NewCut your AI API costs by 30–60%

Your AI costs
are too high.

BTL Runtime sits between your app and AI providers like OpenAI and Anthropic. It automatically routes to the cheapest option, caches repeated requests, and shows you exactly how much you saved — on every single call. Change one URL. No other code changes.

live request
POST /v1/chat/completions
model: btl-2
tokens: 18,440 input
-> cache lookup...
OK cache hit (prefix, 18.1k tokens)
provider call: skipped
x-btl-saved: $0.0084
x-btl-customer-charge: $0.0056
x-btl-benchmark-cost: $0.014
routing ...
0%
Our fee on top of provider costs — drops to 2% at scale
$0
Free credit when you sign up — no card required
0 change
Change one URL in your code — nothing else needs to change
0 ways
Use our credits, or connect your own OpenAI / Anthropic keys

Routes across every major AI provider — with no code changes

OpenAIAnthropicAWS BedrockGoogle VertexTogether AIGroqDeepInfraFireworks AIOpenAIAnthropicAWS BedrockGoogle VertexTogether AIGroqDeepInfraFireworks AI
( 01 / 05 )

See what you'll save
before you switch.

Test it on a real request first. See the cost breakdown. Then move your traffic knowing exactly what you'll pay and what you'll keep.

01
Preview the cost before switching

Send a real request to our preview endpoint and get back the cheapest route, what the direct cost would be, what we'd charge, and how much you'd save — before moving anything live.

02
Point your app at BTL

Change the URL your app sends AI requests to. Same SDK, same model names, same response format. Your app doesn't know anything changed — it just gets cheaper.

03
See the savings on every response

Every response includes what the direct cost would have been, what you were charged, and how much was saved. The dashboard keeps a running total over time.

Request lifecycle
live
A
Your app
OpenAI SDK / unchanged
origin
G
BTL Runtime
Auth / billing / audit
gateway
R
Router
Eligibility -> scoring
routing
C
Cache layer
Exact / semantic / prefix
cache
P
Provider
Route / retry / fallback
provider
OK
Response
Plus savings headers
output
( 02 / 05 )

Start with the requests
you send over and over.

The biggest savings come from repeated requests — support bots, document Q&A, agents re-reading the same context. That's where the numbers get obvious fast.

Line itemBenchmark directVia BTL Runtime
First request$0.014$0.014
Repeat request (cache hit)$0.014$0.0056
Saved on repeat request$0.00$0.0084
100,000 similar repeats / month$1,400$560
Annualized effective cost$16,800$6,720
Benchmark direct cost$16.8kannual cost at 100K similar repeated requests per month
Repeat request benchmark$0.014
Saved per repeat$0
Annual$16,800
With BTL Runtime$6.7kannual effective cost with $10,080 preserved
Customer charge per repeat$0.0056
Saved per repeat$0.0084
Annual$6,720
_

What you can verify before you commit.

See the cost breakdown before and after you switch — not buried in a blended monthly invoice three weeks later.

Can verify now
See the cost before you switch

Send a real request to our preview endpoint and get back what the cheapest route would cost you, what we'd charge, and how much you'd save — before moving any live traffic.

Test on a real payload with zero commitment.
Can verify now
Every response shows what you saved

Each response includes the direct provider cost, what you were charged, and the amount saved — visible in the response headers and on your dashboard.

Per-request proof, not a blended monthly total.
Can verify now
Use our credits or your own API keys

Add paid Runtime credits, or connect your existing OpenAI, Anthropic, or Google keys and keep the same workspace, logs, and savings tracking.

No free model spend is granted automatically.
( 03 / 05 )

Estimate your savings.
Then test with a real request.

Use this to get a rough number, then send one of your real requests to our preview endpoint to get the exact cost before moving anything live.

Rough savings calculator

Model the repeated traffic first.

Use repeated or cache-friendly routes for the first estimate. Then run a real request through /v1/account/quote to see route and pricing details on your actual payload.

