Provider: Inference.net

Use Respan Gateway to call Inference.net models while keeping unified observability (logs, cost, latency, reliability) in Respan.

Quick setup

Send your first request

Pick the integration that matches your stack. The base URL is https://api.respan.ai/api and the only key needed is your RESPAN_API_KEY.

1
from openai import OpenAI
2
3
client = OpenAI(
4
    api_key="YOUR_RESPAN_API_KEY",
5
    base_url="https://api.respan.ai/api",
6
)
7
8
response = client.chat.completions.create(
9
    model="inference_net/meta-llama/llama-3.1-70b-instruct/fp-16",
10
    messages=[{"role": "user", "content": "Hello, Inference.net!"}],
11
)
12
print(response.choices[0].message.content)

More integrations

Inference.net models work with every Respan gateway integration:

• OpenAI SDK
• OpenAI Agents SDK
• LangChain
• LangGraph
• LlamaIndex
• Pydantic AI
• Respan native (OTel)

Switch models

Change the model parameter to call any supported model through the same client. Use the inference_net/ prefix to disambiguate when routing across providers. Browse the full list on the Models page.

1
client.chat.completions.create(model="inference_net/meta-llama/llama-3.1-70b-instruct/fp-16", messages=messages)
2
client.chat.completions.create(model="openai/gpt-5.5", messages=messages)
3
client.chat.completions.create(model="anthropic/claude-sonnet-4-5", messages=messages)
4
client.chat.completions.create(model="gemini/gemini-3.5-flash", messages=messages)

Use your own Inference.net key (BYOK)

Credits are the default path. If you’d rather bill Inference.net directly, attach your own provider key.

Override credentials per model (Optional)

Use credential_override when one model on a request should use a different Inference.net key than the default.

1
{
2
  "customer_credentials": {
3
    "inference_net": { "api_key": "YOUR_INFERENCE_NET_API_KEY" }
4
  },
5
  "credential_override": {
6
    "inference_net/meta-llama/llama-3.1-70b-instruct/fp-16": { "api_key": "ANOTHER_INFERENCE_NET_API_KEY" }
7
  }
8
}

Log without proxying (Optional)

Already calling Inference.net directly? Send logs to Respan asynchronously to track cost, latency, and performance for those external calls.

1
import requests
2
3
requests.post(
4
    "https://api.respan.ai/api/request-logs/create/",
5
    headers={
6
        "Authorization": "Bearer YOUR_RESPAN_API_KEY",
7
        "Content-Type": "application/json",
8
    },
9
    json={
10
        "model": "inference_net/meta-llama/llama-3.1-70b-instruct/fp-16",
11
        "prompt_messages": [{"role": "user", "content": "Hello, how are you?"}],
12
        "completion_message": {"role": "assistant", "content": "Hello from Inference.net through Respan."},
13
        "cost": 0.001,
14
        "generation_time": 1.2,
15
        "customer_params": {"customer_identifier": "user_123"},
16
    },
17
)

See the logging guide for the full setup.