textlambda/hermes-agent-reasoning-tracesagentstool-callingreasoningsftfunction-callinghermessharegptfine-tuningkimiglm

Hermes Agent Reasoning Traces

Name: Hermes Agent Reasoning Traces
License: https://huggingface.co/datasets/lambda/hermes-agent-reasoning-traces

Free

Open dataset

Sample structure: 100 / 100

3 download links issued

Seller: DataBazaar

Already have an account? Log in

Agent? Connect your account →

Source, license and coverage

Supplier documentation. These claims are separate from the automated sample score. A listing edit date is not a data freshness date.

License: apache-2.0
Source / creator: lambda/hermes-agent-reasoning-traces
Collection method: Traces were generated by running the Hermes Agent harness against two frontier open-weights models (Moonshot Kimi-K2.5 and Zhipu GLM-5.1-FP8). Each conversation is a genuine end-to-end agent run: the model produces reasoning inside `<think>` tags, emits tool calls, and the harness executes those tools and feeds real results back into the context. The publisher (Lambda) stored these as ShareGPT-formatted parquet shards, partitioned by source model.
Coverage start: Not documented
Coverage end: Not documented
Data last updated: Not documented
Update schedule: Not documented

Source documentation ↗

License terms ↗

Rows are evenly spaced within the inspected input, subject to the preview byte budget; this is not a new random sample of the full dataset. The dataset card does not detail filtering, deduplication, or tool-environment coverage; buyers should validate empirically. Only two source models are represented, so reasoning-style and tool-use diversity is bounded by Kimi-K2.5 and GLM-5.1 behaviors. English-only. Tool results were captured at generation time and may reference time-sensitive content (web pages, APIs) whose ground truth has since changed. Source does not document quality-grading or safety filtering of trajectories.

Sample structure score: 100 / 100

This automated check describes the inspected sample, not factual accuracy, legal rights, representativeness, or the quality of the entire dataset. It is not a customer rating.

Assessed 5 sample records (JSON) on 2026-10-09. All records in the provided sample were checked.

Check	Points	Evidence
Populated cells	50 / 50	30 of 30 top-level cells contain a value. Null, absent and blank values count as missing; zero and false count as populated.
Consistent value types	30 / 30	30 of 30 populated cells match their column's most common observed type. Types are inferred, not checked against real-world truth.
Consistent record shape	20 / 20	5 of 5 records have the expected fields. CSV/TSV use the header width; JSON uses the union of observed keys.

Field-level findings and improvements

Check missing cells and mixed types below. Document intentional missing values or mixed types in your field descriptions. Do not fill legitimate unknowns with invented values just to increase this score.

Field	Missing cells	Most common type	Other populated types
id	0 / 5	string	0 / 5
conversations	0 / 5	object	0 / 5
tools	0 / 5	string	0 / 5
category	0 / 5	string	0 / 5
subcategory	0 / 5	string	0 / 5
task	0 / 5	string	0 / 5

How the score is calculated, its limitations, and how to correct an assessment →

About this data

Multi-turn agent tool-calling trajectories with step-by-step reasoning traces and real tool execution results, generated from Kimi-K2.5 and GLM-5.1 models via the Hermes Agent harness.

Retrieve with your agent or Python

Create an account and configure DATABAZAAR_API_KEY. This example retrieves free or already purchased data; it never makes a purchase. For a multi-file dataset, choose a file index from the manifest.

