash script to monitor memory usage."}],
stream=True
)
for chunk in stream:
token = chunk.choices[0].delta.content or ""
print(token, end="", flush=True)
/sandboxes · POST/sandboxesCloud Sandboxes (Hardware-Isolated Linux MicroVMs)
VERBAREX Cloud Sandboxes run dedicated Linux MicroVMs with persistent NVMe storage at /workspace, full Python 3.14.4 and Bash 5.2 runtimes, and active outbound NAT egress. Compute is billed per second at $0.17 per active hour with zero cold-starts.
List Active Sandboxes
curl https://verbarex.com/api/v1/sandboxes \
-H "Authorization: Bearer $VERBAREX_API_KEY"
Launch a New MicroVM Sandbox
curl -X POST https://verbarex.com/api/v1/sandboxes \
-H "Authorization: Bearer $VERBAREX_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"name": "agent-runtime-1"
}'
/sandboxes/{sandbox_id}/commandsExecute Code & Commands
Executes shell commands or Python scripts inside your running microVM sandbox. You can pass the 26-character sandbox ULID, the sandbox name, or default to target your active sandbox. Returns stdout, stderr, process exit code, and execution latency.
curl -X POST https://verbarex.com/api/v1/sandboxes/test-sandbox-agent/commands \
-H "Authorization: Bearer $VERBAREX_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"command": "python3 -c \"import platform; print(platform.uname())\""
}'
Response Structure
{
"data": {
"id": "01M5YVGP...",
"status": "succeeded",
"exit_code": 0,
"output": {
"stdout": "uname_result(system='Linux', node='verbarex-microvm', release='6.12.103-x86_64', ...)\n",
"stderr": ""
},
"output_truncated": false,
"duration_ms": 380
}
}
Integrate into Python Autonomous Agents
Connect your autonomous agents directly to VERBAREX Cloud Sandboxes for safe code interpretation, web browsing, data analysis, and script execution.
import requests
class VerbarexSandbox:
def __init__(self, api_key: str, sandbox_id: str, base_url: str = "https://verbarex.com/api/v1"):
self.api_key = api_key
self.sandbox_id = sandbox_id
self.base_url = base_url
self.headers = {
"Authorization": f"Bearer {api_key}",
"Content-Type": "application/json"
}
def run_command(self, cmd: list, timeout: int = 60) -> dict:
url = f"{self.base_url}/workspaces/{self.sandbox_id}/commands"
resp = requests.post(url, headers=self.headers, json={"command": cmd, "timeout_seconds": timeout})
resp.raise_for_status()
return resp.json()["data"]
# Example usage:
sandbox = VerbarexSandbox(api_key="vx_live_...", sandbox_id="sbx-1793108128")
res = sandbox.run_command(["python3", "-c", "import sys; print(f'Running Python {sys.version.split()[0]} in microVM')"])
print(res["output"]["stdout"])
Instant BDT Wallet Top-Up
VERBAREX natively supports developers and enterprises in Bangladesh with instant automated top-ups via secure payment checkout. Top up using bKash, Nagad, Rocket, Upay, and local/international Visa & Mastercard.
Metered Pay-As-You-Go Billing
All resources on VERBAREX operate on transparent pay-as-you-go billing with per-token and per-second micro-settlement. There are no monthly subscriptions, minimum spends, or hidden platform fees.
Standardized JSON Error Envelopes
All API errors return consistent JSON error envelopes with traceable request IDs:
{
"error": {
"code": "insufficient_scope",
"message": "This API key is scoped for AI Gateway only. Please use a Cloud Sandbox API key for agent workflows.",
"details": {}
},
"request_id": "01M5YV..."
}
Hardware-Isolated MicroVM Architecture
Unlike shared container runtimes where containers share a single host Linux kernel, VERBAREX Cloud Sandboxes run inside dedicated hardware microVMs with individual Linux kernels. Memory, CPU, and process tables are hardware-isolated. Outbound networking is protected via NAT egress rules, and API keys are verified with constant-time token hashing.