DEDUPED Hot AI repos, turned into use cases Get the Monday 5

How to trace every Hermes turn, tool call and cost in Langfuse with Hermes Official Plugins

Install the Langfuse plugin to send each turn, model call and tool call to Langfuse for review.

Short answer: Run hermes plugins info langfuse to see what it needs and which commit it pins; then run hermes plugins install langfuse; set up your Langfuse Cloud or self-hosted project and add the credentials the plugin asks for. All steps.

Credentials. Langfuse keys let the plugin write trace data to your project, so store them with the same care as other API keys.

Who it is forAn engineer running Hermes in production who needs to see what the agent did, what each model call cost, and where time was spent.

The problemWithout traces it is hard to tell which model call or tool call caused a bad answer or an unexpected bill.

The outcomeEach Hermes turn appears in Langfuse as a span, with generations for model calls and observations for tool and subagent calls.

What you need first

  • Hermes Agent installed (the README gives a curl install script in the hermes-agent repo)
  • The hermes CLI available in your terminal
  • Accounts and credentials for the specific service you connect, such as Home Assistant, Spotify, Langfuse, Snyk or X
  • Each plugin's own requirements, shown by hermes plugins info <name>

Steps

  1. Run hermes plugins info langfuse to see what it needs and which commit it pins. (inferred, check the README)
  2. Run hermes plugins install langfuse.
  3. Set up your Langfuse Cloud or self-hosted project and add the credentials the plugin asks for.
  4. Choose the content capture mode; the README says capture is sanitized by default, with metadata-only and full modes.
  5. Run a Hermes task and open the Langfuse project to check the new trace.

Prompt for your agent

In the repo NousResearch/hermes-official-plugins, install the langfuse plugin, then run one Hermes task that uses a tool. Open the Langfuse project <project-name> and report the span, generation and tool observation for that run. Verify that the reported usage and cost fields are present on the generation.

Caveat. Full content mode can send prompts and outputs to Langfuse, so pick metadata-only mode if your turns contain private data.

Use this in your AI agent

Ask your agent

Using the Deduped MCP, get the setup steps for NousResearch/hermes-official-plugins and "Trace every Hermes turn, tool call and cost in Langfuse"

Add Deduped to Claude Code once

claude mcp add --transport http deduped https://dedupedai.com/mcp

Cursor, ChatGPT, others

Share: X LinkedIn Hacker News Email Copy link

Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer