Quick Start

Connecting to Induction is a one-line change: swap your agent’s base URL. You keep your own provider API key. It is forwarded with each request.

What you need

Your Induction API key and agent key
Your provider API key (OpenAI or Fireworks)

Get your keys

Sign in at app.induction.ai . Induction creates a default API key and a default agent for you on first sign-in, so there is nothing to set up before your first request.

Open Connect and both are already selected. Choose your provider, add your provider API key to the code sample, and run it. The sample arrives with your Induction API key and agent key filled in, so it works as pasted.

Two pages manage the keys once you’re past that first request:

API Keys lists your Induction API keys. The Key column holds the value your base URL needs.
Agents lists your agents. Each agent key identifies one of your AI agents, and the Dashboard reports cost and savings per agent, so give each agent its own key to keep them apart.

Your base URL

Every agent has its own base URL:


https://proxy.induction.ai/api/{induction_api_key}/{agent_key}

Induction supports the OpenAI Responses API only. Chat Completions is not supported. It’s served with one segment per provider: /openai for OpenAI models, /fireworks for open models hosted on Fireworks. Both use the same SDK and the same request shape. Only the segment, key, and model name change.

OpenAI

Python


from openai import OpenAI
 
client = OpenAI(
    api_key="sk-provider-api-key",
    base_url="https://proxy.induction.ai/api/{induction_api_key}/{agent_key}/openai",
)
 
response = client.responses.create(
    model="gpt-5.6-sol",
    input="Hello! How are you?",
)
 
print(response.output_text)

TypeScript


import OpenAI from "openai";
 
const client = new OpenAI({
  apiKey: "sk-provider-api-key",
  baseURL:
    "https://proxy.induction.ai/api/{induction_api_key}/{agent_key}/openai",
});
 
const response = await client.responses.create({
  model: "gpt-5.6-sol",
  input: "Hello! How are you?",
});
 
console.log(response.output_text);

curl


curl https://proxy.induction.ai/api/{induction_api_key}/{agent_key}/openai/responses \
  -H "Authorization: Bearer sk-provider-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.6-sol",
    "input": "Hello! How are you?"
  }'

LiteLLM


from litellm import responses
 
response = responses(
    model="openai/gpt-5.6-sol",
    input="Hello! How are you?",
    api_key="sk-provider-api-key",
    api_base="https://proxy.induction.ai/api/{induction_api_key}/{agent_key}/openai",
)
 
print(response.output_text)

AI SDK


import { createOpenAI } from "@ai-sdk/openai";
import { generateText } from "ai";
 
const openai = createOpenAI({
  apiKey: "sk-provider-api-key",
  baseURL:
    "https://proxy.induction.ai/api/{induction_api_key}/{agent_key}/openai",
});
 
const { text } = await generateText({
  model: openai.responses("gpt-5.6-sol"),
  prompt: "Hello! How are you?",
});
 
console.log(text);

Fireworks

Same Responses API, same OpenAI SDK. Point it at the /fireworks segment with your Fireworks key.

Python


from openai import OpenAI
 
client = OpenAI(
    api_key="fw-provider-api-key",
    base_url="https://proxy.induction.ai/api/{induction_api_key}/{agent_key}/fireworks",
)
 
response = client.responses.create(
    model="accounts/fireworks/models/kimi-k3",
    input="Hello! How are you?",
)
 
print(response.output_text)

TypeScript


import OpenAI from "openai";
 
const client = new OpenAI({
  apiKey: "fw-provider-api-key",
  baseURL:
    "https://proxy.induction.ai/api/{induction_api_key}/{agent_key}/fireworks",
});
 
const response = await client.responses.create({
  model: "accounts/fireworks/models/kimi-k3",
  input: "Hello! How are you?",
});
 
console.log(response.output_text);

curl


curl https://proxy.induction.ai/api/{induction_api_key}/{agent_key}/fireworks/responses \
  -H "Authorization: Bearer fw-provider-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "accounts/fireworks/models/kimi-k3",
    "input": "Hello! How are you?"
  }'

LiteLLM


from litellm import responses
 
response = responses(
    model="openai/accounts/fireworks/models/kimi-k3",
    input="Hello! How are you?",
    api_key="fw-provider-api-key",
    api_base="https://proxy.induction.ai/api/{induction_api_key}/{agent_key}/fireworks",
)
 
print(response.output_text)

AI SDK


import { createOpenAI } from "@ai-sdk/openai";
import { generateText } from "ai";
 
const openai = createOpenAI({
  apiKey: "fw-provider-api-key",
  baseURL:
    "https://proxy.induction.ai/api/{induction_api_key}/{agent_key}/fireworks",
});
 
const { text } = await generateText({
  model: openai.responses("accounts/fireworks/models/kimi-k3"),
  prompt: "Hello! How are you?",
});
 
console.log(text);

Check it’s working

Open the Dashboard in app.induction.ai and choose your agent. Every request your agent makes lands in the request history, newest first, with the model it used, the compression level it ran under, what it cost, and what it saved. Turn on live updates to watch requests arrive while your agent runs.

If nothing appears, check the base URL. The two key segments must match the Induction API key and agent key that the Connect page shows for that agent, and requests sent to the wrong agent land in that agent’s history instead.

Next, find the right compression level for this agent. Higher levels cut cost more, but can affect accuracy in some cases, so try High first and assess the results. Compression Levels covers the four settings and how to test between them.