SELF-HOSTED · MULTILINGUAL TRANSLATION API

Four translation models. One inference gateway.

Build multilingual translation into your own applications with an authenticated, asynchronous API and four independently configured model backends.

A single-tenant developer API, not a public translation website. Job submission and retrieval require Bearer authentication; the model catalog is public. GPU inference is operator-controlled.

ARCHITECTURE
Your applicationHTTPS
CPU API gatewayFastAPI · Auth
On-demand GPU inferenceModal
Hy-MT2-30B-A3BL40S · Q8_0
Index-Translate-35BL40S · Q8_0
Gemma 4 31B ITL40S · Q8_0
Qwen3.5-35B-A3BL40S · Q8_0
API OVERVIEW
GET /health   ·   GET /v1/models · public

Health and the model catalog are public; job endpoints require a Bearer token. The catalog does not check live GPU readiness or activate GPUs.

Four models. One consistent API.

Select a single translation model or request multiple model runs through an asynchronous job. Each worker has its own pinned GGUF model.

MODEL 01 · TENCENT

tencent/Hy-MT2-30B-A3B

hy_mt2_30b

Hy-MT2-30B-A3B · GGUF Q8_0 · dedicated L40S worker

MODEL 02 · INDEXTEAM

IndexTeam/Index-Translate-35B-A3B-preview

index_35b

Index-Translate-35B-A3B-preview · GGUF Q8_0 · dedicated L40S worker

MODEL 03 · GOOGLE

google/gemma-4-31B-it

gemma4_31b_it

Gemma 4 31B IT · GGUF Q8_0 · dedicated L40S worker

MODEL 04 · QWEN

Qwen/Qwen3.5-35B-A3B

qwen35_35b

Qwen3.5-35B-A3B · GGUF Q8_0 · dedicated L40S worker

Start building

Deploy the gateway, connect your client and run asynchronous translation jobs.

01

Configure your gateway

Configure Modal, PostgreSQL, private Secrets and verified GGUF Volumes.

Configuration guide ↗
02

Connect securely

Submit and retrieve jobs using your own Bearer token. Model catalog is public.

API security guide ↗
03

Handle async jobs

Submit a translation, poll its job status and handle GPU cold starts.

Operations guide ↗