Live — a real gateway, not a mockup

Cut your AI coding bill — and prove it to the penny.

One self-hosted gateway in front of Claude Code, Cursor and your agents. It shows you who spends what, puts every model under your policy, shrinks prompts on the wire — and hands your CFO cryptographically signed proof of the savings. You're looking at a live instance right now.

Evaluate free on your own infrastructure. If it doesn't save, you don't pay a variable fee.

01 — Sound familiar?

You rolled out AI coding tools in weeks. Then came three conversations.

If you lead engineering or platform at an AI-heavy org, you've probably had all three this quarter.

“The AI budget doubled again. What are we actually getting for it?”
Your CFO
The gateway meters every request per person, team, project and model into your own database — and turns it into a value report finance can read, with a savings figure they can independently verify.
“Every team wired up a different model. We have no idea what's out there.”
Your platform lead
Engineers call org model names like acme/fast; which real model answers is your decision, in one config line — with budgets, rate limits and a model ceiling enforced org-wide.
“Which tool sees our code? Where does it go? Who approved it?”
Your CISO
Self-hosted, on your infrastructure. Provider keys leave the laptops and live in the gateway. Every change to models or policies is a reviewable config diff — there is no writable web admin to compromise.
One container in front of your model providers answers all three — with evidence, not opinions.
02 — How it works

One container. Two env vars per developer. That's the whole rollout.

No plugins, no IDE changes, no new workflow for your engineers — the gateway speaks the providers' own wire protocols.

Deploy it in front of your providers

One self-hosted container (Docker, k8s or air-gapped) between your coding tools and Anthropic, OpenAI, Azure AI Foundry or local models. lean-ctx gateway init scaffolds config, keys and compose in minutes. Provider keys move off laptops, into the gateway.

Every request is metered, compressed, routed

Each developer gets a personal key (ANTHROPIC_BASE_URL + token — done). The gateway books every request per person and project, compresses prompts on the wire, and resolves your org's model names to whatever you've approved.

You see, steer and prove

The console shows spend live. Policies enforce budgets and model ceilings org-wide. And every saving is written to an Ed25519-signed ledger your finance team verifies offline — no trust in us required.

03 — What you get

Four deliverables. All of them are open below — live.

This page runs on a real instance. The credentials are published on purpose: the admin API is GET-only and inference endpoints are not routed to the internet. Click through — what you see is what your rollout gets.

For your platform leadThe org console

Live spend, verified savings and the avoided-cost baseline — broken down by person × project × model × provider. The single pane your rollout is missing today.

Admin token (read-only API)
86f63be6a9491ddcc372ef4adef86c2ba96e5bfc7c9caf2c1d307dc173ce90eb
Open the console →

For every engineerA personal /me view

Each developer signs in with their own key and sees exactly their usage, cost and savings — never anyone else's. Cost awareness without surveillance.

Guest gateway key
gk-guest-demo-leanctx-com-aefc73310ee74382dcfb9a1411fc01969243b27627589f49
Open /me →

For your CFOThe value report

A printable ROI document: actual spend, verified savings, the counterfactual baseline and the methodology — regenerated from the live ledger every 15 minutes on this instance.

Open the report →

For your CISOGovernance as code

Budgets, rate limits, model ceilings and provider config live in a reviewable file under git — every change is a diff with an audit trail. The web console stays read-only by design: a browser session can never reroute your org's traffic.

Open the handbook →
Young numbers, honest numbers: this instance went live on July 6, 2026 and accumulates real history day by day — nothing is backfilled. It runs small open-source models priced at a transparent local shadow rate, so absolute dollar amounts are small; the savings share and the mechanics are exactly what a production deployment shows at enterprise scale.
04 — The 30-day proof

Get your own number — before you spend a franc.

The evaluation runs on the free, Apache-2.0 binary on your own infrastructure. At the end you hold a report about your traffic, your models and your savings — signed, and verifiable by your own auditors.

Day 1
Deploy

gateway init → container up next to your providers. Point 5–10 engineers at it: two env vars each. Their workflow doesn't change.

Weeks 1–4
Measure in shadow mode

The gateway meters and compresses real traffic. Together we freeze the baseline — which model each task class would have used without it. Measuring only; no customer traffic depends on it.

Day 30
Decide on evidence

gateway report renders your value report; gateway evidence exports signed aggregates. Numbers good → we talk rollout. Not good → you keep the free gateway, we part friends.

When you write us
What happens next

A personal reply within one business day, a 30-minute technical call (your questions, our architecture), then the deploy plan. No sales sequence, no discovery-call theater.

05 — Pricing

Start free. Pay a fixed setup — then a share of proven savings.

No per-seat toll on the core. The commercial parts are priced so the math always works in your favour: the fee is funded by money you verifiably stopped spending.

Free core
$0 forever
evaluate — or run a single team indefinitely

The gateway is part of the Apache-2.0 open-source binary. This demo runs 100% on it.

