Langfuse launches AI Gateway Learn more

Production checklist for self-hosted Langfuse

A Docker Compose trial of Langfuse runs in minutes. A production deployment needs a few more decisions: which managed services to use, how people sign in, how data is encrypted and backed up, and how you will notice problems. This checklist collects them in one place, with links to the configuration reference for each item.

TL;DR: Deploy with the Helm chart or a cloud Terraform module, not Docker Compose. Use managed Postgres, Redis, and S3, and ClickHouse Cloud, BYOC, or the ClickHouse Kubernetes Operator. Serve Langfuse over HTTPS, set ENCRYPTION_KEY, disable open sign-up, and configure SSO. Back up Postgres and ClickHouse, monitor health endpoints, set a retention policy, and plan a regular upgrade cadence.

Deployment architecture
UI, API, and SDK clients connect to Langfuse Web. Web connects to PostgreSQL, Redis or Valkey, ClickHouse, and S3 or blob storage. Redis queues jobs for Langfuse Worker, which connects to PostgreSQL, ClickHouse, and S3. Both containers can optionally connect to LLM providers: Web for the playground, Worker for evaluations. The optional Langfuse AI gateway receives model requests from applications and coding agents, resolves gateway keys and uploads traces through Langfuse Web, and relays the requests to the LLM providers. The application containers, gateway, and storage run in your VPC or on-premises. The LLM providers can also run in the same VPC or a peered VPC.VPC / on-premisesUI, API & SDKsApps & agentsModel requestsResolve + tracesRequestsPlaygroundEvaluationsRelayAI GatewayOptional (beta)langfuse/ai-gatewayLangfuse WebUI & API serverlangfuse/langfuseLangfuse WorkerAsync event processinglangfuse/workerPostgreSQLTransactional dataRedis / ValkeyCache & job queueClickHouseObservability dataS3 / Blob storageEvents & attachmentsLLM providersOptional (bring your own)
Web connectionsWorker connectionsOptional

The optional AI gateway relays model requests from your apps and coding agents to your LLM providers and records them in Langfuse. LLM providers can also run in the same VPC or a peered VPC.

Infrastructure

  • Deployment method: Kubernetes with Helm, or the AWS, Azure, or GCP Terraform modules. Docker Compose has no high availability, scaling, or backups.
  • Postgres: a managed service with automated backups. See Postgres.
  • ClickHouse: ClickHouse Cloud, BYOC, or the ClickHouse Kubernetes Operator, on a supported version. See ClickHouse for self-hosted Langfuse.
  • Redis or Valkey: Redis 7 or later, with maxmemory-policy=noeviction so queue jobs aren't evicted. See cache.
  • Blob storage: S3, GCS, Azure Blob Storage, or a compatible service. See blob storage.
  • UTC everywhere: Postgres and ClickHouse must run in UTC.

Sizing

  • At least the minimums: web and worker 2 CPU and 4 GiB each, Postgres 2 CPU and 4 GiB, Redis 1 CPU and 1.5 GiB, ClickHouse 2 CPU and 8 GiB. See scaling.
  • More than one replica of web and worker, so one restart doesn't stop ingestion or the UI.
  • A load estimate from a typical day of traces, counted in observations, not requests.

Security

  • HTTPS terminated at the load balancer or Ingress. See encryption.
  • ENCRYPTION_KEY set and stored in your secrets manager. It encrypts stored LLM connections and gateway provider connections, and changing it makes them unreadable.
  • Encryption at rest on the databases and blob storage. See encryption at rest.
  • Sign-in: disable open sign-up, configure SSO, and enforce SSO for your email domains. See hardening and authentication and SSO.
  • Network exposure: keep databases private, and expose only the web container and, if you use it, the AI Gateway. See networking.
  • Outbound URL allowlists for webhooks and LLM connections, to limit server-side requests. See outbound URL allowlists.
  • PII: mask data in the SDKs before it is sent, or with server-side masking on Enterprise. See data masking.

Reliability

  • Backups for Postgres and ClickHouse, and a tested restore. See backups.
  • Health checks on the readiness and liveness endpoints. See health and readiness endpoints.
  • Monitoring of the Langfuse containers with OpenTelemetry. See observability.
  • Disk alerts on ClickHouse and blob storage, and ClickHouse system log tables capped. See ClickHouse system log tables.
  • Data retention per project, so storage doesn't grow forever. See data retention (Enterprise on self-hosted).

Operations

  • Upgrade cadence: follow the upgrade guide, and read the release notes before major versions. Subscribe to the OSS newsletter for breaking changes.
  • Telemetry decision: OSS sends usage telemetry unless TELEMETRY_ENABLED=false. Enterprise sends license telemetry, Government sends none. See telemetry.
  • Headless setup: create the first organization, project, and API keys from environment variables for repeatable deployments. See headless initialization.
  • One deployment or several: decide whether teams share one instance or get their own. See deployment strategies.
  • Optional features: LLM connections for the Playground and evaluations, Halo, and the AI Gateway.

Langfuse does not require internet access. Without it, the update check fails gracefully. Features that call a model, such as LLM-as-a-Judge and the Playground, need a reachable LLM API. See air-gapped LLM observability.

FAQ

What is the smallest production-ready setup?

Two replicas each of the web and worker containers on Kubernetes, managed Postgres, managed Redis, S3, and a small ClickHouse Cloud service or an operator-managed ClickHouse with at least 2 CPUs and 8 GiB of memory.

Do I need an Enterprise license for production?

No. Open source Langfuse runs in production without limits, including enterprise SSO and organization-level roles. An Enterprise license adds features such as audit logs, data retention, project-level roles, server-side masking, and support. See the license key page.


Was this page helpful?