Repeated requests / month
70,000
Non-repeated requests / month
30,000
Projected annual view
$7,056
estimated yearly savings on the repeated or cache-friendly share alone
Direct annual benchmark$16,800
Projected Runtime annual cost$9,744
Projected monthly savings$588.00
This is a directional planning tool, not billing truth. Runtime already exposes exact per-request proof through quote preview, response headers, and the usage ledger.
( 04 / 05 )

Why teams choose BTL
over the alternatives.

Different tools solve different problems. Runtime is for teams that want to see the exact cost and savings on every request — not just a lower bill in theory at the end of the month.

01
See exactly what you saved — per request
Preview the cost before you switch, then read the actual savings back on every live response. No waiting for a monthly invoice.
02
One line of code to switch
Change the URL your app sends requests to. Everything else — model names, auth, response format — stays exactly the same.
03
Start free or use your own API keys
Use our $5 signup credit to try it out, or connect your existing OpenAI/Anthropic keys and keep the savings layer on top.
04
The fee gets smaller as you grow
Starts at 4% on top of provider costs, drops to 2% at scale. Caching reduces the amount we charge on — so as your hit rate rises, the effective cost drops.
BTL vs LiteLLM
Managed savings proof instead of another server to maintain.

LiteLLM is a good fit when you want to self-host the routing layer. Runtime is for teams that want cost previews, per-request savings proof, and a dashboard already built — without running another piece of infrastructure.

BTL vs OpenRouter
Your costs stay visible instead of disappearing into a black box.

OpenRouter is a good fit when access to many models is the main goal. Runtime is for teams that want to see exactly what each request costs, what they charged customers, and how much was saved — on every call.

BTL vs DIY
Get the cost-saving layer without building it from scratch.

Building it yourself makes sense if you want to own every provider integration and billing edge case. Runtime is for teams that want the savings to be part of the product on day one, not a six-month infrastructure project.

_

Who this is for.
And why it works.

Works best for
Apps that send the same or similar requests repeatedly.
Customer support bots, document Q&A, and AI agents that re-read the same context over and over.
Teams that want to see the cost breakdown before moving live traffic.
Products that need per-request savings data in the dashboard, not just a lower bill at month end.
Still useful even without heavy repetition
Routing and cost visibility work for any AI traffic.
Automatic failover to a backup provider if your primary one goes down.
Connect your own OpenAI or Anthropic keys and still get cost tracking and a usage dashboard.
Start with your highest-volume routes where the savings are obvious, then expand.
Built by
Bad Theory Labs

Runtime is a live product from the team behind RetainDB. The goal is to give AI teams a cost layer that is actually visible and verifiable, not just promised in a sales deck.

Already shipped
Dashboard, keys, cost previews, and request history

Everything shown here — workspace, API keys, cost previews, savings tracking, provider connections, and request logs — is already built and live. Not on the roadmap.

Fee structure
Our fee shrinks as your volume grows

We charge 4% on top of provider costs, dropping to 2% at scale. Caching reduces the amount we charge on. We only make more money if you actually spend more — so the incentive is to make requests as cheap as possible.

( 05 / 05 )

Start free.
Scale when it makes sense.

You see every provider cost — we add a small fee on top that shrinks as you grow. Or connect your own API keys and keep the savings layer without paying us per-token.

Hosted
Pay-as-you-go

$5 free on signup. No card required. You see every provider cost.

  • $0–$500: 4% fee on top of provider costs
  • $500–$5,000: 3% fee
  • $5,000+: 2% fee
  • Caching and cost savings included
  • No card required to start
Create workspace
Your own API keys
Bring your keys

Connect your existing OpenAI, Anthropic, or Google keys and keep the savings layer.

  • Works with OpenAI and Anthropic key formats
  • Your provider bill stays with you
  • Routing and caching still apply
  • Cost tracking and savings dashboard stay on
  • No code changes beyond the URL
See the docs
Enterprise
Custom

Dedicated instance, SLA-backed. SSO, audit logs, priority support.

  • Everything in Hosted
  • Negotiated fee tier
  • SLA and uptime guarantee
  • SSO / SAML
  • Dedicated onboarding
Talk to the team

Start saving on
your first request.

Create a workspace, test one real request to see the cost breakdown, and move your first traffic in minutes.