OpenAI-compatible inference gateway

One endpoint for every model. One ledger for every token.

InferClear is an OpenAI-compatible AI inference API gateway. Point your base URL at us and every frontier model arrives behind one key — with the per-request cost ledger, hard budgets and audit trail no provider dashboard will ever give you.

https://api.inferclear.com/v1

Zero retention by default — prompts and completions are never written to disk. Read the data policy

Get an API key See the two-line switch
Per-request
Cost ledger
Multi-route
Failover
Zero
Retention

Unified model access

Route once. Swap models forever.

Text, vision, image, audio and video — same request shape, same key, same invoice. Changing model is changing a string.

New frontier releases land in the catalog as providers ship them — request a model if yours is missing.

One API layer

Two lines of diff. Nothing else changes.

InferClear speaks the OpenAI wire format — streaming, tool calls, structured outputs, vision, prompt caching. Keep your SDK, your retries and your tests. Change the base URL and the key. 

from openai import OpenAI

client = OpenAI(
    base_url="https://api.inferclear.com/v1",   # ← this line
    api_key=os.environ["INFERCLEAR_API_KEY"],     # ← and this one
)

stream = client.chat.completions.create(
    model="anthropic/claude-opus-5",
    messages=[{"role": "user", "content": "Summarise the incident."}],
    stream=True,
)

Works with what your team already runs

Anything that accepts an OpenAI- or Anthropic-compatible base URL points at InferClear without a plugin.

  • Claude Code
  • Codex CLI
  • Cline
  • Roo Code
  • Cursor
  • Continue
  • Cherry Studio
  • LangChain
  • LlamaIndex
  • Vercel AI SDK
  • OpenClaw

One key per person, revocable in a click — instead of a shared provider key pasted into eleven .env files nobody can inventory.

Enterprise-ready

The routing is easy. The accounting is the product.

Any gateway can forward a request. The hard part starts the month your AI line item outgrows a personal card: who spent it, on what, against whose budget — and can you prove it to an auditor.

The ledger

Every request priced the moment it is served — model, route, input, output and cached tokens, latency, and the person behind the key. Queryable, exportable, reconcilable against the invoice.

Time Model Tokens Cost
09:41:02 anthropic/claude-opus-5 12.4k → 1.9k $0.2364
09:41:07 google/gemini-3.8-flash 88.1k → 0.4k $0.0291
09:41:11 openai/gpt-5.6-luna 6.0k → 3.2k $0.0118
09:41:14 anthropic/claude-haiku-4.5 2.2k → 0.6k $0.0041
Today · engineering $1,284.06 / $2,000.00

Budgets that actually stop

Hard caps per key, per member, per team, per month. The request is refused at the cap — not flagged in an email three days after the overrun.

Failover across routes

Several upstreams per model, plus our own inference capacity. A route that degrades drains out of rotation before your pager does anything about it.

An audit trail, not a dashboard

Key issued, key rotated, quota raised, model enabled, request served — each event timestamped and attributed to a person. Export it and hand it over.

One org, one invoice

Every team and every agent under a single account with unified billing and permissions. No more nine provider accounts and a reimbursement queue.

Zero retention by default

We meter what a request cost, never what it said. Prompts and completions are held in memory for the life of the call and never written to disk.

FAQ

Questions engineers ask first

Short answers here, the long ones in the docs.

What exactly is InferClear?

InferClear is an OpenAI-compatible API gateway for AI inference. One base URL and one key reach text, vision, image, audio and video models across many providers, and every request is metered into a single cost ledger with per-member budgets and an exportable audit trail.

How do I migrate an existing app?

Change two values: point your client's base URL at https://api.inferclear.com/v1 and swap the API key. Streaming, tool calls, structured outputs, vision and prompt caching keep the OpenAI request and response shapes, so your SDK, retry logic and tests stay as they are.

Do you store my prompts or completions?

No. InferClear runs zero data retention by default: request and response bodies live in memory for the duration of the call and are never written to disk. We keep operational metadata only — model identifier, token counts, latency, cost, timestamp, account and key — because billing, rate limiting and abuse prevention cannot work without it. When a request is routed to a third-party provider, that provider's own terms govern their handling; we prefer routes configured for no-training and zero retention wherever a provider offers them.

What happens when a provider goes down?

Where a model is available on more than one route, InferClear retries the request on a healthy route. Degraded upstreams drain out of rotation on error rate and latency rather than waiting for a status page. Failover is bounded by what providers actually serve — a model with a single upstream cannot be made redundant by a gateway, and we will not pretend otherwise.

How is usage billed?

Usage is metered per request against the selected model's published rate — by input, output and cached tokens, or by image, second or character where a model is priced that way — and drawn from your account balance in US dollars. Every line is visible in the ledger the moment it is served, so the invoice at the end of the month is a total you have already watched accumulate.

Can I raise my rate limits?

Yes. Requests per minute and tokens per minute are set per organization and can be raised for production workloads, subject to the models involved and upstream capacity. Tell us the model, the shape of the traffic and the peak you need to survive, and we will tell you what we can hold.

Who is InferClear for?

Teams whose AI spend has outgrown a shared personal key. A single developer can use InferClear to reach every model from one key; the reason to move an organization onto it is the moment finance asks which team spent the money and nobody can answer from a provider dashboard.

Put your org's inference on one ledger

InferClear is onboarding teams now. Tell us where to send your key and what you are running — we reply to every request ourselves.

No newsletter, no drip sequence — one reply from a person.