Open Langfuse -> Settings -> API Keys and create a new key pair for the project you want to sync.
Copy both the public key (pk-lf-...) and the secret key (sk-lf-...). The secret is shown once.
Set host to your instance base URL - https://cloud.langfuse.com (or the US/EU regional variants) for Langfuse Cloud, or your self-hosted origin (no trailing slash).
Store the secret as a secret and reference it from config as secretKey: secret("LANGFUSE_SECRET_KEY"), alongside the plaintext publicKey.

Configuration

Field	Type	Required	Description
`publicKey`	string	Yes	Langfuse public API key for the project (starts with `pk-lf-`). Created in Langfuse -> Settings -> API Keys.
`secretKey`	secret	Yes	Langfuse secret API key for the project (starts with `sk-lf-`). Issued alongside the public key in Langfuse -> Settings -> API Keys.
`host`	string	No	Langfuse instance base URL. Use https://cloud.langfuse.com (US) or https://us.cloud.langfuse.com / https://eu.cloud.langfuse.com for Langfuse Cloud, or your self-hosted origin. No trailing slash.
`lookbackDays`	number	No	How many calendar days of history to backfill on a full sync. Defaults to 30.
`resources`	array	No	Which Langfuse resources to sync. Omit to sync all of them.

Resources

langfuse_trace (entity) - LLM traces in the project, keyed by id, with name, owning user/session, optional release/version, aggregate cost in USD, aggregate latency in milliseconds, and the createdAt timestamp.
- Endpoint: GET /api/public/traces
- Traces upsert by id on every run. Trace input/output payloads are not stored.
- name: Trace name set by the SDK.
- projectId: Langfuse project id the trace belongs to.
- userId: Attached userId, if any.
- sessionId: Attached sessionId, if any.
- release: Release identifier from the SDK, if set.
- version: Version identifier from the SDK, if set.
- totalCost: Aggregate trace cost in USD across all observations.
- latencyMs: End-to-end trace latency in milliseconds.
- createdAt: ISO timestamp of trace creation.
langfuse_observations_per_day (metric) - Daily LLM observation volume, total tokens, and total cost rolled up by model from the Langfuse daily metrics endpoint. One sample per (day, model) over the lookback window.
- Endpoint: GET /api/public/metrics/daily
- Unit: observations
- Granularity: Daily (UTC)
- Dimensions: model
- Measures: inputTokens, outputTokens, totalTokens, costUsd
- Rollup metrics are stamped at UTC midnight of the day they cover.
langfuse_scores (metric) - Daily Langfuse score rollups by score name. One sample per (day, name): the mean numeric value across that day and the count of scores written.
- Endpoint: GET /api/public/scores
- Unit: scores
- Granularity: Daily (UTC)
- Dimensions: scoreName
- Measures: count
- Only numeric scores contribute to the average; non-numeric scores still increment the count.

Example

import {
  defineConfig,
  defineDashboard,
  defineMetric,
  secret,
} from '@rawdash/core';

const langfuse = {
  name: 'langfuse',
  connectorId: 'langfuse',
  config: {
    publicKey: 'pk-lf-...',
    secretKey: secret('LANGFUSE_SECRET_KEY'),
    host: 'https://cloud.langfuse.com',
  },
};

export default defineConfig({
  connectors: [langfuse],
  dashboards: {
    llm: defineDashboard({
      widgets: {
        daily_observations: {
          kind: 'timeseries',
          title: 'LLM observations per day',
          window: '30d',
          metric: defineMetric({
            connector: langfuse,
            shape: 'metric',
            name: 'langfuse_observations_per_day',
            fn: 'sum',
          }),
        },
      },
    }),
  },
});

Rate limits

Langfuse Cloud applies per-project rate limits (around 1000 requests/min on paid plans); 429 responses with Retry-After are honored.

Limitations

One key pair scopes the sync to a single Langfuse project; sync multiple projects by adding one connector instance per project.
Trace bodies (input/output payloads) are not synced - only the trace envelope plus aggregated cost / token / latency.
Session and dataset endpoints are out of scope for the initial release.

License

Apache-2.0