CrowLLM
An OpenAI-compatible API aggregation gateway that aggregates models from multiple mainstream AI providers through a single endpoint and unified key. It supports switching providers by modifying the model string, failover, and usage-based billing.
Timeline (last 90 days)
CrowLLM provides an endpoint fully compatible with the OpenAI Chat Completions API specification, so users only need to change the model string to switch upstream AI providers, with no code changes required.
A single CrowLLM API key can call the Claude series, Gemini series, GPT series, and DeepSeek models as well as other models in the catalog.
CrowLLM provides a /v1/messages endpoint to support clients using the Anthropic format, including the count_tokens functionality.
When an upstream request fails, CrowLLM automatically retries the same request by routing it to other available channels serving the same model (failover).
CrowLLM promises that when an upstream returns a 429 status code, that failed attempt is not billed (failed requests are not charged repeatedly).
CrowLLM promises to pass the provider's cache read discount through to the user's bill, that is, cache reads are billed at the cache price.
CrowLLM does not enforce IP binding, and API keys can be used in any network environment.
CrowLLM divides model access into three tiers: a free tier (permanently free, no card binding required, Discord login), a credits tier (earning credits through Discord interaction to redeem calls), and a paid tier (billed by token, with pricing kept as low as possible below official vendor prices).
CrowLLM currently has about 7,000 users, its Discord community remains active, and the website is still under construction.
CrowLLM has not yet registered a company, and the author says he plans to consider incorporation only after reaching 100,000 users.
4 more timeline events
All Facts (14)
CrowLLM provides an endpoint fully compatible with the OpenAI Chat Completions API specification, so users only need to change the model string to switch upstream AI providers, with no code changes required.
50%UnverifiedA single CrowLLM API key can call the Claude series, Gemini series, GPT series, and DeepSeek models as well as other models in the catalog.
50%UnverifiedCrowLLM provides a /v1/messages endpoint to support clients using the Anthropic format, including the count_tokens functionality.
50%UnverifiedWhen an upstream request fails, CrowLLM automatically retries the same request by routing it to other available channels serving the same model (failover).
50%UnverifiedCrowLLM promises that when an upstream returns a 429 status code, that failed attempt is not billed (failed requests are not charged repeatedly).
50%UnverifiedCrowLLM promises to pass the provider's cache read discount through to the user's bill, that is, cache reads are billed at the cache price.
50%UnverifiedCrowLLM does not enforce IP binding, and API keys can be used in any network environment.
50%UnverifiedCrowLLM divides model access into three tiers: a free tier (permanently free, no card binding required, Discord login), a credits tier (earning credits through Discord interaction to redeem calls), and a paid tier (billed by token, with pricing kept as low as possible below official vendor prices).
50%UnverifiedCrowLLM currently has about 7,000 users, its Discord community remains active, and the website is still under construction.
50%UnverifiedCrowLLM has not yet registered a company, and the author says he plans to consider incorporation only after reaching 100,000 users.
50%UnverifiedCrowLLM was built independently by an 18-year-old developer.
50%UnverifiedCrowLLM's compatibility covers clients such as Cursor, Cline, and JanitorAI and SillyTavern, which are commonly used by the role-playing community.
50%UnverifiedCrowLLM's /status page shows only real traffic data, and the author says he would rather show a bad week than fabricate good-looking numbers.
50%UnverifiedCrowLLM's long-term stability and sustainability still need to be observed, its status page data depends on the developer's own statements, and at the current stage it may not be suitable as the sole dependency for enterprise-grade scenarios.
50%