Langfuse
for LuumenAI
Read traces, scores and prompts from Langfuse, then record your own scores
Langfuse holds the traces, scores and prompts behind your LLM application. Connect it with an API key and Luumen can read that history directly in the terminal: which generations ran slow, how a prompt version is scored, what an experiment produced. When you want to record a judgment of your own, Luumen attaches a score to the trace it just showed you.
The Langfuse toolbox
10 tools: 9 read, 1 write. Reads answer instantly. Writes require approval by default. Everything is logged.
- ReadGet PromptRetrieve a Langfuse text or chat prompt by name, deployment label, or exact version, with optional dependency resolution.
- ReadList Annotation QueuesList annotation queues configured for human evaluation in the connected Langfuse project.
- ReadList Dataset ItemsList dataset items, optionally scoped to a dataset, source trace or observation, or historical dataset version.
- ReadList DatasetsList datasets available to the connected Langfuse project.
- ReadList ExperimentsList Langfuse experiments active in a required time range, optionally including metadata and directly attached scores.
- ReadList ModelsList Langfuse-managed and project-custom model pricing definitions used for usage and cost calculation.
- ReadList ObservationsSearch Langfuse observations such as generations, spans, events, agents, and tool calls, selecting only the field groups needed.
- ReadList ScoresSearch numeric, boolean, categorical, text, and correction scores with subject and annotation context.
- ReadQuery MetricsRun an aggregate Langfuse metrics query over observations or numeric, boolean, or categorical scores.
- WriteCreate ScoreAttach an evaluation score to exactly one Langfuse trace, session, or dataset run, optionally narrowing a trace score to an observation. Approval by default
One prompt, start to finish
What a governed Langfuse run looks like inside Luumen.
Questions
Authorize once with API token. Luumen lists the scopes each action needs before you approve the connection, and credentials never appear in the chat.
Read actions answer immediately. Anything that writes — create score — is shown as a plan and requires approval by default. Administrators configure that per tool, so you decide exactly which actions can ever run unattended.
You decide. Actions are granted per agent, skill, and team, and per environment — production is not staging. Read access can be broad while writes stay narrow.
Every call to Langfuse — read or write, approved or declined — is recorded with the actor, the input, and the result, and can be linked to the ticket or change record.
Often connected alongside
All integrations
GitHubRepos, actions, and runbook code as context and as actions.
GitLabMerge requests and pipelines proposed, triggered, and rolled back.
BitbucketRead commits, diffs, and branches in Bitbucket, then open the fix as a held write.
Azure DevOpsAzure DevOps repos and pipelines as context — and as approved actions.Put Langfuse to work with Luumen
Connect in minutes. Every action scoped, approved, and audited from day one.