Log & debug LLM agents
Log all your prompts and results, see how agents are performing in production.
- Traces & error stack traces
- Instant search & filters
- Live tail
- Label data for fine-tuning
Stay on top of costs
Monitor requests and costs segmented by user and model.
- Auto cost & token calculations
- Optimize costs
- Costs by user, prompts & agents
Debug complex agents
Replay agent executions with traces and find out what went wrong and where.
- Trace agent executions
- Error stack traces
- Filter & instant search
Experiment with prompts and models to find the best performing ones.
- Fully no-code
- AI-powered assertions & tests
- Test open-source models
Replay user chats
Record user conversations and identify gaps in your chatbot's knowledge.
- Capture user feedback
- React hooks
- Dev-friendly frontend integration
Iterate on prompts
Create templates and collaborate on prompts with non-technical teammates.
- Clean your source-code
- A/B testing
Lunary is easy to integrate
Secure hosting for your data
Self-host on your servers
Keep the data inside your company for maximum data security. Learn more →
Inspect our source code
Our source code is 100% open source and available on GitHub. Learn more →
Data hosted in the EU
The cloud servers we use are located in Europe for GDPR compliance.
Full access to the data
Use our complete API or export directly from the dashboard. Learn more →
What our users say
Developers also love these
Live TailNew logs and runs are shown in realtime.
Instant SearchSearch across your whole data in milliseconds.
Label dataTag your data to easily filter and fine-tune.
Self HostSetup on your own servers for maximum data security.
PromptsStore, version and collaborate on prompts.
Data APIFetch your data to integrate it in other services.
AlertsSetup alerts to be notified for outlier results and errors.
Frontend SDKTrack chats and feedback direcly in your frontend.
PlaygroundTest your prompts on 10+ models.
Get started in minutes.
Run it on your own or get started in minutes with our hosted version.