Our AI bill grows faster than revenue
Send less, pay less
Foreman V2 picks the context each task actually needs instead of resending everything. Route routine work to cheaper models you approve.
For SaaS & enterprise teams
AI is your fastest-growing cost. Put a gateway in front of it.
AetherGate sits between your product and your AI providers. Cut wasted context, cap spend for every app or customer, keep provider keys sealed, and fail over when a provider goes down. On your own provider accounts, with 0% token markup.
- Works with the OpenAI SDK you already use
- Bring your own provider keys
- 0% markup on provider usage
- Monthly billing, cancel anytime
Your plan updates as you answer the three questions on the rightbelow.
Fit check
3 questions · nothing is sentYour fit
Enterprise Growth
$999 /month
- $60,000/month sits in the Enterprise Growth range (~$25K–$100K/month).
- Includes 250 gateway keys, so each of your 11-50 apps or customers can have its own limit.
At a 30% provider-cost reduction
$17,001/mo net
Recommended key protection
L5 · Bank Vault
Your plan, worked out
What Enterprise Growth does for a $60,000/month AI bill.
The money
5% · conservative60%60% · Foreman V2 internal token average
- Current provider spend
- $60,000/mo
- Modeled provider savings
- −$18,000/mo
- AetherGate Enterprise Growth
- +$999.00/mo
- Net modeled savings
- $17,001/mo
$204,012 a year, before counting the security and outage controls.
Break-even: a 1.7% drop in provider cost pays for the subscription.
Illustrative. Foreman V2 averaged 60% fewer tokens in AetherGate's internal benchmark tests; fewer tokens do not always mean an equal drop in your bill. Your provider bills you directly and AetherGate adds no markup. See the benchmarks.
The keys
A key that can reach $60,000 a month is a bank vault, not a drawer. Here's how we'd seal your production keys.
sk-prod-••••4e5c
Level 5 · Bank Vault
- Sealed with its own AES-256 key, bound to your workspace
- Every change written to a tamper-evident audit log
- A total spend limit is required
- A daily spend limit is required
- Expires within 30 days
- Your password is required to change, replace, or remove it
- Pauses itself if an hour's spend passes half the daily limit
- Raising a limit, extending expiration, or lowering the level waits 24 hours
- Only app keys you approve can use it
- Tightening is always instant
Provider keys are shared across apps and people
One sealed key per provider
Your provider keys stay in the vault. Every app, service or customer gets its own revocable AetherGate key with its own limit.
Switching cost
Live in an afternoon.
- 01
Connect your provider accounts
Paste your OpenAI, Anthropic, Google or OpenRouter keys once. Each is sealed with its own AES-256 key and never shown again.
- 02
Give every app its own key
Issue an AetherGate key per service, environment or customer, each with a spend limit and expiry. Revoke one without touching the rest.
- 03
Change one base URL
Keep your OpenAI SDK and your code. Requests now pass through your budgets, routing and backups.
Technical terms: OpenAI-compatible /v1/chat/completions endpoint, server-sent event (SSE) streaming, bring-your-own-key (BYOK) provider credentials, per-key cost reservations and pre-stream failover via fallback_models.
Read the setup guideimport OpenAI from "openai"; const client = new OpenAI({ baseURL: "https://api.openai.com/v1", baseURL: "https://aethergate.site/api/v1", apiKey: process.env.AETHERGATE_API_KEY, // one key per app});Streaming, tool calls and image inputs work through the same endpoint for compatible models.
Before you pick a gateway
Pick a bad day. See who stops it.
Four things that actually go wrong with AI keys, and what each gateway's own documentation says happens next.
A stolen login tries to raise a key's limit to $50,000
Someone gets into a signed-in browser and goes straight for the spend limit.
AetherGate
Vault asks for the password again. Bank Vault also holds the raise for 24 hours, in plain view on the key, and you can cancel it.
OpenRouter
Key limits are editable. No password re-check or delay on changes is described in the public docs.
