lean-ctx Enterprise is the self-hosted control plane for AI coding usage: per-person cost attribution, an org-wide model catalog, on-the-wire context compression and savings evidence your CFO can verify. Everything below is served by a running instance — open the console and check the numbers yourself.
One gateway between your agents and every model provider — because three problems show up in every org the month after AI tools roll out:
Provider invoices arrive as one line item. The gateway meters every request per person, team, project and model into your own Postgres — the console and a printable value report answer the CFO question with measured data, not estimates.
Clients call org aliases like demo/fast;
which real model answers is a gateway decision. Swap the model for 800 seats with
one config line — plus budgets, rate limits and a model ceiling under signed org policy.
Prompts are compressed on the wire; every event books actual cost and the counterfactual cost at a frozen reference model. Ed25519-signed evidence exports let finance and auditors verify the delta offline. On this instance, right now: measured live above.
Public, read-only access to the same three surfaces a customer deployment has. The credentials below are published on purpose: the admin API is GET-only and inference endpoints are not routed to the internet.
Spend, verified savings, the avoided-cost baseline and a live breakdown by person × project × model × provider — what a platform lead sees.
Each engineer signs in with their own gateway key and sees exactly their usage and savings — never anyone else's. Try the shared guest identity.
The printable ROI artifact finance gets: actual spend, verified savings and
the counterfactual baseline — regenerated every 15 minutes from the live
ledger with lean-ctx gateway report.
The gateway ships inside the Apache-2.0 open-source binary
(gateway-server feature, since v3.9.0).
Running the evaluation costs you nothing but a docker compose up —
you find out what you'd save before you spend anything.
lean-ctx gateway init scaffolds config, secrets, keys and compose;
Helm and Terraform templates exist for k8s and cloud. Runs entirely on
your infrastructure — provider keys never reach laptops.
Two env vars per IDE or agent (ANTHROPIC_BASE_URL + personal key).
No plugin, no client rollout — it speaks the providers' own wire protocols.
Real traffic accrues in your Postgres: per-person costs, compression savings, the counterfactual baseline against the model you'd otherwise pay list price for.
gateway report renders your value report;
gateway evidence exports Ed25519-signed aggregates your finance team
verifies offline. Your numbers — not our marketing.
One gateway between your coding agents and every model provider. Four verbs, all live in this demo:
SEE — per-person keys, metering to PostgresROUTE — org aliases demo/fast, demo/standardREMEMBER — prompts compressed on the wirePROVE — savings vs. a frozen reference modelEverything on this page is in the open-source binary — self-hosting the gateway is free.
# scaffold a complete instance (config, secrets, compose, keys) lean-ctx gateway init ~/gateway --org="Your Org" --seats=100 --person=you@example.com # bring it up (gateway + Postgres) and check it end-to-end cd ~/gateway && docker compose up -d lean-ctx gateway doctor # point any IDE or agent at it export ANTHROPIC_BASE_URL=http://your-host:8484 export ANTHROPIC_AUTH_TOKEN=<your personal gateway key>