LangDB (Singapore) is a managed AI gateway that exposes an OpenAI-compatible endpoint for models from different providers. The idea is to concentrate routing, policy and measurement in one place, without the application knowing each provider.
What it does
- Calls models from different providers through one OpenAI-compatible endpoint
- Auto Router picks the model by cost, speed or availability
- Fallback targets and virtual models group equivalent models
- Guardrails run before and after the call, with partner providers
- Tracing and analytics per project, request and model
- Cost control per project, with unified credits and dynamic limits
- Uses your own provider keys or the service’s credits
How it works
- The application points at the service endpoint and picks a model or virtual model
- The gateway resolves the provider, applies policies and logs the request
- Routing rules can weigh cost, latency and availability
- Tracing follows OpenTelemetry on the plans that export to external tooling
- The self-hosted edition runs in a container inside the customer infrastructure
Models and providers
- A declared catalog of hundreds of models
- Commercial providers and open models on the same endpoint
- Per-provider credentials or the service’s unified credits
Availability and license
- Proprietary managed service, with free, Professional, Business and Enterprise plans
- Self-hosted container execution available
- The gateway’s open-source code was discontinued: the linked repository now points to another product under a fair-code license
Points to note
- The gateway is no longer open source; the original self-hosted version is frozen
- The free plan caps logs and retention
- Load balancing across instances is not documented as a dedicated feature