Credit — USD balance
https://api.korlogic.com/v1
Every request debits a USD balance on the key at the rates the operator set.
Korlogic — API gateway
OpenAI + Anthropic compatible
ONE URL. 14 MODELS.
An OpenAI- and Anthropic-compatible gateway. Point your existing SDK at one base URL and keep the rest of your code.
01 — Endpoints
Two base URLs. One key format.
Token — Prepaid tokens
https://api.korlogic.com/token/v1
Every request draws down a prepaid token budget — input plus output, counted as tokens.
Authorization: Bearer kl_live_… — keys are issued by the operator; the billing mode is stored on the key.
02 — Drop-in
Your SDK already speaks it.
Two request shapes
Chat Completions and Messages both go through. Use whichever SDK your codebase already has.
Streaming passes through
Streamed responses are proxied straight back to your client, chunk by chunk.
Errors keep their shape
Upstream status codes come through unchanged, and gateway rejections use the OpenAI error body.
python — openai
from openai import OpenAI
client = OpenAI(
base_url="https://api.korlogic.com/v1",
api_key="kl_live_…",
)
stream = client.chat.completions.create(
model="claude-opus-5",
messages=[{"role": "user", "content": "ping"}],
stream=True,
)
Two lines change. The client, the model call and the streaming loop do not.
03 — Catalog
The whole catalog, one key.
14 models · a key can be scoped to any subset of them · rates are set by the operator, so there is no public price list
- 01
claude-haiku-4-5-20251001
Claude Haiku 4.5
- 02
claude-opus-4-6
Claude Opus 4.6
reasoning - 03
claude-opus-4-7
Claude Opus 4.7
reasoning - 04
claude-opus-4-8
Claude Opus 4.8
reasoning - 05
claude-opus-5
Claude Opus 5
1M ctxvisionreasoning - 06
claude-sonnet-5
Claude Sonnet 5
reasoning - 07
fable-5
Fable 5
1.05M ctxreasoningcache - 08
gpt-5.3-codex
GPT-5.3 Codex
- 09
gpt-5.4
GPT-5.4
- 10
gpt-5.4-mini
GPT-5.4 mini
- 11
gpt-5.5
GPT-5.5
- 12
gpt-5.6-luna
GPT-5.6 Luna
1.05M ctxcache - 13
gpt-5.6-sol
GPT-5.6 Sol
1.05M ctxcache - 14
gpt-5.6-terra
GPT-5.6 Terra
1.05M ctxcache
04 — Pipeline
What happens on every request.
- 01
Your app
Any OpenAI or Anthropic client, pointed at the base URL for the key you hold.
- 02
korlogic.com
The key is checked, the model is checked against its allowlist, and funds are checked.
- 03
Upstream
The request is proxied with the operator’s own parent key. Yours never leaves the gateway.
- 04
Ledger
Tokens, cost, status and latency are written per request, then the key is debited.
01
Two protocols, one gateway
OpenAI Chat Completions and Anthropic Messages are both accepted and both metered.
02
Streaming, not buffering
Stream responses exactly as you would upstream; usage is parsed out of the stream.
03
Credit or token billing
A key carries either a USD balance or a prepaid token budget. The mode is fixed on the key.
04
Per-key model allowlist
A key can be scoped to a subset of the catalog. Anything outside it is refused.
05
Time-based expiry
Keys can be issued with an expiry date, after which requests stop.
06
Usage you can check
Paste a key into the usage checker for balance, consumption and recent requests.
Point your SDK at korlogic.com.
There is no signup. Keys are issued by the operator — paste one into the usage checker to see balance, consumption and recent requests.