DigiCore AI

One base URL for every model your team calls

Token Factory is an OpenAI-compatible gateway. Each request is checked against its project's models, rate limits and budget, then billed line by line. Open-weight models such as DeepSeek run in our own GPU data centers, so they cost less than the official APIs.

Create accountSign in to the console
POST /v1/chat/completionsreq_8b3d02e1
  • API key belongs to payments-assist
  • deepseek-v3.1 is in the project's catalog
  • Within 600 requests per minute
  • Project budget has room
  • Served by deepseek-v3.1 on our own GPUs, 1.84 s
Input2,184 tokens
$0.000401
Cache read, in input1,536 tokens
$0.000077
Output412 tokens
$0.000618
Billed amount
$0.001019
An example request. Prices are for illustration.

Keep your code. Change one line.

  1. 1

    Create a project

    Choose which models it may call. Anything outside that list is refused.

  2. 2

    Create an API key

    Each key belongs to one project, so every request is attributed and limited by it.

  3. 3

    Change your base URL

    Keep the OpenAI SDK or client you already use. Only the base URL and key change.

client.ts
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://proxy.token-factory.digicore-ai.com/v1",
  apiKey: process.env.TOKEN_FACTORY_API_KEY,
});

What happens to a request on its way through

  1. Admit

    • API keys scoped to one project, with rotation
    • Tokens from your own identity provider, checked against the project's OIDC contract

  2. Check

    • The model must be in the project's catalog
    • Request and token ceilings per minute and per hour, for each key or project
    • Budgets and your credit balance

  3. Route

    • Fallback models and traffic splits
    • Routing on request attributes your application sends
    • Open-weight models served from our own GPU data centers
    • Your own OpenAI-compatible endpoints next to the hosted providers

  4. Record

    • A log entry for every request
    • Usage by API key and by model
    • The billed amount, itemized

Priced below the official APIs

We run open-weight models such as DeepSeek on GPUs in our own data centers instead of reselling the model maker's API. Running the hardware ourselves is what lets us charge less per token.

Each request is priced from what it used: input tokens (cache reads and writes included), output tokens, and per-image charges for image models.

Buy credit up front and top up as you go.

Create a project and send your first request.

Create accountTalk to us