Hermes Agent Integration — Dunetrace docs

Quick Start

pip install dunetrace hermes-agent

from run_agent import AIAgent
from dunetrace import Dunetrace
from dunetrace.integrations.hermes import DunetraceHermesPlugin

dt = Dunetrace()   # local dev, no API key needed
plugin = DunetraceHermesPlugin(dt, agent_id="my-agent", model="hermes-3-llama-3.1-70b")
plugin.attach()   # registers hooks with the Hermes global PluginManager, once per process

agent = AIAgent(api_key=os.getenv("OPENAI_API_KEY"), model="gpt-4o-mini", quiet_mode=True)
result = agent.run_conversation("What is the capital of France?")

dt.flush()

What this does

Dunetrace hooks into Hermes Agent's plugin system — every LLM call, tool call, and session boundary is captured with no changes to your agent code. Each run_conversation() call becomes one Dunetrace run (Hermes's turn_id is used as the run_id).

Verification

SCENARIO=tool_loop PYTHONPATH=packages/sdk-py python packages/sdk-py/examples/hermes_agent.py

Check the dashboard at http://localhost:3000 — the run and the TOOL_LOOP signal should appear within ~15 seconds. Other scenarios: happy, retry, abandon, edge, real, real_loop, all (default).

Advanced (optional)

Persistent plugin (CLI users)

Install the plugin once so it activates for every hermes CLI session — see ~/.hermes/plugins/ in the Hermes docs for the plugin registration format. Configure via env vars: DUNETRACE_API_URL, DUNETRACE_API_KEY, DUNETRACE_AGENT_ID.

Data handling

User message, tool arguments, and error messages are sent to the backend as-is (input_text, args, error); token counts and latency as plain numbers.