Deliver OpenClaw observability data to Agent-lens and All-in-One Observability

Observability capabilities

openclaw-exporter-to-langfuse provides two complementary capabilities:

Capability	Data pipeline	Delivery component	Primary use case	Viewing tool
Deliver to Agent-lens	OpenClaw → openclaw-exporter-to-langfuse → Langfuse	openclaw-exporter-to-langfuse	Analyze individual Agent call chains and LLM interaction results	Langfuse
Deliver to All-in-One Observability	OpenClaw → diagnostics-otel → otelcol-contrib → ClickHouse	otelcol-contrib	Analyze OpenClaw runtime metrics, Gateway logs, and overall health status	HyperDX

Open source repository: https://github.com/aliyun/openclaw-exporter-to-langfuse

Limits

openclaw-exporter-to-langfuse requires Node.js 18 or later.
All-in-One Observability delivery requires the diagnostics-otel plugin. Use OpenClaw v2026.2.19 or later.
otelcol-contrib uses ports 4317 and 4318 by default. Ensure these ports are available.

Prerequisites

An ApsaraDB for ClickHouse cluster is created with the following information:
- VPC endpoint
- Username (for example, openclaw) with read and write permissions
- Password
Unified Observability is enabled for the ClickHouse cluster in the console.
The Agent Observability Suite (Agent-lens) is enabled for the ClickHouse cluster, and a project is created with the following information:
- VPC endpoint
- Public Key (in the format of pk-lf-xxx)
- Secret Key (in the format of sk-lf-yyy)
OpenClaw is installed and running.

Considerations

By default, traces go to Agent-lens. Metrics and logs go to All-in-One Observability.
To send traces to All-in-One Observability, add --diag-traces true during installation.
Set the ClickHouse cluster time zone to UTC. A non-UTC time zone causes data display anomalies in Agent-lens.

Procedure

Install through OpenClaw

Use the OpenClaw conversational interface to install. OpenClaw auto-detects the local configuration path, installation directory, and runtime environment.

Download the installation instructions

Enter the following in the OpenClaw chat:

Download the installation file (overwrite if it already exists): https://ck-langfuse-public.oss-cn-beijing.aliyuncs.com/openclaw-exporter-to-langfuse/INSTALLATION.md, and follow the prompts to install.

OpenClaw returns the installation parameters.

Configure and install

Enter the following in the OpenClaw chat, replacing placeholders with your values:

Install both openclaw-exporter-to-langfuse and otelcol-contrib.

openclaw-exporter-to-langfuse configuration:
endpoint = http://<LANGFUSE_ENDPOINT>:3000/api/public/otel/v1/traces
pk = <PK>
sk = <SK>
serviceName = <SERVICE_NAME>
debug = true

otelcol-contrib configuration:
ck-endpoint = http://<CK_ENDPOINT>:8123
username = <USERNAME>
password = <PASSWORD>

Parameters:

Parameter	Description
`endpoint`	Langfuse trace endpoint. Replace `<LANGFUSE_ENDPOINT>` with the Agent-lens VPC endpoint.
`pk`	Public key of the Agent-lens project.
`sk`	Secret key of the Agent-lens project.
`serviceName`	Name of the Agent service. Identifies the data source.
`debug`	Enables debug logging. Set to `true` during initial setup or troubleshooting.
`ck-endpoint`	ClickHouse cluster endpoint. Replace `<CK_ENDPOINT>` with the cluster VPC endpoint.
`username`	Username for the ClickHouse cluster.
`password`	Password for the ClickHouse cluster.

When OpenClaw returns a success message, both openclaw-exporter-to-langfuse and otelcol-contrib are configured.

Verify the installation

Enter the following in the OpenClaw chat:
```
View the configuration of both components
```
The expected output includes:
- openclaw-exporter-to-langfuse (Langfuse plugin): enabled is true, with the configured endpoint, authorization, and serviceName values.
- diagnostics-otel + otelcol-contrib (All-in-One Observability): diagnostics-otel is enabled with its endpoint, and otelcol-contrib is running with gRPC/HTTP receiver ports and ClickHouse endpoint.
Both components showing as enabled with correct parameters confirms a successful installation.

Verify data synchronization

View traces in Agent-lens

Log on to the Agent-lens console.
Open the target project.
Open the Traces page.
Start an Agent conversation in OpenClaw.
Verify that a new trace record appears.

New trace records appear in the Traces list. Click a trace to view the call chain tree (nodes such as enter_openclaw_system, invoke_agent_main, and tool_call), with latency, token consumption, input/output, and LLM interaction details per node.

Verify data in ClickHouse

Run the following SQL statements in ClickHouse to verify data ingestion:

SHOW DATABASES;

SHOW TABLES FROM clickobserve_service;

SELECT count() FROM clickobserve_service.otel_logs;

SELECT count() FROM clickobserve_service.otel_metrics_sum;

Analyze data in All-in-One Observability

Log on to the All-in-One Observability console.
Use AI Notebook to analyze data.

Example query:

Show OpenClaw error logs from the past 1 hour

AI Notebook interprets the query, generates SQL, and returns results with timestamps, sources, and log content.

FAQ

Traces are not visible in All-in-One Observability

By default, traces go only to Langfuse (Agent-lens). To also send them to All-in-One Observability, reinstall the plugin with:

--diag-traces true