Agent-ready controls
- Model allowlist
- Tool permission
- Maximum output tokens
- RPM, TPM and concurrency limits
- Daily and monthly budgets
A safe rollout
- Begin with one tool and deterministic arguments.
- Validate tool input on your server.
- Cap loop depth and retry count.
- Log request IDs and tool outcomes.
- Require human approval for irreversible actions.
Framework compatibility
Frameworks such as LangChain, CrewAI and other OpenAI-compatible clients can be evaluated through the same base URL. Framework support is not a substitute for testing the actual model and tool schema.
agent-tool.tstypescript
const response = await client.chat.completions.create({
model: "qwen3.7-plus",
messages: [{ role: "user", content: "What is the order status?" }],
tools: [{
type: "function",
function: {
name: "get_order_status",
description: "Return the current order status",
parameters: {
type: "object",
properties: { order_id: { type: "string" } },
required: ["order_id"]
}
}
}]
});Common questions
Which model should I use for an AI agent?
Choose from models that advertise tool support, then test your actual schemas and language requirements. There is no universal best agent model.
Can I stop an agent from overspending?
Use bounded loops in application code and enforce project and API-key budgets, rate limits and concurrency limits at the gateway.
Do I need to change my agent framework?
Many frameworks accept an OpenAI-compatible base URL. Confirm the framework's exact configuration and response expectations in a small test first.