Native functions & variants
Define a task once, run several prompts/models/strategies behind it, with retries and cross-model fallback built in. This is the LSD-native way to call a model. Learn more.
LSD
An LLM gateway for Unix, speaking to every major provider. One binary, one database.
LSD (LLM Service Daemon) is an LLM gateway for Unix systems. It speaks natively to every major model provider, plus OpenAI- and Anthropic-compatible APIs so your existing clients work unchanged, with routing, fallbacks, and retries built in. Inferences, feedback, and configuration all live in a single PostgreSQL database, so there is no extra datastore to operate.
One database
Inferences, feedback, and config live in PostgreSQL. No separate analytics store or cache to run.
Built for Unix
Linux, FreeBSD, DragonFlyBSD, and OpenBSD are all supported. Lightweight enough for modest hardware.
Telemetry-free
No phone-home analytics, no usage tracking. The only network calls LSD makes are the ones you configure.
Open source
Apache 2.0 licensed, no CLA required. Free to run, modify, and self-host.
OpenAI/Anthropic-compatible
An easy on-ramp, not the destination. Point existing OpenAI or Anthropic clients at LSD unchanged, then move to the native API for functions, variants, and trips.
Capabilities
Section titled “Capabilities”Functions & Variants
Structured outputs, tool use, retries, and cross-model fallback, all native to LSD. Learn more.
Gateway
Call any of ~19 supported providers natively, or through the OpenAI/Anthropic-compatible APIs, with routing, fallbacks, retries, and batch inference. Learn more.
Experimentation
A/B test variants with static weights, or let an adaptive bandit shift traffic toward the best performer. Learn more.
Observability
Every inference and feedback signal is stored in Postgres and queryable programmatically. Learn more.
Optimization
Use collected metrics and feedback to optimize prompts, models, and inference strategies. Learn more.
Evaluations
Benchmark individual inferences or end-to-end workflows with heuristics and LLM judges. Learn more.
Get started
Section titled “Get started”The Quickstart walks through standing up an LLM application on LSD in minutes, calling the native /inference endpoint from the very first request.