ϟfluxgateAPI / 0.9

BUILDING THE INFERENCE LAYER

AI inference.
Your traffic.
One API.

Run language models through a single inference API. Stream tokens to your app, generate embeddings, or send a million prompts in a batch.

OpenAI-compatible request format · JSON over HTTPS

endpoint / chat.completionsSTREAMING
01POST /v1/chat/completions02Authorization: Bearer $FLUXGATE_KEY03Content-Type: application/json04 05{06  "model": "stride-8b-instruct",07  "stream": true,08  "messages": [{ "role": "user",09    "content": "Explain token streaming." }]10}
Example output / SSE

Responses arrive one token at a time.▌

Responses arrive one token at a time.
Illustrative API response
01 / INTERACTIVEToken streamingServer-sent events, as they arrive
02 / ASYNCHRONOUSJSONL batchesUp to 50,000 requests per file
03 / SEARCHText embeddings768 or 1,536 dimensions
Find your model ↗

FROM REQUEST TO RESPONSE

Less plumbing.
More building.

Keep your application code. Route requests by model ID, set token limits, and separate production keys from experiments.

Explore the batch endpoint →
01Your applicationHTTP / Python / JS
↓
02Fluxgate routerkey scope · model ID · limits
↓
Stride 8BSummit 32BEmbed 768

Make inference do useful work.

THREE COMMON STARTING POINTS
01

Support agents

Stream answers with structured tool calls. Keep conversation context inside a 32K token window.

POST /v1/chat/completions
02

Document enrichment

Classify and summarize a large corpus with JSONL inputs and a downloadable result file.

POST /v1/batches
03

Semantic retrieval

Turn passages into vectors for a search index. Choose output dimensions for your storage budget.

POST /v1/embeddings

EARLY ACCESS, THOUGHTFULLY SIZED

Have a workload in mind?

Talk capacity ↗