Platform comparison
Last verified 2026-11-16 from each project’s docs or repository; the source for every cell is in evidence/comparison-table.md. “Not documented” means the docs checked do not say, not that the feature is missing. OTel GenAI conventions means the platform documents reading gen_ai.* span attributes on ingest.
This table scrolls sideways on a narrow screen.
| Platform | Licence | Self-host | OTLP ingest | OTel GenAI conventions | Evals built in |
|---|---|---|---|---|---|
| Langfuse | MIT, open-core (ee/) |
Yes | Yes | Yes | Yes |
| Arize Phoenix | Elastic License 2.0 | Yes | Yes | Yes | Yes |
| LangSmith | Closed; MIT SDKs | Enterprise add-on | Yes | Yes | Yes |
| Opik | Apache-2.0 | Yes | Yes | Yes | Yes |
| LangWatch | Apache-2.0, open-core (enterprise/) |
Yes | Yes | Not documented | Yes |
| Laminar | Apache-2.0 | Yes | Yes | Yes | Yes |
| Agenta | MIT, open-core (ee/) |
Yes | Yes | Not documented | Yes |
| Pydantic Logfire | Closed; MIT SDK | Enterprise plan | Yes | Yes | Yes |
| W&B Weave | Apache-2.0 | With a W&B licence | Yes | Yes | Yes |
| MLflow Tracing | Apache-2.0 | Yes | Yes | Yes | Yes |
| Braintrust | Closed | Data plane only (BYOC) | Yes | Yes | Yes |
| AgentOps | MIT SDK; app Elastic License 2.0 | Yes | Not documented | Not documented | Not documented |
| Noveum Trace | Apache-2.0 SDK; hosted backend | Not documented | Not documented | Not documented | Not documented |
| Agnost AI | Closed; MIT SDKs | Not documented | Yes | Yes | Yes |
Sources
Last verified 2026-11-16. Each cell of the table in “Tracing and observability platforms” comes from the source below. All pages returned HTTP 200 on 2026-11-16. Licences come from the GitHub API (license.spdx_id) and the repository’s LICENSE file. Where a repo shows NOASSERTION, the README or LICENSE text was read. “Not documented” means the pages listed for that platform, plus its docs index (llms.txt) or README, did not say. It doesn’t mean the feature is absent.
Column definitions:
- Self-host: the vendor documents running the backend on your own infrastructure.
- OTLP ingest: the backend documents an OTLP endpoint that accepts spans from any OpenTelemetry SDK.
- OTel GenAI conventions: the docs say the backend reads
gen_ai.*attributes from incoming spans. - Evals built in: the product runs evaluations or scorers itself.
This table scrolls sideways on a narrow screen.
| Platform | Licence | Self-host | OTLP ingest | OTel GenAI conventions | Evals built in |
|---|---|---|---|---|---|
| Langfuse | MIT outside ee/ (repo LICENSE) |
https://langfuse.com/self-hosting (“can be self-hosted using Docker”; some add-ons need a license key) | https://langfuse.com/integrations/native/opentelemetry (/api/public/otel) |
same page (“Langfuse maps the received OTel traces”, gen_ai.*); ingest code packages/shared/src/server/otel/OtelIngestionProcessor.ts |
https://langfuse.com/docs/evaluation/overview |
| Arize Phoenix | Elastic License 2.0 (repo LICENSE) | https://arize.com/docs/phoenix/self-hosting | https://arize.com/docs/phoenix/tracing/concepts-tracing/otel-openinference/exporter | https://arize.com/docs/phoenix/integrations/llm-providers/google-gen-ai/gemini-go-sdk (“Phoenix ingests gen_ai.* attributes”) | https://arize.com/docs/phoenix/get-started/get-started-evaluations |
| LangSmith | Closed backend; langsmith-python and langsmith-javascript MIT |
https://docs.langchain.com/langsmith/self-hosted (“an add-on to the Enterprise plan”) | https://docs.langchain.com/langsmith/trace-with-opentelemetry (<LANGSMITH_ENDPOINT>/otel) |
same page, “Supported OpenTelemetry attribute and event mapping” (gen_ai.operation.name, gen_ai.usage.*) |
https://docs.langchain.com/langsmith/evaluation |
| Opik | Apache-2.0 | https://www.comet.com/docs/opik/self-host/overview | https://www.comet.com/docs/opik/integrations/opentelemetry (/api/v1/private/otel) |
same page, mapping table from OpenTelemetry GenAI semantic conventions (gen_ai.prompt, gen_ai.request.*, …) |
https://www.comet.com/docs/opik/evaluation/overview |
| LangWatch | Apache-2.0 outside enterprise/ |
