Open Source
LiteLLM
by BerriAI
A Python SDK and a self-hosted proxy, the LiteLLM AI Gateway, that let apps call more than 100 model providers through one OpenAI-format API, with virtual keys, budgets, fallbacks and spend tracking.
Our verdict
A Python SDK and a self-hosted proxy, the LiteLLM AI Gateway, that let apps call more than 100 model providers through one OpenAI-format API, with virtual keys, budgets, fallbacks and spend tracking.
Best for: Platform teams running a shared gateway for several apps that need per-team keys, budgets and spend logs across providers.
Choose something else if: you cannot pin versions, patch quickly and rotate keys. LiteLLM's release pipeline was compromised twice in 2026, in March and October.
Source review
Using LiteLLM in practice
Where it runs
Self-hosted with Docker, Helm or Terraform. Prompts go only to the providers you configure, and telemetry is off by default when self-hosted.
How to install
For a quick test, install the proxy with uv and start it on one model. For a shared gateway, use the Docker Compose quickstart with Postgres, set a master key and pin an explicit version. Clients call http://localhost:4000.
Install commands
uv tool install 'litellm[proxy]'
litellm --model gpt-4oAs published in the linked sources. Not run by us.
What it costs
The open-source gateway is free to self-host. You pay your model providers directly, plus your own hosting. Enterprise adds governance and support, priced through sales on annual gateway capacity and never per token.
Prices and plans change. Check the linked sources before committing.
Why consider it
- Very wide provider coverage behind one OpenAI-format API, from Python or over HTTP.
- The open-source proxy includes virtual keys, budgets, rate limits, fallbacks, logging callbacks and an admin UI.
What to watch for
- On 21 October 2026 a compromised release, 1.107.0 on PyPI and the Docker latest and main-stable tags, carried credential-stealing malware for a few hours. LiteLLM says to pin 1.106.2 or an earlier explicit version.
- Budgets and virtual keys need a Postgres database.
- Single sign-on beyond five users, audit logs and several guardrails need a paid Enterprise licence.
A first task to try
Start it locally with one provider key, send one request to http://localhost:4000, then find that request in the spend log.
Suggested exercise, not an installation test. We reviewed the linked documentation; no host compatibility is implied.
Other options to understand
Related tools can solve different parts of the same job. Compare fit before treating them as replacements.
Key facts
What we have checked
- Latest version
- 1.106.4 (released 25 Oct 2026)
- GitHub stars
- 59,647
Each value shows when we last confirmed it and where it came from. Values we have not confirmed are left out, never guessed.
Stars and usage
How LiteLLM is growing
48,950 to 59,647
Alternatives
Alternatives to LiteLLM
| Record | Type | How we know | Checked |
|---|---|---|---|
| AI GatewayAlternative to this | Open source | Named in the listing | Not dated |
| AI studioAlternative to this | Open source | Named in the listing | Not dated |
| BifrostAlternative to this | Open source | Named in the listing | Not dated |
| BitRouterAlternative to this | Open source | Named in the listing | Not dated |
| Continuum-AI-Corp/orcarouter-mcp-serverAlternative to this | MCP server | Named in the listing | Not dated |
| Correctover/mcp-serverAlternative to this | MCP server | Named in the listing | Not dated |
| EntrolyAlternative to this | Open source | Named in the listing | Not dated |
| GlideAlternative to this | Open source | Named in the listing | Not dated |
| GoModelAlternative to this | Open source | Named in the listing | Not dated |
| HypersigilAlternative to this | Open source | Named in the listing | Not dated |
| JamesANZ/cross-llm-mcpAlternative to this | MCP server | Named in the listing | Not dated |
| ManifestAlternative to this | App | Named in the listing | Not dated |
| Maxim AIAlternative to this | Open source | Named in the listing | Not dated |
| MirascopeAlternative to this | Open source | Named in the listing | Not dated |
| Modelz-LLMAlternative to this | Open source | Named in the listing | Not dated |
| OpenRouterAlternative to this | App | Named in the listing | Not dated |
| RelayAlternative to this | Open source | Named in the listing | Not dated |
| ReliableGPT 💪Alternative to this | Open source | Named in the listing | Not dated |
| sachinuppal/modelcostsaverAlternative to this | MCP server | Named in the listing | Not dated |
| smigolsmigol/llmkitAlternative to this | MCP server | Named in the listing | Not dated |
| spanlens/SpanlensAlternative to this | MCP server | Named in the listing | Not dated |
| TeamoRouterAlternative to this | Open source | Named in the listing | Not dated |
| TokenMixAlternative to this | Open source | Named in the listing | Not dated |
| TrueFoundryAlternative to this | Open source | Named in the listing | Not dated |
| Vercel AI GatewayAlternative to this | Open source | Named in the listing | Not dated |
| XiuRouterAlternative to this | Open source | Named in the listing | Not dated |
| ypollak2/llm-routerAlternative to this | MCP server | Named in the listing | Not dated |
| FastChatThis is an alternative to it | Open source | Named in the listing | Not dated |
Compare
Compare LiteLLM side by side
Sources
Where this comes from
- GitHubOfficial
- Repository and READMEOur review
- Docker quickstartOur review
- PricingOur review
- 21 October 2026 security incidentOur review
- github api via nightly crawl, recovered from git history 20e874270887c1008f860ee0198e461a23c9e438Registry
A source shows what was published when we checked it. It does not certify security, performance or a working installation. Read our method.
Changelog
What changed
Released 1.106.4
Released 1.105.5
Released 1.104.7
Released 1.103.9
Released 1.106.3
Released 1.105.4
Released 1.104.6
Released 1.103.8
Released 1.106.2
Released 1.105.3
Released 1.103.7
Released 1.104.5
Released 1.106.1
Released 1.105.2
Get the free Developer’s Field Guide
A 27-page field guide to the AI coding workflow with Claude. Claude Code, MCP servers, the prompt patterns that work, and what to delegate. Free.
Enter your work email. We send it straight over, plus a few short notes worth knowing. Unsubscribe any time.