Keep the key server-side
Load AIWAVE_API_KEY from an environment variable or the agent's secret store. Do not place it in browser code, prompts, screenshots, or source control.
Connect the agent you already run.
Point OpenClaw, WorkBuddy, or your own agent at AIWave with one key and an exact model ID. Start with the OpenAI-compatible route, then follow the native contract when the model family needs it.
The setup has four decisions.
The agent owns memory, tools, permissions, retries, and task state. AIWave provides the model route, usage record, and charge surface.
Choose the verified launch route
Use https://aiwave.live/v1 with /v1/chat/completions for the current verified launch set.
Pin one verified model ID
Start with deepseek-v4-pro, glm-5.1, or kimi-k3. Copy the exact ID from Models.
Run one real task
Keep the response, usage, status, model ID, request ID when available, and charge evidence together before allowing the agent to handle more traffic.
Choose a project with a clear human boundary.
An agent is useful when the input, approval point, and evidence are explicit. These are starting projects, not claims of autonomous operation.
Ship code
Turn a bounded issue into a proposed patch and test summary.
Human approval required. Agent drafts and tests; a maintainer reviews and merges.
Research faster
Turn a defined question into sources, claims, and a decision brief.
Human approval required. Agent gathers and structures; an owner checks sources and decides.
Run support
Draft a reply from approved help material and the current request context.
Human approval required. Agent drafts; a human approves every customer-facing answer.
Automate operations
Prepare a repeatable check, report exceptions, and propose the next step.
Human approval required. Agent observes and proposes; a human approves any production change.
Build a content system
Create a source-backed outline, draft, and review checklist for one topic.
Human approval required. Agent drafts; a human verifies claims and publishes.
Create visual workflows
Plan a prompt, variants, approval points, and export requirements.
Human approval required. Agent prepares the workflow; a human approves assets and any paid render.
Use the adapter your agent already supports.
These examples use placeholders only. Replace the model ID with the exact value you tested. Never paste a live key into the example.
OpenClaw · custom provider
{
models: {
mode: "merge",
providers: {
aiwave: {
baseUrl: "https://aiwave.live/v1",
apiKey: "\${AIWAVE_API_KEY}",
api: "openai-completions",
models: [
{ id: "deepseek-v4-pro", name: "DeepSeek V4 Pro" }
]
}
}
}
}WorkBuddy · custom model
Provider: Custom
Protocol: OpenAI-compatible
Base URL: https://aiwave.live/v1
API key: load from WorkBuddy's secure credential field
Model: deepseek-v4-proYour own agent · OpenAI SDK
import os
from openai import OpenAI
client = OpenAI(
base_url="https://aiwave.live/v1",
api_key=os.environ["AIWAVE_API_KEY"],
)
response = client.chat.completions.create(
model="deepseek-v4-pro",
messages=[{"role": "user", "content": "Reply with one sentence."}],
)
print(response.choices[0].message.content)Use the adapter your agent already exposes.
OpenClaw, OpenAI SDK, Vercel AI SDK, LangChain, CrewAI, AutoGen, Pydantic AI, LlamaIndex, Flowise, n8n, WorkBuddy, and custom HTTP agents commonly expose an OpenAI-compatible configuration path. Check the current tool documentation before copying its field names.
Verified chat route
Use https://aiwave.live/v1 with the OpenAI chat adapter and one tested model ID.
Other provider families
Claude, Gemini, image, audio, and video routes remain route-specific until their public verification and billing facts are complete.
Keep approval outside the model
Use tool permissions, budgets, retries, and human approval in your agent. A model route does not supply those controls.
When the agent fails, preserve the boundary.
Most first failures come from the wrong base URL, model ID, protocol, or parameter shape. Save the error before changing several settings at once.
404 / model not foundCheck the route and exact ID ↗
401 / authenticationCheck the key source and header ↗
400 / invalid requestCheck adapter and parameters ↗
429 / 503 / timeoutRead the recovery sequence ↗
Check the agent's current setup screen.
Agent products change their field names and supported adapters. These links point to the current official configuration references we used for this page.