Backmesh is an open-source backend-as-a-service (BaaS) designed specifically for AI applications, enabling applications to securely call LLM APIs without building their own backend.
Primary purpose The core goal is to address the security and cost risks posed by embedding LLM API keys directly in applications. By storing keys in the Backmesh backend, it prevents keys from being leaked when distributed with clients, thereby preventing massive API charges caused by abuse.
Key capabilities
- Key protection: Protects LLM API keys with military-grade encryption to avoid exposing them in applications.
- JWT authentication: Requests are verified through the JWT of the application's identity provider, ensuring that only legitimate users of the application can access LLM APIs through Backmesh.
- Per-user rate limiting: Per-user rate limits can be configured to prevent abuse, for example limiting each user to at most 5 OpenAI API calls per hour.
- API resource access control: Protects sensitive API resources (such as Files and Threads) so that only the user who created them can continue to access them.
- LLM user analytics: All LLM API calls are instrumented and logged, making it possible to identify usage patterns, reduce costs, and improve user satisfaction without installing additional packages.
How to use Backmesh runs as an open-source, well-tested backend. Developers connect their applications to Backmesh, which proxies LLM API calls; requests carry a JWT from the application's identity provider for verification; per-user rate limits can be configured; sensitive resources are protected by access control. The official text does not provide specific installation commands or code examples, and related details are not yet available.
Use cases
- AI applications that need to call LLM APIs from the frontend or mobile client but do not want to expose keys.
- Scenarios that need to control API call frequency per user and prevent abuse.
- Multi-user AI applications that need to protect users' private resources (such as files and conversation threads).
- LLM applications that want to optimize cost and experience through call data analysis.
Limitations Regarding pricing, supported LLM providers, specific deployment methods, performance metrics, etc., the official text does not yet provide information, and further consultation of the official documentation is needed.
