A backup with rules.
Choose model routes and approved fallbacks across your connected accounts. Compatible failover depends on your configuration and when the upstream failure happens.
Explore routingYour models. Less wasted work.
Keep the intelligence.
Cut the wasted tokens.
Stop paying your AI to reread information it doesn't need. Foreman V2 selects relevant context and coordinates the work. AetherGate brings routing, spending limits and provider security into the same workspace.
Team from $249.99/month · 7-day trial · Provider usage billed separately
Bring your own API keys0% token markupOpenAI-compatible endpoint
Inside a better promptIllustrative workflow
The job
“What did the recovery test show?”
Send everything
- Old meeting notes
- Recovery test results
- Unrelated product plans
- Recovery acceptance criteria
- Repeated background material
Foreman V2 selects
- Recovery test results
- Recovery acceptance criteria
Relevant evidence, with source references.
60% fewer tokens on average
Foreman V2 internal benchmark tests. Workload-dependent; token reduction is not the same as invoice savings.
Inspect the context experimentsPut your API bill in perspective
What would less waste
look like on your bill?
Choose your monthly API spend. Compare it with a reduction scenario, then see what remains after the AetherGate subscription.
$0$100,000 /month
Model a provider-cost reduction
Default: 60%, anchored to Foreman V2's internal average token reduction. Applying it to provider costs is an illustrative assumption. Try a lower rate—or 0%—to pressure-test the case.
60% reduction scenario
Without optimization
$5,000.00 /mo
Modeled provider cost
$2,000.00 /mo
- Modeled provider savings
- $3,000.00/mo
- Team subscription
- $249.99/mo
- Provider cost + subscription
- $2,249.99/mo
Modeled net monthly savings
$2,750.01
$33,000.12 in modeled net annual savings.
Illustration, not a quote or a savings guarantee. The 60% internal result measures token reduction, not a verified 60% invoice reduction. Actual cost depends on input/output mix, model prices, workload, quality and added orchestration calls. Uses standard monthly plan pricing; excludes taxes and migration costs. Plan fit is suggested from baseline spend. Review the experiment and limitations.
Foreman V2 / Context + execution
Your AI doesn't need
the entire backstory.
It needs the right parts.
Long conversations, documents and tool results can follow a job from one call to the next. Sending all of them every time means paying to process the same background repeatedly—even when most of it has nothing to do with the current question.
Foreman V2 builds a smaller, task-relevant context and chooses an execution strategy around the job. It can answer directly, delegate independent work, or plan a multi-step task. More agents and more calls have to earn their place.
01 / FOREMAN V2
Understand the job before spending on it.
Foreman V2 analyzes the task, its requirements and its risk. A straightforward question can take a direct path. Work with independent parts can go to specialists. Planning is used when the job calls for it.
02 / FOREMAN V2
Give each model the context it needs.
Workspace material is split into traceable sections, ranked for relevance and selected for coverage. Foreman builds the prompt from the relevant evidence instead of attaching every available document. If the job needs no workspace context, it can send none.
03 / FOREMAN V2
Check the work without repeating all of it.
Deterministic checks look for incomplete output and unmet requirements. Independent verification is added when risk, ambiguity or the task warrants it. Repairs are bounded, and a complete single-task answer can pass through without a paid synthesis rewrite.
04 / FOREMAN V2
Keep the budget and the receipt together.
Paid steps reserve estimated spend before dispatch. The job stops when required work cannot fit its allowance. The result includes recorded costs, model calls and checks, so you can judge the outcome as well as the bill.
Technical details: how context is selected
Section-aware chunking keeps source IDs and offsets. BM25 relevance ranking uses stemmed, identifier-aware terms; coverage-aware selection handles questions with multiple parts. Relevance gates can exclude unrelated material, while conflict notices flag inconsistent evidence. Instruction-like source content is labelled and fenced as untrusted material. This is context engineering, not a promise that prompt injection is impossible.
Foreman V2 is the mission orchestration workflow. The OpenAI-compatible gateway provides routing, request admission and telemetry; ordinary gateway calls do not automatically run the Foreman context-selection pipeline.
Why AI keeps rereading your historyThe gateway around the work
A smaller prompt is a start.
Control the rest of the request.
Keep your provider relationships. Give your applications one place to enforce routes, limit access and account for usage.
Limits before the invoice.
