Open source LLM gateway with cost routing

RelayPlane is an open source LLM gateway with cost routing built in: install it free under MIT, and it routes each request to the cheapest model that meets the task's complexity, then meters exactly what every request costs in a local ledger.

No paid tier, no fee on your API calls, and a hard spend cap enforced before a runaway job drains the budget. It runs on your machine across 6 native providers plus 300+ more through OpenRouter.

npm install -g @relayplane/proxy
relayplane start
export ANTHROPIC_BASE_URL=http://localhost:4100
# Cost routing is on by default. View the ledger:
relayplane watch

Open source LLM gateways with cost routing, compared

A check means the feature ships and is documented on their public source.

FeatureRelayPlaneLiteLLMBifrostHelicone
LicenseMIT, no paid tierOSS + paid enterprise tierApache 2.0Apache 2.0 (self-host) + paid cloud
Cost routing by task complexity, on by defaultYou configure routing rulesYou configure routing rulesObservability-first, routing is manual
Per-request cost ledger includedSpend logs in the proxy databaseLogs, self-hosted dashboard
Hard spend cap, enforced before the call leavesBudgets per virtual keyRate limits, budget via pluginsAlerts, not a hard block by default
Installnpm install -g @relayplane/proxypip / uv installGo binary / DockerDocker / hosted
Providers reachable6 native + OpenRouter for 300+ more100+ providers12+ providersProxy for major providers

Sources: LiteLLM README, Bifrost README, Helicone README, RelayPlane source (MIT).

Frequently asked questions

Is there a free, open source LLM gateway with cost routing built in?

Yes. RelayPlane is MIT licensed with no paid tier, and cost-based routing is on by default: it classifies each request by complexity and sends routine work to cheaper models automatically, while pricing every request into a local ledger.

What does "cost routing" mean in an LLM gateway?

Cost routing means the gateway picks which model handles a request based on how complex the task is and what each model costs, instead of always using the model your code named. RelayPlane's classifier is authoritative and routes down: a simple request goes to a cheaper model even if your app requested an expensive one, unless you bypass it.

Do open source LLM gateways track cost per request?

It varies. RelayPlane and Helicone both track cost per request out of the box. LiteLLM and Bifrost log spend too, but the detail and where it lives differs, LiteLLM in its proxy database, Bifrost in its self-hosted dashboard. RelayPlane writes a local SQLite ledger with model, tokens, cost, agent, and session for every request, no separate database to run.

What is the difference between RelayPlane and other open source LLM gateways?

RelayPlane is narrower by design: a local, single-machine proxy focused on cost visibility, automatic cost routing, and hard spend caps, no database, no Docker. LiteLLM and Bifrost are built for platform teams with virtual keys and broader provider coverage. Helicone leans toward observability over enforcement. Pick RelayPlane if you want the cost control and routing to just work on one machine for free.

Get the source

MIT licensed. The whole product is in the repo.

View on GitHub