Download the Python example

python3 retrieve-dataset.py 026d7820-e865-4a26-ad18-91a791b9e594 --output dataset.bin

Full supplier documentation

## Overview Hermes Agent Reasoning Traces is a corpus of 14,701 multi-turn AI agent conversations captured from the Hermes Agent harness. Each sample is a real agent trajectory containing step-by-step `<think>` reasoning blocks, tool/function calls, and the actual results returned from tool execution. The dataset is published in Parquet across two configs split by the source model that generated the trace: `kimi` (Moonshot AI Kimi-K2.5, 7,646 samples) and `glm-5.1` (ZhipuAI GLM-5.1-FP8, 7,055 samples). English-language; updated April 2026. ## Schema - conversation / messages — list[struct] — ShareGPT-style turns including system, user, assistant, and tool roles - think blocks — text — embedded `<think>...</think>` reasoning segments inside assistant turns - tool_calls — struct/list — function-calling invocations with names and arguments - tool results — text — actual outputs returned from tool execution - model — string — source model identifier (Kimi-K2.5 or GLM-5.1-FP8) via config (Exact column names follow the ShareGPT/Hermes convention; see the dataset card for the canonical field list.) ## Sources - lambda/hermes-agent-reasoning-traces on Hugging Face — https://huggingface.co/datasets/lambda/hermes-agent-reasoning-traces — Apache-2.0 ## Methodology Traces were generated by running the Hermes Agent harness against two frontier open-weights models (Moonshot Kimi-K2.5 and Zhipu GLM-5.1-FP8). Each conversation is a genuine end-to-end agent run: the model produces reasoning inside `<think>` tags, emits tool calls, and the harness executes those tools and feeds real results back into the context. The publisher (Lambda) stored these as ShareGPT-formatted parquet shards, partitioned by source model. ## Known gaps & limitations The dataset card does not detail filtering, deduplication, or tool-environment coverage; buyers should validate empirically. Only two source models are represented, so reasoning-style and tool-use diversity is bounded by Kimi-K2.5 and GLM-5.1 behaviors. English-only. Tool results were captured at generation time and may reference time-sensitive content (web pages, APIs) whose ground truth has since changed. Source does not document quality-grading or safety filtering of trajectories. ## Intended use & out-of-scope - IS for: SFT and distillation of agentic / tool-calling models, reasoning-trace fine-tuning, building evals over multi-step tool use, studying `<think>`-style chain-of-thought patterns, RAG over real agent transcripts. - NOT for: benchmark training without leakage checks (no dedup against common agent evals like SWE-bench, GAIA, τ-bench is documented); not a substitute for human-graded preference data; not suitable as ground-truth for tool API behavior since captured results may be stale. _Federated dataset: 2 parquet shards, 1.03 GB total. Queries and downloads stream through the DataBazaar API._ _PII signals: email×65, us_phone×7442 present in the sample. Common in public datasets (papers, logs) but worth knowing before joining with private data._ Original supplier listing: Hermes Agent Reasoning Traces (Kimi-K2.5 & GLM-5.1) 14,700+ multi-turn agent tool-calling trajectories with step-by-step reasoning traces and real tool execution results, generated via the Hermes Agent harness from Kimi-K2.5 and GLM-5.1 models. Apache-2.0.

Schema

Name	Type	Description
id	VARCHAR
conversations	STRUCT("from" VARCHAR, "value" VARCHAR)[]
tools	VARCHAR
category	VARCHAR
subcategory	VARCHAR
task	VARCHAR

Sample Data

Preview a sample of the data before downloading.

Public sample only. Sign in to retrieve the full dataset, including free datasets.

For AI Agents

Via MCP Server

# 1. Add to your agent's MCP config (claude_desktop_config.json or similar):
{
  "mcpServers": {
    "databazaar": { "command": "npx", "args": ["databazaar-mcp"] }
  }
}

# 2. Your agent can then call:
search_datasets({ query: "Hermes Agent Reasoning Traces" })
// Found: 026d7820-e865-4a26-ad18-91a791b9e594
get_download_url({ dataset_id: "026d7820-e865-4a26-ad18-91a791b9e594" })  // free — sign in with MCP OAuth first

Via REST API

# Free dataset — sign in or use your account API key:
curl https://api.databazaar.io/datasets/026d7820-e865-4a26-ad18-91a791b9e594/download-url -H "Authorization: Bearer $DATABAZAAR_API_KEY"