IntSen

Developers · Quickstart

Keep your framework.
Change the route. Gain control.

Govern an existing agent without an SDK: two short YAML files, a new base URL, an OAuth token and a few headers. Your agent's logic and tools stay unchanged.

Start the walkthroughFull onboarding guide

The workflow

Nine steps, most of them one line

  1. Step 01

    Get your application service and OAuth client

    Your platform team sets up your team's space in IntSen and an OAuth client (client id and secret) for your agent—or registers your team's identity provider.

  2. Step 02

    Write the Registration

    The use case: owners, maximum data class, the OAuth client, approved models and the capture policy.

  3. Step 03

    Write the Agent

    The agent: owners, description, which registration it belongs to, its client and per-turn limits.

  4. Step 04

    Validate

    Run intsenctl apply with --dry-run to see validation errors and exactly what would change.

  5. Step 05

    Apply

    Apply both files with intsenctl, or through your team's config repository pipeline.

  6. Step 06

    Authenticate the agent

    Get a token with the OAuth client-credentials flow and use it as the API key.

  7. Step 07

    Change the base URL

    Point your model client at the IntSen AI Gateway's OpenAI route.

  8. Step 08

    Add the headers

    Registration and agent once; a new task id for every user message; conversation and user ids if you have them.

  9. Step 09

    Make a call and inspect it

    Send a request, then open the agent in the Console to see the turn, its checks and its cost.

Interactive walkthrough

From YAML to a governed turn

Switch between the files. Every example uses reserved example domains; replace them with your own values.

registration.yaml
# registration.yaml — your use caseapiVersion: gateway.ai.intsen.com/v1kind: Registrationmetadata:  name: field-service-assistant-prod  namespace: field-service                 # your application servicespec:  description: Assistant for field technicians (manuals, parts, work orders)  owners:    - { email: product-owner@example.com }    - { email: tech-lead@example.com }  maxDataClass: confidential               # public | internal | confidential | restricted  clients:    - { clientId: fleet-advisor-prod }     # the OAuth client your agent signs in with  models:    - provider: openai      ids: [ gpt-5.4-mini ]  capture: { bodies: true, retentionDays: 90, redactPii: true }

Field notes

namespace
Your application service—your team's space in IntSen.
maxDataClass
The most sensitive data this use case may send. Models must be cleared for it.
clients
The OAuth client your agent signs in with. Unbound clients are refused.
models
Approved provider and model ids. The first release supports OpenAI.
capture
Store bodies (needed for traces, evals and compliance), retention, PII redaction.

TypeScript

The same change in TypeScript

Set the base URL and default headers on the client; pass turn headers per request.

More integration examples
agent.ts
import OpenAI from "openai";import { ulid } from "ulid"; const client = new OpenAI({  baseURL: "https://gateway.intsen.example/openai/v1",  apiKey: await getIntsenToken(),            // OAuth client-credentials token  defaultHeaders: {    "x-intsen-registration-id": "field-service-assistant-prod",    "x-intsen-agent-id": "field-service/fleet-service-advisor",  },}); // One task id per user messageconst turnHeaders = { "x-intsen-task-id": ulid(), "x-intsen-conversation-id": conversationId };const response = await client.chat.completions.create(  { model: "gpt-5.4-mini", messages, tools },  { headers: turnHeaders },);

Error handling

Every refusal carries a code

Gateway errors use an OpenAI-compatible envelope with a stable code and a request id you can look up in the Console.

403 response body
{  "error": {    "message": "Model 'gpt-5.4' is not allowed for registration 'field-service-assistant-prod'.",    "type": "intsen_gateway_error",    "code": "model_not_allowed",    "param": "model",    "source": "gateway-authorization",    "request_id": "01JB8Q4Z0X9W6Y3T2R1P0N5M4K"  }}
Handling refusals
from openai import APIStatusError try:    response = client.chat.completions.create(        model="gpt-5.4-mini", messages=messages, extra_headers=turn_headers)except APIStatusError as err:    code = (err.body or {}).get("error", {}).get("code")    if code == "task_budget_exceeded":        return "This request needs more steps than allowed. Please narrow it down."    if code in ("model_not_allowed", "provider_not_allowed"):        raise  # configuration problem: check the Registration's models    raise

Want a guided technical walkthrough?

We'll connect one of your agents with your team, live, and walk through the traces, checks and spend it produces.

Book a live demoDiscuss an enterprise pilot