LiteLLM

license

MIT(core)

enterprise-dir

LiteLLM-Commercial-License

source

public

core-gated

gated(the enterprise/ directory ships under the proprietary BerriAI Enterprise License requiring a paid subscription for production use, and the proxy gates premium features on a LITELLM_LICENSE key verified against license.litellm.ai)

An MIT core with a proprietary enterprise directory in the same repository - and the gating is in the code, not merely in BerriAI selling a hosted gateway. The root LICENSE is explicitly split: "All content that resides under the 'enterprise/' directory of this repository, if that directory exists, is licensed under the license defined in 'enterprise/LICENSE'", and everything outside it is MIT. That enterprise license says the software "may only be used in production, if you (and any entity that you represent) have agreed to, and are in compliance with, the BerriAI Subscription Terms of Service ... and otherwise have a valid BerriAI Enterprise license for the correct number of user seats", with copying, distribution and sale forbidden. The directory is not a stub - it ships the enterprise package, hooks, UI and a CloudFormation stack - and the proxy enforces it at runtime, reading a license key from the environment and verifying it against license.litellm.ai before enabling premium features. A license key you must hold for paid features to run is what makes this open core rather than fully open. That is the other side of the line from LangChain, LlamaIndex, Pydantic AI, Zed and Otari, which sell a platform beside a complete open core.

• https://raw.githubusercontent.com/BerriAI/litellm/main/LICENSE recorded 2026-08-11

  Root LICENSE body, read rather than assumed. It opens "Portions of this software are licensed as follows:" and then splits the repository in two - "All content that resides under the 'enterprise/' directory of this repository, if that directory exists, is licensed under the license defined in 'enterprise/LICENSE'" and "Content outside of the above mentioned directories or restrictions above is available under the MIT license as defined below", followed by the full MIT text. So the tier-setting license is MIT and the gate is a carve-out named in the license itself.

• https://api.github.com/repos/BerriAI/litellm/contents/enterprise recorded 2026-08-11

  The carved-out directory exists and is not a stub. It contains LICENSE.md, README.md, __init__.py, cloudformation_stack, dist, enterprise_hooks, enterprise_ui, litellm_enterprise and its own pyproject.toml - hooks, a UI and a deployment stack, i.e. shipped functionality. The repository license endpoint returns NOASSERTION rather than MIT, which is GitHub declining to classify the split root LICENSE.

• https://raw.githubusercontent.com/BerriAI/litellm/main/enterprise/LICENSE.md recorded 2026-08-11

  The enterprise license body. It states the software "may only be used in production, if you (and any entity that you represent) have agreed to, and are in compliance with, the BerriAI Subscription Terms of Service ... or other agreement governing the use of the Software, as agreed by you and BerriAI, and otherwise have a valid BerriAI Enterprise license for the correct number of user seats". Copying for development and testing is allowed; "it is forbidden to copy, merge, publish, distribute, sublicense, and/or sell the Software". Production use behind a seat-counted paid license is functionality withheld from the MIT source.

• https://raw.githubusercontent.com/BerriAI/litellm/main/litellm/proxy/auth/litellm_license.py recorded 2026-08-11

  The enforcement, in code. The module's own header comment is "If litellm license in env, checks if it's valid". LicenseCheck reads a LITELLM_LICENSE environment variable, carries a bundled public_key.pem for air-gapped verification, and otherwise checks the key against base_url https://license.litellm.ai, setting the premium_user flag the proxy tests before serving paid features. A key you must hold for functionality to run, which is the rubric's own wording for a gate.

• https://api.github.com/repos/BerriAI/litellm recorded 2026-08-11

  Repository root. The published tree is the whole gateway and proxy - litellm, gateway, ui, backend, helm, terraform, docker, migrations, tests - and self-hosting it needs nothing that is not in the repo, so the source dimension is public. Retained from the phase-C pass with a real reading in place of the placeholder.

• https://www.litellm.ai/ recorded 2026-06-04

  Vendor site, kept for context only. It establishes that BerriAI sells a hosted product, which under the settled rule is NOT evidence for or against core-gated and is not what the 4 rests on.

741,403,911 PyPI downloads in the trailing 30 days for litellm, which is above 10M and lands in the top band. The figure is internally consistent across windows (207,823,088 weekly, 28,463,811 daily) - litellm is a proxy library pulled on nearly every CI run that touches a model provider. GitHub shows 21.7k dependent repositories, roughly 49.2k stars and 1,342 releases, none of which is the basis for the band.

• https://github.com/BerriAI/litellm recorded 2026-08-15

  Repository page for BerriAI/litellm: 56.4k stars and 10.6k forks. Corroboration only — the level 5 band rests on the measured PyPI count in the source above it, not on stars.

• https://pypistats.org/api/packages/litellm/recent recorded 2026-08-13

  741,403,911 downloads in the trailing 30 days for litellm; 207,823,088 weekly and 28,463,811 daily

A unified gateway to a very large set of LLM providers in OpenAI format, with cost tracking, guardrails, load balancing, virtual keys and budgets, and logging. No standard benchmark applies to a gateway, so this is rated on breadth of provider coverage and production features. The README and docs now say 100+ providers where the recorded feature list says 140+, so the headline count has been restated downward rather than grown; the score does not turn on that number.

• https://github.com/BerriAI/litellm recorded 2026-08-15

  Repository page describing the gateway: a single interface over many provider endpoints, with routing, fallbacks and spend tracking. Establishes the feature breadth the band rests on.

• https://docs.litellm.ai/ recorded 2026-08-15

  'LiteLLM - Getting Started': 'Call 100+ LLMs using the OpenAI Input/Output Format', translating inputs to /chat/completions, /responses, /embeddings, /images, /audio and /batches; consistent output across providers; retry and fallback logic across deployments via the Router; spend tracking and per-project budgets; available as a Proxy Server or a Python SDK.