Feature	LiteLLM Proxy	One API (songquanpeng/one-api)
Primary purpose	AI Gateway / OpenAI-compatible proxy for many providers, with cost tracking + routing features. (GitHub)	OpenAI-compatible “API management & redistribution” gateway with UI, tokens, channels, quotas/billing. (pkg.go.dev)
Docker deploy	Official Docker deployment paths and images; includes a dedicated “database” image for DB-backed mode. (LiteLLM)	Official Docker image (`justsong/one-api`) and docker-compose examples. (GitHub)
DB requirement	Optional overall; but DB-backed features (“Virtual Keys”, spend tracking, etc.) require PostgreSQL via `DATABASE_URL`. (LiteLLM)	Can run with SQLite (default/quick) or MySQL via `SQL_DSN`; docs warn that higher concurrency “must set `SQL_DSN`” (i.e., use MySQL). (GitHub)
SQLite support	❌ (not supported for the proxy’s DB-backed features; Postgres is the documented requirement). (LiteLLM)	✅ (explicit “use SQLite” Docker instructions; `docker-compose.yml` comment says you can comment `SQL_DSN` to use SQLite). (GitHub)
Key management / multi-tenant tokens	“Virtual Keys” (admin master key → generate keys). (LiteLLM)	Token management (expiry, quota, IP allowlist, model allowlist) + user management flows. (pkg.go.dev)
Quotas / budgets	Budget limits via Virtual Keys (DB mode). (LiteLLM)	Quota system + quota details + group multipliers. (pkg.go.dev)
Cost / usage tracking	Explicit “Track Spend” positioning in DB-backed deployment; also “cost tracking” called out in project overview. (LiteLLM)	Quota + billing system with usage tracking / cost calculations across providers/models (as described in project docs). (DeepWiki)
Provider aggregation	“Call 100+ LLM APIs” (gateway/SDK). (GitHub)	Many providers listed (OpenAI/Azure, Claude, Gemini, DeepSeek, Ollama, etc.), plus “third-party proxy services”. (pkg.go.dev)
Load balancing / routing	Load balancing is a first-class feature. (GitHub)	Supports load balancing across channels. (pkg.go.dev)
Streaming	Supported (proxy supports typical OpenAI streaming usage; generally used with OpenAI SDKs). (LiteLLM)	Explicitly supports stream mode. (pkg.go.dev)
OpenAI-client compatibility	Designed to be used with OpenAI SDKs; docs show using OpenAI Python client against the proxy base URL. (LiteLLM)	“Access all models via standard OpenAI API format”; intended for OpenAI-compatible clients by setting API Base + Key. (pkg.go.dev)
UI / admin panel	Has an admin/UI story (esp. in DB-backed/key-management mode). (GitHub)	Built-in web admin UI; repo describes “English UI”. (GitHub)
Notable operational caveat	If you want key management/spend tracking, you’re signing up for Postgres (and usually more “gateway” ops). (LiteLLM)	SQLite is fine for simple/small setups, but docs strongly recommend MySQL (`SQL_DSN`) for higher concurrency. (GitHub)

