Deadline-aware AI inference

BACK/BY

Set the time. Get it back by then.

BackBy selects the lowest-cost eligible route for every request, while preserving the original realtime path to protect delivery.

Start with one API See the routing decision
01

Privacy first

Data class, region, and retention policy are enforced before a provider receives the request.

02

Performance protected

The requested model and original realtime path remain available as the delivery backstop.

03

Savings after

Lower-cost capacity is used only when it remains eligible for the selected delivery window.

One request, one commitment

Routing decisions happen behind the API. The deadline stays visible.

REQUEST

Accept the original call

OpenAI- and Anthropic-compatible interfaces preserve the application’s existing request shape.

POLICY

Protect the boundary

Privacy, region, retention, model, and credential constraints remove ineligible routes.

ROUTE

Use waiting intelligently

Current capacity and latency evidence determine whether flexible capacity can meet the window.

DELIVER

Return by the promise

BackBy reserves enough time to fall back to realtime before the selected delivery window closes.

Developer entry point

Change the endpoint. Keep the request.

Declare when the result is needed. BackBy handles route selection, delivery protection, and auditable evidence.

curl https://api.backby.ai/v1/chat/completions \
  -H "Authorization: Bearer $BACKBY_API_KEY" \
  -H "X-BackBy-Delivery-Within: PT15M" \
  -H "X-BackBy-Data-Class: internal" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.4",
    "messages": [{"role":"user","content":"..."}]
  }'