llmgateway
Route, manage, and analyze your LLM requests across multiple providers with a unified API interface.
The Lens
By Erik Loyd, SaaS CEO and former COO/CFO of an AWS Premier Partner.
Updated Oct 2026
LLM Gateway sits between your app and the model providers: one OpenAI-compatible API that routes to OpenAI, Anthropic, Google and dozens more and tracks tokens and cost in one dashboard. It's an open source answer to OpenRouter, in the same lane as LiteLLM and Portkey, and the core is free under AGPLv3.
Self-hosting is one Docker image with Postgres and Redis inside, or a Helm chart when you mean it. The docs say to move Postgres and Redis to managed services for production, and releases land weekly, so upgrades are a standing chore.
Hosted has no subscription: provider token rates plus a flat 5% fee when you buy credits, and routing with your own keys is free. Enterprise (custom pricing) adds SSO, audit logs, guardrails and full payload retention, and self-hosters need a signed license for those. Solo: free hosted with your own keys. Small teams: self-host free. Larger teams: Enterprise once SSO and audit trails are requirements.
The catch: this layer holds every provider key and sees every prompt. August 2026 alone brought a critical cross-tenant API key bug and a cache that could hand one tenant's responses to another. All patched, but self-hosted, the upgrade cadence is yours.
Free vs Self-Hosted vs Paid
open coreFree: Hosted Free plan at $0 forever with no credit card: access to 200+ models across 40+ providers, API and Lounge chat access, activity logs, auto-routing, budgets, prompt caching and BYOK. Metadata-only logs. One free model rate limited to 20 requests/min; paid models are not rate limited. Community support on Discord.
Self-hosted: Core gateway, API, dashboard and worker are AGPLv3. Run the unified ghcr.io/theopenco/llmgateway-unified image (bundles Postgres and Redis) or Docker Compose, Helm on Kubernetes, and guides for AWS, Google Cloud and Azure. You supply provider API keys and two secrets. No platform fees when self-hosted. Code in ee/ (admin dashboard, audit logs, guardrails) is under a separate commercial license: production use requires a signed Enterprise license token, verified offline. Enterprise licenses bind to one organization; multi-org administration needs a White-label license; both carry a user cap.
Paid: Hosted usage is pay per token at each provider's rates plus a flat 5% platform fee on credit purchases (no seats, no minimums, no subscription). Non-US cards pay an extra 1.5% international fee. BYOK routing carries no platform fee. Enterprise is custom pricing (contact sales, 30-day trial): volume discounts, custom routing, full request and response retention for 30 days (billed $0.01 per 1M tokens), SSO/SAML with SCIM, audit logs, guardrails, contractual SLAs (99.9% uptime), white-label chat app, invoicing and a Slack support channel.
Self-hosting the AGPL core costs nothing, and hosted routing with your own provider keys is free too. Buy credits through them and you pay provider rates plus 5% (1.5% more on non-US cards); SSO, audit logs and guardrails are Enterprise, custom-priced.
What to do by team size
- Solo
- free hosted plan with your own provider keys
- Small team
- free self-hosted, or hosted credits at provider rates plus 5%
- Medium team
- free self-hosted if you can patch it promptly; Enterprise once you need SSO and audit logs
- Large team
- Enterprise or White-label license for SSO/SCIM, audit trails, guardrails and SLAs
Get tools like this every Wednesday
One featured tool, three on the radar. No fluff.
A low score is not a verdict on quality. Young and niche tools start low by design. How we calculate scores
Trust Signals
License: GNU Affero General Public License v3.0
Core is AGPLv3; the ee/ directory (admin, audit logs, guardrails) needs a paid Enterprise license in production.
Commercial use: ✓ Yes
About
- Owner
- theopenco (Organization)
- Stars
- 1,725
- Forks
- 201
Explore Further
More tools in the directory
rea
Reverse engineer anything with agents, from app behavior down to native binaries.
435.8k ★openclaw
Your own personal AI assistant. Any OS. Any Platform. The lobster way. 🦞
394.0k ★everything-claude-code
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
298.1k ★