What the gateway does between your app and the model

Mobiforge sits in front of OpenAI, Anthropic, Google and DeepSeek. It authenticates your key, checks its budget, routes the request, retries on failure and records the cost.

OpenAI-compatible API

Works with the official OpenAI SDKs for Python and Node.js, LangChain, LlamaIndex, the Vercel AI SDK and any client that lets you set a base URL. Claude and Gemini models are translated to the same request and response shape.

POST /v1/chat/completions: Chat and tool calling in OpenAI format, with stream: true for SSE
GET /v1/models: Every available model ID with its current input and output rate
POST /v1/embeddings: Embeddings from supported providers through the same key
GET /v1/usage: Usage and cost per key, model and day, as JSON or CSV

classify.ts

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.mobiforgesystems.store/v1",
  apiKey: process.env.MOBIFORGE_API_KEY,
});

const res = await client.chat.completions.create({
  model: "google/gemini-3.8-flash",
  messages: [{ role: "user", content: "Classify: refund request" }],
});

console.log(res.usage); // prompt, completion, cached tokens

Routing and fallback

Each model can be served by more than one upstream route. On a 429, 5xx or timeout the gateway retries on the next route, up to two retries, before returning. Failed attempts are not billed.

Pin a request to one provider to turn retries off
Set a fallback model list, e.g. Sonnet 5.5 then GPT-6.1 Sol
Every attempt appears in the request log with its status code

Request trace with a timeout and retry

openai/gpt-6.1-sol · route A · upstream timeout after 30s 09:41:02
Retry 1 · route B · 200 OK in 1.8s 09:41:04
Billed 1,204 input + 388 output tokens 09:41:04

Illustrative product interface

Key settings with caps and model allowlists

Key policies

Key policies
Key	Allowed models	Monthly cap
prod-chat Owner	anthropic/, openai/	$1,000
intern-sandbox Developer	-flash, -luna	$25
nightly-evals Developer	deepseek/*	$150

Illustrative product interface

Budgets you set per key

Caps reset monthly. You get an email alert at 80%, and once a key hits its cap it returns HTTP 402 until you raise the limit. Allowlists stop a sandbox key from calling a $50-per-million-token model by accident.

See the saving, not just the spend

Usage reports show what you paid next to what the same tokens would cost at provider list price, broken down by provider, model and key. Export to CSV for your finance team.

Monthly spend versus provider list price

Spend by provider, USD Mobiforge (Pro) Provider list price

Illustrative product interface

Teams and Enterprise

Roles: Owner, Developer and Billing roles per workspace. Billing users see invoices but not keys.
Audit log: Key creation, revocation, cap changes and member changes, with who and when.
SAML 2.0 SSO: Sign in through Okta or Microsoft Entra ID on the Enterprise plan.
Provider rules: Restrict routing to providers that meet your requirements on data retention and training use.
Bring your own key: Use your own provider keys alongside credits; that usage is billed by the provider.
Invoicing: Monthly invoices with usage detail; Enterprise can pay by invoice instead of prepaid credits.

Compare plans Talk to us