| Feature | LiteLLM Proxy | One API (songquanpeng/one-api) | | ------------------------------------ | --------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------- | | Primary purpose | AI Gateway / OpenAI-compatible proxy for many providers, with cost tracking + routing features. ([GitHub][1]) | OpenAI-compatible “API management & redistribution” gateway with UI, tokens, channels, quotas/billing. ([pkg.go.dev][2]) | | Docker deploy | Official Docker deployment paths and images; includes a dedicated “database” image for DB-backed mode. ([LiteLLM][3]) | Official Docker image (`justsong/one-api`) and docker-compose examples. ([GitHub][4]) | | DB requirement | **Optional** overall; but DB-backed features (“Virtual Keys”, spend tracking, etc.) require **PostgreSQL via `DATABASE_URL`**. ([LiteLLM][3]) | Can run with **SQLite** (default/quick) or **MySQL** via `SQL_DSN`; docs warn that higher concurrency “must set `SQL_DSN`” (i.e., use MySQL). ([GitHub][4]) | | SQLite support | ❌ (not supported for the proxy’s DB-backed features; Postgres is the documented requirement). ([LiteLLM][3]) | ✅ (explicit “use SQLite” Docker instructions; `docker-compose.yml` comment says you can comment `SQL_DSN` to use SQLite). ([GitHub][4]) | | Key management / multi-tenant tokens | “Virtual Keys” (admin master key → generate keys). ([LiteLLM][5]) | Token management (expiry, quota, IP allowlist, model allowlist) + user management flows. ([pkg.go.dev][2]) | | Quotas / budgets | Budget limits via Virtual Keys (DB mode). ([LiteLLM][5]) | Quota system + quota details + group multipliers. ([pkg.go.dev][2]) | | Cost / usage tracking | Explicit “Track Spend” positioning in DB-backed deployment; also “cost tracking” called out in project overview. ([LiteLLM][3]) | Quota + billing system with usage tracking / cost calculations across providers/models (as described in project docs). ([DeepWiki][6]) | | Provider aggregation | “Call 100+ LLM APIs” (gateway/SDK). ([GitHub][1]) | Many providers listed (OpenAI/Azure, Claude, Gemini, DeepSeek, Ollama, etc.), plus “third-party proxy services”. ([pkg.go.dev][2]) | | Load balancing / routing | Load balancing is a first-class feature. ([GitHub][1]) | Supports load balancing across channels. ([pkg.go.dev][2]) | | Streaming | Supported (proxy supports typical OpenAI streaming usage; generally used with OpenAI SDKs). ([LiteLLM][3]) | Explicitly supports stream mode. ([pkg.go.dev][2]) | | OpenAI-client compatibility | Designed to be used with OpenAI SDKs; docs show using OpenAI Python client against the proxy base URL. ([LiteLLM][7]) | “Access all models via standard OpenAI API format”; intended for OpenAI-compatible clients by setting API Base + Key. ([pkg.go.dev][2]) | | UI / admin panel | Has an admin/UI story (esp. in DB-backed/key-management mode). ([GitHub][8]) | Built-in web admin UI; repo describes “English UI”. ([GitHub][9]) | | Notable operational caveat | If you want key management/spend tracking, you’re signing up for Postgres (and usually more “gateway” ops). ([LiteLLM][3]) | SQLite is fine for simple/small setups, but docs strongly recommend MySQL (`SQL_DSN`) for higher concurrency. ([GitHub][4]) | [1]: https://github.com/BerriAI/litellm?utm_source=chatgpt.com "GitHub - BerriAI/litellm: Python SDK, Proxy Server (AI Gateway) to call ..." [2]: https://pkg.go.dev/github.com/songquanpeng/one-api?utm_source=chatgpt.com "one-api command - github.com/songquanpeng/one-api - Go Packages" [3]: https://docs.litellm.ai/docs/proxy/deploy?utm_source=chatgpt.com "Docker, Helm, Terraform | liteLLM" [4]: https://github.com/songquanpeng/one-api/blob/main/README.md?plain=1&utm_source=chatgpt.com "one-api/README.md at main · songquanpeng/one-api · GitHub" [5]: https://docs.litellm.ai/docs/proxy/virtual_keys?utm_source=chatgpt.com "Virtual Keys - liteLLM" [6]: https://deepwiki.com/songquanpeng/one-api/3.4-quota-and-billing?utm_source=chatgpt.com "Quota and Billing | songquanpeng/one-api | DeepWiki" [7]: https://docs.litellm.ai/docs/providers/openai_compatible?utm_source=chatgpt.com "OpenAI-Compatible Endpoints | liteLLM" [8]: https://github.com/BerriAI/litellm/blob/main/docs/my-website/docs/proxy/virtual_keys.md?utm_source=chatgpt.com "litellm/docs/my-website/docs/proxy/virtual_keys.md at main - GitHub" [9]: https://github.com/songquanpeng/one-api?utm_source=chatgpt.com "GitHub - songquanpeng/one-api: LLM API 管理 & 分发系统，支持 OpenAI、Azure ..."

skobkin pinned this

2026-01-07 04:15:22 +03:00

skobkin unpinned this

2026-05-01 01:11:00 +03:00