Self-hosted and open source - verified
LiteLLM Proxy (AI Gateway)
LiteLLM Proxy is listed under self-hosted and open source. Its first listed fee, in its own words: open source (MIT, outside enterprise/ directory), self-hosted software. It states OpenAI-compatible and Anthropic-compatible requests. Routing stated: fallbacks and load balancing. Read from the vendor's own pages on October 18, 2026.
Read from 4 official sites. Docs - Pricing - Compare every gateway
Status
- Status, as we read it
actively maintained (October 2026 town hall announced in docs)inferred
Our reading: The page implies this status without labelling it.
“LiteLLM October Town Hall.”
docs.litellm.ai
How LiteLLM Proxy charges
Every fee line LiteLLM Proxy states, in its own words and units, with the quote. Nothing is converted.
| Product | Fee, in the vendor's words | Applies to | Quote and source |
|---|---|---|---|
| LiteLLM Proxy (AI Gateway) | open source (MIT, outside enterprise/ directory) | self-hosted software | “available under the MIT license as defined below” |
| Custom pricing | Enterprise (SSO/SCIM, audit logs, SLAs) | “Custom pricing, procurement, and a walkthrough for your team.” |
Access and keys
- Free tier
- Not stated
Open-source software; free tier not applicable.
- Bring your own key
yes (self-hosted; you configure provider credentials and secret manager)inferred
Our reading: Read as bring your own key from key management with your own secret manager on a self-hosted server.
“We handle key management and work with the secret manager you already run”
litellm.ai - BYOK fee
- Not stated
Not applicable to self-hosted software.
API formats
- OpenAI-compatible
yes
“It gives your applications one OpenAI-compatible endpoint for 100+ LLM providers, MCP tools, and A2A agents.”
docs.litellm.ai - Anthropic-compatible
yes (Anthropic v1/messages format)
“Use LiteLLM to call all your LLM APIs in the Anthropic v1/messages format.”
docs.litellm.ai - OpenAI base URL
http://localhost:4000 (self-hosted default)
“base_url="http://localhost:4000"”
docs.litellm.ai - Anthropic base URL
http://0.0.0.0:4000/v1/messages (self-hosted example)
“curl -L -X POST 'http://0.0.0.0:4000/v1/messages'”
docs.litellm.ai
Routing
- Fallbacks
yes (load balancing, fallbacks, retries)
“Load balancing, fallbacks, and retries move traffic between deployments when a provider has a problem.”
docs.litellm.ai - Load balancing
yes
“Load balancing, fallbacks, and retries move traffic between deployments when a provider has a problem.”
docs.litellm.ai - Automatic model routing
- Not stated
Not stated in the checked pages.
Caching
- Exact-match cache
yes (redis, s3, gcs, local, disk)
“Exact-match caches (redis, s3, gcs, local, disk) key on a hash of the whole request”
docs.litellm.ai - Semantic cache
yes (Redis, Valkey, Qdrant semantic)
“Semantic caches embed the prompt and serve the closest”
docs.litellm.ai
Controls and observability
- Spend controls
budgets and rate limits per virtual key
“Each key has its models, budgets, and rate limits.”
docs.litellm.ai - Rate limits
per virtual key
“Each key has its models, budgets, and rate limits.”
docs.litellm.ai - Guardrails
content filters, PII masking, policies
“Apply content filters, PII masking, and policies to requests and responses.”
docs.litellm.ai - Observability
spend tracking, logs, Admin UI
“The gateway records the cost of each request.”
docs.litellm.ai - Telemetry and log export
Langfuse, Datadog, OpenTelemetry, Prometheus and others
“Send logs and metrics to Langfuse, Datadog, OpenTelemetry, Prometheus, and other tools.”
docs.litellm.ai
Data, deployment and license
- Data logging and retention
- Not stated
Self-hosted; not stated.
- Regions
- Not stated
Self-hosted; the enterprise page mentions multi-region and air-gapped deployments.
- Self-hosting
yes (Docker, Helm, Terraform)
“The LiteLLM AI Gateway is a self-hosted server.”
docs.litellm.ai - License
MIT (content under enterprise/ uses a separate enterprise license)
“Content outside of the above mentioned directories or restrictions above is available under the MIT license as defined below.”
raw.githubusercontent.com
Vendor-stated counts
Model and provider counts as LiteLLM Proxy states them, dated. When two pages disagree, both are shown.
140+ providers (homepage) (vendor-stated, Oct 18 2026)
“One OpenAI-compatible API to 140+ providers and 1,800+ models.”
litellm.ai 1,800+ models (homepage) (vendor-stated, Oct 18 2026)
“One OpenAI-compatible API to 140+ providers and 1,800+ models.”
litellm.ai 100+ LLM providers (docs) (vendor-stated, Oct 18 2026)
“one OpenAI-compatible endpoint for 100+ LLM providers”
docs.litellm.ai
Research notes
Homepage (140+ providers) and docs (100+ providers) state different counts; both recorded.
FAQ
How much does LiteLLM Proxy cost?
In LiteLLM Proxy's own terms: open source (MIT, outside enterprise/ directory) (self-hosted software); Custom pricing (Enterprise (SSO/SCIM, audit logs, SLAs)). As of October 18, 2026; fees are not converted.
Does LiteLLM Proxy support bring your own key?
Yes (self-hosted; you configure provider credentials and secret manager) (our reading of the quote), per the vendor: "We handle key management and work with the secret manager you already run". BYOK fee: not stated. Not applicable to self-hosted software.
Is LiteLLM Proxy OpenAI-compatible or Anthropic-compatible?
OpenAI: yes, base URL http://localhost:4000. Anthropic: yes (Anthropic v1/messages format), base URL http://0.0.0.0:4000/v1/messages.
Sources
Last verified from each vendor's own docs, pricing pages, READMEs and license files; every quote was string-matched against the page that day. Fees are shown in the vendor's own words and change without notice, so check the linked source before you commit.
Newsletter
Know when a gateway changes its fees
One email per week. New model releases, routing and pricing changes and what they mean for what you build.