Portkey
Budget limits can't be edited once set; a new limit means a new integration. No password re-check is described.
LiteLLM
Budgets are set and updated through the key API. No password re-check or delay is described.
Checked against public docs on October 16, 2026: OpenRouter: create API key, Portkey: budget limits, LiteLLM: budgets and rate limits. “Not in public docs” means we looked and didn't find it.
AetherGate fully covers 4 of 4. No other gateway fully covers more than 1.
See every control side by side · verified October 16, 2026
Security review
The questions your security team will ask. Answered straight.
Plain answers about what AetherGate enforces today, including what it doesn't do.
Technical terms: AES-256-GCM envelope encryption with workspace-bound data keys, HKDF-derived versioned master key, hashed gateway keys, RBAC (Owner/Admin/Auditor), SHA-256 hash-chained audit log, pattern-based prompt redaction.
Read the full security page- Each provider key is sealed with its own AES-256-GCM data key, bound to your workspace, so a copied database row won't decrypt elsewhere. Your apps never see provider keys; they get hashed, revocable AetherGate keys.
- From Vault level up, changing, replacing or removing a key asks for the account password again. At Bank Vault, raising a limit, extending expiry or lowering the level waits 24 hours in plain view, and you can cancel it. Tightening is always instant.
- Every request reserves its estimated cost before it's sent, and parallel calls only run while those reservations fit under the key's cap. Vault and above pause a key if an hour's spend passes half its daily limit.
- Shared workspaces enforce Owner, Admin and Auditor roles. Administrative actions record actor, role, before/after state and request ID in a per-workspace SHA-256 hash chain, exportable as JSON or CSV.
- Emails, card numbers and common API-key formats are redacted from prompt text before it reaches the provider by default. Responses, tool-call arguments and image pixels are not redacted, and your provider's own retention terms still apply.
- If a provider fails before a response starts, the request routes to a backup model you approved. A stream that has already started is never silently replayed on another model.
- These are scoped and agreed per contract under Custom. They aren't included automatically, and we don't claim attestations we don't hold. Ask us and we'll tell you exactly where things stand.
A flat subscription. Not a percentage of your growth.
Your providers bill you directly at their prices. The highlighted plan matches your fit check.
Team
$249.99 /month
Up to ~$25K/month in model spend
- Keys
- 50
- Req/min
- 5,000
- Rules
- 100
- Foreman V2 context optimization
- Per-key spend controls
- All five protection levels
Enterprise Growth
$999 /month
~$25K–$100K/month in model spend
- Keys
- 250
- Req/min
- 10,000
- Rules
- 500
- Everything in Team
- Workspace roles and audit records
- Provider, credential and app usage reporting
Enterprise Scale
$2,499 /month
~$100K–$300K/month in model spend
- Keys
- 1,000
- Req/min
- 20,000
- Rules
- 1,000
- Everything in Enterprise Growth
- Capacity for large app portfolios
- All five protection levels at scale
Custom
Talk to us
Above $300K/month, or VPC, SLA and compliance needs
- Keys
- Scoped
- Req/min
- Scoped
- Rules
- Scoped
- Volume and capacity planning
- VPC or specialized deployment discussed
- Security and procurement reviews
Team includes a 7-day free trial for new customers; a payment method is required and it renews at $249.99/month unless canceled. Request limits are per-key plan ceilings; provider quotas still apply. Provider API usage is billed separately by your providers. Full plan details.
Talk to a builder, not a sales queue
Get a rollout plan for your stack.
Tell us what you run. We reply by email with where AetherGate would cut cost in your traffic, how we'd seal your keys, and what switching takes.
- 1
Your fit check comes with it
No re-explaining. We see the spend, key count and problems you picked.
- 2
A written plan back
Which workloads to route where, which keys get which level, and the expected break-even.
- 3
Switch in an afternoon
Change one base URL in your OpenAI SDK. Keep your provider accounts.
Ready now? Start on Enterprise Growth and we'll still send the plan.