Replace LangSmith.
11 open source alternatives, ranked by health. We re-check every project weekly, so you don’t migrate to something abandoned.
Ranked by health
- LangfuseLLM engineering platform for tracing, evaluations, prompt management and metrics of LLM apps.★ 35k custom TypeScript self-host managedNot scored yet
- MLflowPlatform for tracing, evaluating and versioning LLM apps, agents and ML models, plus a prompt registry.★ 28k Apache-2.0 Python self-host managedNot scored yet
- OpikPlatform for tracing, evaluating and monitoring LLM applications, RAG systems and agents.★ 22k Apache-2.0 Python self-host managedNot scored yet
- PhoenixLLM tracing, evaluation and prompt experimentation platform built on OpenTelemetry.★ 12k Elastic License 2.0 Python self-host managedNot scored yet
- HeliconeAI gateway and LLM observability platform that logs requests, traces agent sessions and tracks cost and latency.★ 6.2k Apache-2.0 TypeScript self-host managedNot scored yet
- Coze LoopPlatform for developing, debugging, evaluating and monitoring AI agents, including prompt management and tracing.★ 5.8k Apache-2.0 Go self-host managedNot scored yet
- LangWatchPlatform for tracing, evaluating and simulation-testing LLM apps and AI agents.★ 4.9k Apache-2.0 TypeScript self-host managedNot scored yet
- LaminarObservability platform for AI agents with tracing, evaluations and analytics, self-hosted or cloud.★ 3.3k Apache-2.0 TypeScript self-host managedNot scored yet
- PezzoLLMOps platform for prompt management, versioning, delivery and request observability.★ 3.3k Apache-2.0 TypeScript self-hostNot scored yet
- OpenLITOpenTelemetry-native platform for tracing, evaluating and guarding LLM apps and agents, including GPU monitoring.★ 2.8k Apache-2.0 TypeScript self-hostNot scored yet
- Future AGISelf-hostable platform for tracing, evaluating, simulating and guarding LLM and agent applications.★ 2.1k Apache-2.0 Python self-host managedNot scored yet
No alternatives match those filters.
How the ranking works
Health blends commit recency, release cadence, issue throughput, contributor breadth and star growth. Stars are shown but never ranked on. The full method.