Self-hosted AI gateway · one binary · your perimeter
Gatelodge sits in front of every LLM your teams use — cloud or on-prem — behind a single OpenAI-compatible endpoint, with the keys, budgets, and tamper-evident audit trail your security team already demands. The data never leaves your estate.
models · alias chains
| alias | routed to | p50 | $ / 1k | |
|---|---|---|---|---|
| smart | anthropic · sonnet 4.5 | 82ms | 3.00 | |
| fast | groq · gpt-oss-20b | 41ms | 0.15 | |
| vision | openai · gpt-4.1 | 119ms | 2.50 | |
| local | embedded · qwen3-8b | 128ms | 0.00 | |
| draft | unlisted provider | — | — |
don't-display: the unlisted model never renders — its requests return the same unknown-model error as any typo
The gatehouse in operation — one console page, the whole AI estate
Checkpoint 01 — The problem
Every organization building AI features hits the same wall — usually in the same order.
No rotation, no per-team accounting, no idea who can spend what.
→ Gateway keys per team and role · central credential vault · provider keys never appear in config or logs
No budgets, no cost attribution, no levers to pull when the number doubles.
→ Monthly token budgets with hard stops · per-team & per-request cost accounting · semantic cache — hits never bill
Compliance (SOC 2, EU AI Act) is arriving and AI usage leaves no record.
→ Per-request structured logs · hash-chained audit trail — tamper-evident by construction, not by policy
Every provider change means application surgery across the codebase.
→ One OpenAI-compatible endpoint in front of OpenAI, Anthropic, Gemini, Azure, Vertex, Bedrock, RunPod & local models · swap per-task with alias chains — zero app changes
Shadow AI: ungoverned, unlogged, unowned.
→ Self-hosted gateway, allowlisted backends · unapproved models don’t even display — they can’t be clicked
One upstream hiccup and the feature pages go dark.
→ Hedge routing — a slow primary loses the race to its fallback · chain plans with automatic failover
Checkpoint 02 — Capabilities
Everything below ships in the binary you download — the free edition carries the full governance stack.
Routing & scale
OpenAI-compatible, Anthropic Messages, Gemini Interactions, OpenAI Responses, Azure OpenAI, Vertex AI, Bedrock Converse, RunPod pods & serverless, and embedded local GGUF inference. Alias chains route per task; hedges and fallbacks ride along. Tier D replicas share state over Redis.
Governance
Role-based access (admin, read-only ops, key-manager), SSO/OIDC, SCIM provisioning, team workspaces, per-team and per-session budgets, full key lifecycle with revocation — and an append-only, hash-linked audit chain.
Cost control
A semantic cache where hits never bill. Monthly token budgets that stop overruns cold. Per-request, per-team, per-backend cost accounting. Route the cheap tasks to small or local models — flagship only when it matters.
Observability
OTLP tracing export, Prometheus metrics, per-request JSONL logs, and a live admin console — health, performance, syslog, governance, and license in one pane. The audit-chain widget renders tamper evidence visually.
Security posture
Self-hosted — prompts never leave your perimeter. Closed distribution: a machine-code binary, no source to fork or leak. Secrets resolve from env, files, commands, or the console-managed store — never committed, never logged.
Also at the desk
MCP tool-call governance · one-click provider presets · model finder and downloader · RunPod deploy wizard · hot-reloading config and licenses · tamper-evident revocation list · Windows & Linux · single-file install.
Checkpoint 03 — Return on investment
Direct savings from caching and routing, cost avoidance from audit readiness, and one governance layer across every provider.
Estimate your payback
| gross monthly savings | $2,913 |
|---|---|
| gateway cost (enterprise) | $975 |
| net monthly | $1,938 |
| return multiple | 3.0× |
estimate — rates validated with design partners. community edition: the gateway itself is free; every dollar saved is kept.
Checkpoint 04 — Editions
One product, two editions. The free binary is the full governance stack — not a demo.
Community — free
Free · 30-day trial of serving, one-time key removes the limit
Enterprise — paid
from $39 /user/month · or tiered per deployment ($499–$1,499/mo) · volume breaks
Self-hosted always. No SaaS tier — your prompts never transit our infrastructure, because there is no “our infrastructure.”
Checkpoint 05 — Channel partners
For inference platforms without a governance layer — which is all of them — Gatelodge ships as a 90-day co-branded trial for your customers. One binary, a license carrying your name, your endpoints pre-entitled.
A governed gateway configured for your endpoints locks the customer’s workflow onto your platform. Governance is the stickiness raw API access never had — and it gets you into procurement conversations that tokens alone never see.
Budgets, audit trail, key management, cost accounting — on day one, self-hosted, in their perimeter. Post-trial they buy the key; you keep the inference and a share of the license.
Checkpoint 06 — Get Gatelodge
The Community binary is the full gatehouse — governance, observability, cost-saver, local inference — free for a 30-day evaluation from first boot. No license file, no feature walls during the trial. Enter your email and take the gate home.
By downloading you agree to the End-User License Agreement.
CLEARED
Your download is ready.
The binary runs a full-feature 30-day trial from first boot — no license file needed.
Download gatelodge-community.exe156 MB · Windows · SHA-256 verified build
We also emailed your download link — keep that message for later.
The lodge — Contact
Request the community binary, book an enterprise walkthrough, or talk channel trials. We answer within a business day.