Langfuse LLM Observability
Langfuse 4.56.0 on Ubuntu 24.04 LTS, running the full single-node stack in your own AWS account. Langfuse is an open source platform for tracing, evaluating and debugging LLM applications. Your trace data stays on an EBS volume you control, every service listens on the instance itself, and you open the web UI through AWS Systems Manager port forwarding. This page covers what is on the image, the defaults it ships with and how billing works.
- AMI release
- 0.2.4 (Langfuse 4.56.0)
- Recommended
- t3.large
- Storage
- 40 GiB+ encrypted gp3
- Access
- Session Manager, no inbound ports
- Released
- Oct 12, 2026
- Delivery
- Amazon Machine Image in your own AWS account
Trace, evaluate and debug LLM applications with Langfuse 4.56.0, ingesting through OpenTelemetry or the v4 Langfuse SDKs.
Trace data stays on your EBS volume, and each instance creates its own credentials and admin password on first boot.
The UI and data services bind to localhost; you reach the UI through Session Manager port forwarding.
Quick start
Full guide →What is on the image¶
- Langfuse 4.56.0 web and worker, with Postgres, Redis,
ClickHouse and MinIO, in Docker Compose with pinned images. The
Compose file is
/opt/langfuse/docker-compose.yml. - A systemd service,
langfuse, that starts the stack at boot. - First-boot setup that creates this instance's database and service passwords, encryption keys, an admin password and a first project with its API keys. Nothing secret is baked into the image.
- A walkthrough script,
/opt/sushi/docs/demo.sh, that checks the stack end to end, including sending a trace and reading it back.
Telemetry is off. Enterprise (ee/) features are not included.
Defaults¶
| Setting | Default |
|---|---|
| Web UI and API | http://127.0.0.1:3000 (loopback only) |
| Admin sign-in | admin@localhost.local, password in /etc/langfuse/admin.password |
| First project | Sushi Pilot, in the organization Sushi Pilot |
| Project API keys | /etc/langfuse/project-api.env, root only |
| Health | http://127.0.0.1:3000/api/public/health |
| Trace ingestion | OpenTelemetry (OTLP/HTTP) at /api/public/otel/v1/traces |
This is Langfuse's v4 data model. Traces arrive through OpenTelemetry or a current Langfuse SDK, and you read them with the Observations API v2. The legacy batch ingestion endpoint used by older SDKs does not store traces, see Send traces.
Ports¶
All ports are published on 127.0.0.1 only. No inbound security group rule
is needed.
| Port | Service |
|---|---|
| 3000 | Langfuse web UI and API |
| 3030 | Langfuse worker |
| 5432 | Postgres |
| 6379 | Redis |
| 8123, 9000 | ClickHouse HTTP and native protocol |
| 9090, 9091 | MinIO API and console |
Recommended instance¶
t3.large (8 GiB of memory) with an encrypted gp3 root volume of at least
40 GiB and IMDSv2. The stack needs at least 8 GiB of memory. Grow the volume
as your trace volume grows.
Known issue¶
The bundled ClickHouse is the stable 25.12 release. Its static binary embeds OpenSSL 3.5.0, which has a published CVE. ClickHouse listens only on loopback and is not reachable from the network. A new version of the image will follow once a stable ClickHouse release ships the fixed OpenSSL. Move to it when it is released.
Billing¶
You pay an hourly software fee while the instance runs, as shown in the offer terms on the listing, which also shows any free trial. AWS bills EC2, EBS and data transfer separately.
Where to go next¶
- Getting started: subscribe, launch, sign in and send your first trace.
- Configuration: files, settings and how to apply them.
- Operations: security model, service management, backups, updates and support.
- Troubleshooting: common errors and how to fix them.
Upstream project¶
Langfuse is published under the MIT license, except for the separately
licensed ee/ directories. See the
Langfuse documentation for tracing, SDKs,
evaluations and prompt management. Software Sushi packages the image; the
Langfuse project does not endorse it, and upstream support is not included.
Last updated October 13, 2026 · Suggest an improvement