https://langwatch.ai/docs/self-hosting/overview | https://langwatch.ai/docs/observability/trace-vs-activity-ingestion (/api/otel/v1/traces) |
Not documented: the semantic-conventions page says it “follows the OpenTelemetry semantic conventions” but lists only langwatch.* attributes, so gen_ai.* ingest isn’t stated |
https://langwatch.ai/docs/evaluations/overview |
| Laminar | Apache-2.0 | https://laminar.sh/docs/self-hosting/overview (OSS images; Signals and Slack not included) | https://laminar.sh/docs/tracing/otel (OTLP/gRPC, HTTP/proto, HTTP/json, “cloud and self-hosted”) | same page (“Fully compatible with the gen_ai semantic conventions”) |
https://laminar.sh/docs/agent-evaluations/introduction |
| Agenta | MIT outside ee/ (repo LICENSE text) |
https://agenta.ai/docs/self-host/quick-start | https://agenta.ai/docs/1.0/observability/quick-start-opentelemetry (/api/otlp/v1/traces) |
Not documented: https://agenta.ai/docs/1.0/observability/concepts lists OpenInference, OpenLLMetry, and PydanticAI conventions, not OTel GenAI | https://agenta.ai/docs/evaluation/overview/ |
| Pydantic Logfire | SDK MIT; backend closed | https://pydantic.dev/docs/logfire/get-started/faq/ (“Logfire Enterprise: Self-hosted”) | https://pydantic.dev/docs/logfire/guides/alternative-clients/ (/v1/traces) |
https://pydantic.dev/docs/logfire/instrument/typescript/frameworks/vercel-ai/ (GenAI semantic-convention spans are “what Logfire renders and prices best”) | https://pydantic.dev/docs/logfire/instrument/typescript/evals/ |
| W&B Weave | Apache-2.0 | https://docs.coreweave.com/products/wandb/weave/guides/platform/weave-self-managed (W&B Self-Managed with a “Weave-enabled license”) | https://docs.coreweave.com/products/wandb/weave/guides/tracking/otel (/otel/v1/traces) |
same page (“Weave parses span attributes according to the OpenTelemetry semantic conventions”, gen_ai prefix) |
https://docs.coreweave.com/products/wandb/weave/guides/core-types/evaluations |
| MLflow Tracing | Apache-2.0 (mlflow/mlflow) |
https://mlflow.org/docs/latest/self-hosting/ | https://mlflow.org/docs/latest/genai/tracing/app-instrumentation/opentelemetry (“OTLP endpoint at /v1/traces”) |
https://mlflow.org/docs/latest/genai/tracing/opentelemetry/genai-semconv (“Ingesting GenAI Semconv Traces to MLflow”: the /v1/traces endpoint “automatically recognizes traces that use GenAI Semantic Convention attributes”); exporting as gen_ai.* is opt-in via MLFLOW_ENABLE_OTEL_GENAI_SEMCONV |
https://mlflow.org/docs/latest/genai/eval-monitor/ |
| Braintrust | Closed (Autoevals MIT) | https://docs.braintrust.dev/admin/self-hosting (you run the data plane; Braintrust provides the managed UI and platform). That page now calls the self-hosted data plane legacy and points new organizations to https://docs.braintrust.dev/admin/deployment/byoc (Bring your own cloud) | https://docs.braintrust.dev/integrations/sdk-integrations/opentelemetry/send-traces-and-logs (/otel/v1/traces) |
https://docs.braintrust.dev/integrations/sdk-integrations/opentelemetry/attributes (“Braintrust implements the OpenTelemetry GenAI semantic conventions”) | https://docs.braintrust.dev/evaluate |
| AgentOps | SDK MIT; app Elastic License 2.0 (list entry, unchanged) | https://docs.agentops.ai/v2/self-hosting/overview | Not documented in https://docs.agentops.ai/llms.txt | Not documented | Not documented |
| Noveum Trace | SDK Apache-2.0 (README badge and LICENSE); backend hosted | Not documented in the SDK README | Not documented: the README describes sending to the Noveum API or a custom endpoint, not OTLP | Not documented | Not documented in the SDK README |
| Agnost AI | Python and TypeScript SDKs MIT; backend closed (list entry, unchanged) | Not documented in https://docs.agnost.ai/llms.txt | https://docs.agnost.ai/otel (“hosted OTLP collector at otel.agnost.ai”) |
same page (“GenAI convention attributes”: gen_ai.request.model, gen_ai.usage.*) |
https://docs.agnost.ai/online-evals |
No platform was run for the table. Every cell comes from the docs or source listed above.