Set application-key caps and workspace budgets. Requests reserve estimated cost before they leave, so parallel calls compete for the same available allowance.
Understand spending controlsCredentials with boundaries.
Encrypt stored provider keys, issue scoped gateway keys, and apply expiry and stronger credential policies. Five configurable protection levels let you tighten access as exposure grows.
Review key protectionA bill you can investigate.
See request history, provider usage, estimated costs, latency and fallback events. Find the models, credentials and applications behind your spending instead of working from one total.
Explore the workspaceStart with one workload
Your providers.
One controlled entry point.
Connect your accounts, issue a scoped application key and point a supported client at AetherGate. Test a representative slice of traffic before rolling it out further.
- 1
Connect. Add your provider API keys in the workspace.
- 2
Configure. Choose model routes, approved backups and spending limits.
- 3
Compare. Measure cost, useful output and latency against your direct baseline.
OpenAI-compatible gateway / Python example
import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["AETHERGATE_API_KEY"],
base_url="https://aethergate.site/api/v1",
)
# Use a model enabled on your account.
response = client.chat.completions.create(
model=os.environ["AETHERGATE_MODEL"],
messages=[{
"role": "user",
"content": "Summarize this request."
}],
)This example routes a chat-completions request. To use Foreman V2 planning and context selection, start a Foreman job in the workspace.
Evidence, then expansion
Make the case
with your own work.
Our internal tests make context-heavy tasks a useful starting point. They do not replace a comparison on the prompts, providers and quality requirements your product actually uses.
60%
Average token reduction in internal tests
This is the token-efficiency figure used to anchor the calculator's default scenario. Cost benchmarks, context experiments and latency tests measure different things. Complete-job spending can rise when added verification costs more than context selection saves.
One subscription. No token markup.
Keep the savings
as your usage grows.
Team
$249.99 /month
For growing teams, typically up to roughly $25K/month in provider spending.
50 provider connections · 50 gateway keys
Explore the 7-day trialEnterprise Growth
$999 /month
For companies with roughly $25K–$100K/month in provider spending.
250 provider connections · 250 gateway keys
View plan detailsEnterprise Scale
$2,499 /month
For larger portfolios with roughly $100K–$300K/month in provider spending.
1,000 provider connections · 1,000 gateway keys
View plan detailsProvider usage is billed separately. Spend ranges describe plan fit, not provider billing limits. Team trial requires a payment method and renews at $249.99/month after 7 days unless canceled. Discuss a custom workload.
Choose by the job
Compare the whole stack.
Gateway routing and Foreman orchestration solve different parts of the problem. Compare deployment, credential custody, subscriptions and complete-job cost alongside the features you need.
Before your first request
A few things,
made clear.
The mechanism, the bill and what to test.
Does Foreman V2 just switch me to a cheaper model?
Context selection and model routing are separate decisions. Foreman can reduce irrelevant context while keeping the lead model. A different model is considered under capability, cost and outcome rules; the cheapest model is not automatically the right one.
Will every request save 60%?
No. The 60% figure is an average token reduction from internal benchmark tests. History-heavy work has more context to remove; a short, already efficient prompt may gain little. Verification and orchestration add calls, so compare complete-job cost, answer quality and latency on your own tasks.
Is context selection the same as prompt injection?
Foreman selects relevant evidence and includes it in the model's prompt. We call that context selection or context engineering. Prompt injection is an attempt to make untrusted content override instructions. Foreman labels and fences instruction-like workspace content; that is a mitigation, not a guarantee against every attack.
Do I keep my provider accounts and pricing?
Yes. Connect your own provider API keys. Your providers bill their usage directly, and AetherGate adds no token markup or per-request surcharge. The AetherGate subscription is separate; provider quotas and terms still apply.
Do I need to replace my whole application?
The gateway exposes an OpenAI-compatible chat-completions endpoint for supported requests. Start with one workload, connect its providers and compare behavior. Foreman is a separate job workflow; routing existing API traffic through the gateway does not automatically turn every call into a Foreman job.
What should I measure during the trial?
Use representative tasks and the same acceptance criteria. Compare provider-reported input and output usage, complete-job cost, useful output, failure handling and latency. Include the subscription and any planning, verification or retry calls in your business case.
Start with the waste you can measure
Your next AI bill
deserves a better plan.
Connect your providers. Put limits around the work. Give Foreman V2 a representative task and see what changes.