Currai
LLM observability and prompt A/B testing platform.
āĻ¸ā§āĻā§āĻ°
60
āĻŦā§āĻ¯āĻŦāĻšāĻžāĻ°āĻāĻžāĻ°ā§āĻ° āĻ°ā§āĻāĻŋāĻ
3.0
āĻŦāĻŋāĻ¨āĻžāĻŽā§āĻ˛ā§āĻ¯ā§/āĻā§āĻ°āĻžāĻ¯āĻŧāĻžāĻ˛ āĻ¸āĻšāĻžāĻ¯āĻŧāĻ¤āĻž
āĻ¸āĻŽāĻ°ā§āĻĨāĻŋāĻ¤ āĻ¨āĻ¯āĻŧ
āĻļā§āĻ°ā§āĻ° āĻŽā§āĻ˛ā§āĻ¯
--
āĻ¸āĻ°ā§āĻŦāĻļā§āĻˇ āĻāĻĒāĻĄā§āĻ
āĻāĻāĻ¸ā§āĻ 04, 2026
āĻ¸āĻāĻā§āĻˇāĻŋāĻĒā§āĻ¤āĻ¸āĻžāĻ°
Currai āĻā§?
Currai is an observability and tracing platform designed for LLM applications. It helps developers move away from basic print statement debugging by tracing every prompt, token, and tool call in a single view. The platform enables monitoring production traffic, calculating costs based on processed bytes, running evaluations, and A/B testing prompt versions live without requiring any local infrastructure to run. It offers drop-in Python and TypeScript SDKs and is byte-compatible with Langfuse and OpenTelemetry, making it easy to integrate or migrate existing projects.
Currai āĻā§ āĻāĻžāĻā§ āĻŦā§āĻ¯āĻŦāĻšā§āĻ¤ āĻšāĻ¯āĻŧ?
āĻļā§āĻ°ā§āĻˇ āĻŦā§āĻļāĻŋāĻˇā§āĻā§āĻ¯
- Full tracing of prompts, tokens, and tool calls in one view
- Production prompt A/B testing and evaluation scoring
- Background trace batching and flushing to prevent application slowdown
- Byte-compatible drop-in support for Langfuse and OpenTelemetry
- Pricing model calculated on processed data volume (bytes) instead of flat trace counts
Currai āĻā§āĻāĻžāĻŦā§ āĻŦā§āĻ¯āĻŦāĻšāĻžāĻ° āĻāĻ°āĻŦā§āĻ¨?
To use Currai, install the SDK via pip or npm, initialize it using your public and secret keys, and wrap your LLM calls using the trace and generation methods to start capturing inputs, outputs, and token usage.
āĻŦāĻŋāĻāĻ˛ā§āĻĒ āĻā§āĻ˛
āĻā§āĻ¨ āĻŦāĻŋāĻāĻ˛ā§āĻĒ āĻā§āĻ˛ āĻĒāĻžāĻāĻ¯āĻŧāĻž āĻ¯āĻžāĻ¯āĻŧāĻ¨āĻŋ
āĻ¸ā§āĻŦāĻŋāĻ§āĻž & āĻ āĻ¸ā§āĻŦāĻŋāĻ§āĻž
āĻā§āĻ¨ āĻĄā§āĻāĻž āĻ¨ā§āĻ
āĻŦā§āĻ¯āĻŦāĻšāĻžāĻ°ā§āĻ° āĻā§āĻˇā§āĻ¤ā§āĻ°ā§
- Debugging complex, nested LLM application requests and chains
- Tracking and monitoring real-time API token usage and daily expenditures
- A/B testing new prompt variations against live production traffic
- Stitching multi-turn conversations together to track user sessions
āĻŦā§āĻ¯āĻŦāĻšāĻžāĻ°āĻāĻžāĻ°ā§āĻ° āĻĒāĻ°ā§āĻ¯āĻžāĻ˛ā§āĻāĻ¨āĻž
3.0
ā§Ģ āĻāĻ° āĻŽāĻ§ā§āĻ¯ā§
3.0
āĻāĻāĻ¨āĻ āĻā§āĻ¨ āĻĒāĻ°ā§āĻ¯āĻžāĻ˛ā§āĻāĻ¨āĻž āĻ¨ā§āĻ
āĻāĻ āĻā§āĻ˛ā§āĻ° āĻĒā§āĻ°āĻĨāĻŽ āĻĒāĻ°ā§āĻ¯āĻžāĻ˛ā§āĻāĻ¨āĻžāĻāĻžāĻ°ā§ āĻšāĻ¨!