  • Metering per person, project, model — into your Postgres
  • Model routing with org aliases, incl. local models
  • Wire compression — the part that cuts the bill
  • Console, /me, value report and signed evidence
  • Budgets & rate limits on your instance, GDPR tooling
Enterprise license
Org rollout self-hosted, annual
when the whole organization runs through it

Everything a fleet needs that a single team doesn't:

  • SSO (OIDC) & SCIM — joiners/leavers sync from your IdP
  • Org-wide enforced policies — budgets & model ceilings pushed centrally
  • Multi-tenant console — strict separation per department or client project
  • Long audit retention + signed compliance exports
  • Support, SLA, security updates — a liable counterparty
The engagement
Setup + success fee capped
how we work with you

One-time fixed setup: deployment on your infra, SSO wiring, keys & policies, provider setup and the jointly signed baseline calibration (2–4 weeks shadow mode). Fixed price up front — no day-rate creep.

Then a capped share of verified savings — attributed request by request against the frozen baseline, signed, offline-verifiable. Renegotiated yearly as list prices fall.

What that looks like: in our standard pilot contract, a period with $100k of verified savings and a 20% take rate results in a ~$21k fee — capped so it can never exceed the agreed share of your actual provider-bill reduction. (Illustrative example from the contract template, not a customer figure.) Your number comes out of the 30-day proof. hello@leanctx.com

06 — Security & compliance

The answers your security review asks for first.

Built to pass a CISO's checklist before the first meeting — self-hosted, key-hygienic, auditable end to end.

Your code never leaves your infraGateway, database and (optionally) models run on your machines. Air-gapped deployment is supported — nothing phones home.
Provider keys off laptopsOrg keys live in the gateway's environment. Engineers hold personal, individually revocable keys, stored as SHA-256 hashes.
No writable admin surfaceThe console API is GET-only; configuration changes are git-reviewed file diffs applied on the host. Rollback is git revert.
Enforcement with honest errorsModel ceilings, per-person budgets and rate limits answer in the provider's own wire shape (403/429 + Retry-After) — tools degrade gracefully.
GDPR built inPer-person export and erasure (gateway gdpr export|delete), optional pseudonymization, bounded retention (90 days on this demo).
Fail-open for availabilityIf the metering database is briefly down, traffic keeps flowing. Bookkeeping never becomes your outage.
Identity at org scaleThe enterprise plane plugs into your IdP: SSO (OIDC) for sign-in, SCIM provisioning so joiners/leavers sync automatically — a leaver loses model access the moment HR offboards them.
No credential-vault blast radiusA single Rust binary with a GET-only console API — no writable admin panel in the data path, no Python plugin surface. Personal keys are stored as hashes; org provider keys stay in the gateway's environment.
07 — Questions your team will ask

Asked in every evaluation. Answered up front.

Can we just run the free version and never pay you?

Yes — genuinely. A team can self-host the open-source gateway forever; a public CI gate enforces that free capabilities never move behind a paywall. Organizations come back for the licensed plane when the rollout needs SSO/SCIM, centrally enforced policies, strict separation between departments or client projects, long audit retention and an SLA — plus the calibrated baseline and signed savings engagement. If the free core solves your problem, use it with our blessing.

Do our developers have to change how they work?

No. Each developer sets two environment variables (ANTHROPIC_BASE_URL and a personal key) — Claude Code, Cursor and scripts work unchanged, because the gateway speaks the providers' own wire protocols. No plugin, no proxy certificates, no new tool to learn.

Does our code or our prompts leave our infrastructure?

Only to the model providers you already use — and through your gateway on your infrastructure, under policies you set. lean-ctx (the company) sees nothing: the metering database is yours, the ledger stores aggregates, and air-gapped deployments are supported. The savings evidence is designed so auditors can verify it offline.

How is this different from LiteLLM, Portkey or a cloud AI gateway?

Routing, budgets and cost dashboards are commodity — good free tools exist, and if that's all you need, use them. Two things are unique here: on-the-wire context compression that reduces the bill itself (not just re-routes it), and the cryptographically signed savings ledger that turns "we think it helps" into evidence finance and auditors can verify independently. That proof is also what makes risk-free, savings-based pricing possible.

We run local models — what do "savings" even mean then?

Local inference is never booked at $0, because hardware and power are real. It's costed at a transparent, configurable shadow rate (this demo: $0.25 per million tokens, stated in the report's methodology). Compression still pays off in the local world: fewer tokens means less GPU time and more capacity per cluster.

Can our engineers audit the code before we trust it?

Please do. The engine is Apache-2.0 on GitHub — 8,500+ automated tests, a CI pipeline with a zero-warnings policy, cross-platform builds and CodeQL analysis. And the pilot is structured so trust is earned, not assumed: the first weeks run in shadow mode, read-only, with nothing critical depending on the gateway.

08 — Next step

Find out what your number is.

Two sentences are enough — your stack, your seat count, what hurts. Personal reply within one business day, then a 30-minute technical call if it fits.

hello@leanctx.com