Introduction

Getting started

Create a service token, choose its scopes, and make your first request.

Base URL

Send every API request to the base URL assigned to your organisation. Ask your administrator for the exact host. Examples in this documentation use:

text

https://api.inferada.com

Choose the right token

Inferada issues two kinds of bearer token. Their prefixes identify where you can use them:

Service API tokens start with inf_. Use these from your applications. They authorise the service APIs documented here, including completions, language, PII, Decisions, and audio.
Portal session tokens start with infp_. You get one by signing in to the portal. They authorise portal management endpoints, not the service APIs documented here.

Create a service token

Sign in to the portal.
Open the user menu in the bottom-left and choose API Tokens.
Select Create token. Enter a name, choose the required scopes, and optionally set an expiry or IP allowlist.
Create the token and copy its value immediately. The portal shows it only once.

Authentication

Pass the service token in the Authorization header as a bearer token on every request.

http

Authorization: Bearer inf_MjQ...

Every response includes an x-request-id header. Error bodies also include a request_id field. Include this value in a support request so Servada can identify the call.

Token scopes

Scopes determine which endpoints a service token can call. Assign only the scopes the application needs.

completionsscope: Chat completions. Endpoints: /v1/chat/completions, /v1/responses. (Listing models via /v1/models only needs a valid token, not a specific scope.)
embeddingsscope: Text embeddings. Endpoint: /v1/embeddings.
decisionsscope: Typed decisions. Endpoint: /v1/decisions.
languagescope: Grammar and spell checking. Endpoints: /v1/language/correct, /v2/check, /v2/languages.
piiscope: PII analysis and redaction. Endpoints: /v1/pii/analyse, /v1/pii/redact, /v1/pii/restore.
sttscope: Speech-to-text transcription and English translation. Endpoints: /v1/audio/transcriptions, /v1/audio/translations.
ttsscope: Text-to-speech synthesis. Endpoints: /v1/audio/speech, /v1/audio/voices.

A request to an endpoint outside the token's scopes returns a 403.

To restrict a token by source address, add IP addresses or CIDR ranges when you create it. Requests from other addresses return 403. The gateway rejects unbounded ranges (0.0.0.0/0 and ::/0); leave the list empty to allow any address.

Your first request

First, list the models available to the token. This endpoint accepts any valid token and filters the result to the models that token may use.

bash

curl https://api.inferada.com/v1/models \
  -H "Authorization: Bearer inf_YOUR_TOKEN"

Once you have a model ID, send a chat completion:

bash

curl https://api.inferada.com/v1/chat/completions \
  -H "Authorization: Bearer inf_YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.5-35b",
    "messages": [
      { "role": "user", "content": "Say hello in one word." }
    ]
  }'

Next steps

Chat completions — create OpenAI-compatible chat requests with optional Inferada extensions.
Processors — add redaction, output correction, policy checks, and routing controls to a completion.
Prompt templating — render the date and other gateway values in system prompts.
Rate limits — understand limit enforcement and handle 429 responses.
Usage accounting — understand which metrics each scope records.