OpenAI-compatible AI API gateway with team budget controls.
InferenceDirect is a unified AI inference API that routes requests
to Claude, DeepSeek, Llama, Qwen, and dozens more — all through one OpenAI-compatible endpoint.
Pay per token, no subscription. Every call is metered and accounted for at
api.inferencedirect.com.
A unified LLM API — swap the base URL, keep your SDK
InferenceDirect is an AI inference API gateway: a single OpenAI-compatible
endpoint that gives your team access to multiple LLM providers. No SDK migration,
no per-developer provider keys, no subscription. Point any OpenAI client at
api.inferencedirect.com and every call is metered against
your team's prepaid budget.
The /n/{team}
path selects the organisation node per request. Keys are bound to a node;
a request can narrow the node but never escape it.
Streaming included — token usage is metered from provider costs
either way, and every call lands on the ledger with model, tokens
in/out, latency, and cost.
LLM cost management for teams
Budget caps that mirror your org chart
Departments, teams, sub-teams — set AI spend limits at arbitrary depth.
Each level caps usage, warns or hard-blocks at the cap, rolls unused budget
forward, and restricts which models its members may call.
No more shared API keys and manual cost reconciliation.
Acme Corp $2,000/mo
├─ Engineering $1,200/mohard partition
│ ├─ ML-Platform $500/morollover: capped
│ │ └─ [email protected]$100/moper-team cap
│ └─ Internal-Tools $150/modeepseek-ai/* only
├─ Legal $200/mowarn at cap
└─ Individuals personal auto-nodes
Choose how each division
distributes budget downward: overcommit (children
capped, parent also binds), hard partition
(children must sum within the parent), or
shared pool (children draw freely from one pot).
Caps stop real money: a blocked budget returns a clean 402 the
moment it's spent, and strict mode reserves before the call ever
reaches a provider.
Built for AI infrastructure teams
Everything teams need to run AI APIs at scale
Pay per token — no subscription
Prepaid credit, no monthly fee, no surprise invoices. Add funds once;
usage draws the balance down. The remaining credit is always visible in
the admin console.
Hard-block spend caps
Per-member, per-team, and per-department monthly budgets with
warn-only or hard-block enforcement — plus one-off bonuses when a
team genuinely needs more headroom.
Strict reserve-then-settle
Opt-in per account: spend is reserved before each call and
settled to actuals after, so concurrent traffic can never
overshoot a contractual cap.
Per-model access policy
Allow-lists per division — route specific teams to specific models.
Budget says how much; policy says which models. Point cost-sensitive
workloads at cheaper open models automatically.
Service keys for CI & back-ends
Machine principals for automated workloads, each bound to the
billing node it charges, stored hashed at rest, and revocable
in one click without rotating shared secrets.
MFA & delegated administration
TOTP is mandatory for the portal. Capability grants separate
finance from operations, scoped to a division, with delegation
rights only the account owner can assign.
Security, safety & integrations
Guardrails, BYOK, and the integrations your stack already speaks
Content guardrails
Block or redact prompts with built-in injection, jailbreak, and PII
packs — plus your own regex rules. Applied inline, streaming included.
Bring your own provider key
Route through your own provider API key with an optional per-key
spend cap and automatic fall-back to the platform key.
OpenAI & Anthropic compatible
One endpoint speaks the OpenAI API and Anthropic's
/v1/messages. Sign in from the CLI with
device-code login — Claude Code and agents work natively.
Single sign-on for your apps
Register OAuth2 apps (Authorization Code + PKCE) so your own tools
authenticate users against your tenant — no shared secrets.
Per-call observability
Every call lands on the ledger with model, tokens, latency, and cost.
Fetch the record for any completion via
/v1/generation/{id}.
Abuse-shielded at the edge
Forged tokens, expired keys, and out-of-funds accounts are rejected
at Cloudflare's edge — before they reach, or bill, your workloads.
Optional add-ons — pay only if you need them
Two metered add-ons, on the same prepaid balance
Enterprise Security — metered
Lock the API down with cryptographic request signing
(Ed25519, per tenant / user / node) and per-request access
policy: required claims, time-of-day windows, JSON schema
validation, and a mandatory-signing rule. Prepaid at
$5 per 1,000,000 calls — billed only while it's on.
Analytics & chargeback — metered
Usage analytics and chargeback reporting: per-user
and per-team cost, org-tree drill-down, CSV export. Usage-based pricing
from your prepaid balance; scales from D1 to R2 for high volume.
Add-ons are independent — fund either, both, or neither. There's still no subscription.
Getting started
Three steps from sign-up to your first AI API call
Create an account
Sign up with Google, set up two-factor authentication, and create
your first tenant. Add credit to activate your prepaid balance.
Get an API key
Your key is bound to your billing node. Everything it spends is
attributed to that node automatically — no tagging, no manual tracking.
Point your client and call
Change the base_url in any OpenAI-compatible
client and start making calls. Usage, cost, and budget standing appear
on the dashboard in real time.
Ready to centralise your team's AI API spend?
Pay per token. No subscription. Team budgets built in from day one.