Solutions

Anyone who buys tokens
is already a customer.

Six shapes of team, one integration. Whether you are shipping an AI product, running agents overnight, generating media at volume or answering to a European regulator, the gateway underneath is the same.

Get an API key Talk to sales

AI apps and workspaces

Every model in one account, on release day.

Products that let people choose a model live or die by how fast a new one appears in the picker. Here it is a catalog entry, not a sprint, and the best price for each job is found per call rather than per procurement cycle.

apps.conf

# The picker is just a list of ids.
MODELS = [
    "anthropic/claude-opus-5",
    "openai/gpt-5.4-mini",
    "google/gemini-3-flash-preview",
    "deepseek/deepseek-v4-pro",
]

reply = client.chat.completions.create(
    model=user_choice,      # any id in the catalog
    messages=messages,
    stream=True,
)

Agent and coding tools

An outage stops being a pager at 3am.

Long-running agents are the workload least able to tolerate a provider incident and most able to burn a budget by accident. Failover handles the first problem; a per-key spend ceiling handles the second.

agents.conf

# Key policy, enforced by the key rather than the agent.
routing:   fastest
budget:    $250 / month   # hard stop, reserved pre-dispatch
models:    allowlist
expires:   90 days
alerts:    50% / 80% / 100%

Creative platforms

Image, video, audio and music, signed on the way out.

Generate across modalities on one key, and attach provenance in the same call that produced the asset. No second pipeline, no second vendor, no reconciliation between the thing you made and the record that it was made.

creative.conf

job = client.videos.create(
    model="<video-model-id>",
    prompt="a lighthouse at dawn, slow push in",
    sign=True,     # C2PA manifest + watermark
)

# Video runs as a job you poll.
result = client.videos.retrieve(job.id)

EU-regulated teams

Residency you can point a procurement team at.

EU-only routing and no-training are properties of the key, enforced per request, with a retention class on every artifact and an hourly purge proof behind it. One contract instead of ten, and no tier upgrade to get in-region routing.

regulated.conf

# Key policy: eu-only
routes:     EU endpoints only
training:   never
retention:  zero (no prompt or output content stored)
proof:      hourly purge attestation
fallback:   fail closed, never out of region

Product companies

AI features with a budget line per project.

CRM, e-commerce, HR, legal, education: teams adding AI features to software that already has customers. Mint a key per project, give it a ceiling, and read the spend back per project rather than per company.

product.conf

# One key per project, one invoice for all of them.
keys:
  - name: crm-summaries      budget: $400/mo
  - name: support-triage     budget: $150/mo
  - name: contract-review    budget: $900/mo   eu_only: true

Resellers and platforms

Prepaid AI credits you can sell on.

Telcos, ISPs and integrators bundling AI into an existing customer relationship. Buy credit batches at a discount, issue sub-keys with their own budgets, and report usage per customer.

resellers.conf

# Sub-keys under one reseller account.
POST /v1/keys
{
  "label": "customer-4821",
  "budget_usd": 50,
  "models": ["openai/gpt-5-mini"],
  "expires_in_days": 30
}

Not sure which one you are?

Most teams are two of them at once. Tell us your models, volumes and rules and we will scope it with you in half an hour.

Anyone who buys tokens is already a customer.