Lunary
Open-source prompt management and observability for LLM apps
Lunary is profiled here as a Prompt Management tool for engineering teams. Read about features, pricing, and how it compares to related options in the tools directory.
Description
Lunary is an open-source platform for managing and monitoring LLM applications, built by Hugh Bzn and Vince Hardy. It pairs a prompt registry with full request logging, so a team can edit prompts, deploy new versions, and trace what production traffic actually did against them from one place. The codebase is Apache 2.0 and runs self-hosted, with a managed cloud for teams that skip operating it. Integration needs only a wrapper around the existing model client. The Apache-licensed core runs on a team's own servers, which suits organizations with data-residency requirements that rule out sending traffic to a third party.
Key Capabilities:
Prompt registry with versioning, templates, and environment-based deployment
Request tracing across completions, tool calls, and multi-step chains
Cost, latency, and token analytics per user, prompt, and model
LLM-as-judge and custom evaluators for output scoring
PII masking and a separate self-hosted analytics option for data control
Python and JavaScript SDKs with OpenAI-compatible logging
Alternative tools
- BAML
A domain-specific language for typed LLM functions
- Langtail
Collaborative prompt playground with testing and deployment
- W&B Weave
Trace, evaluate, and monitor LLM applications systematically
- Traceloop
OpenTelemetry-native tracing for LLM applications
- LangChain
The standard open-source framework for LLM applications
- Portkey
AI gateway with routing, guardrails, and prompt management