datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
agent-llm-traces-v2
Exgentic Agent LLM Traces v2 — Agent Chat Only
OpenTelemetry-shaped execution traces for 10,057 agent runs across 6 benchmarks (AppWorld, SWE-bench, BrowseCompPlus, τ²-bench Airline/Retail/Telecom), filtered to the agent under test's chat-only LLM calls. This is the dataset for replay testing, behavioral analysis, or any task where you care about what the benchmarked model actually did — not the eval scaffolding around it.
This v2 release expands upon Exgentic/agent-llm-traces… See the full description on the dataset page: https://huggingface.co/datasets/Exgentic/agent-llm-traces-v2.agent-traces
Trace Commons — Agent Traces
Trace Commons is one open, public dataset of coding-agent sessions — the
back-and-forth between a developer and an AI coding agent, including prompts,
model responses, tool calls, and command output — contributed voluntarily as an
open resource for studying, evaluating, and building on how these agents
actually work.
Every trace here was donated only from a public, open-source repository, was
anonymized on the contributor's own machine before upload… See the full description on the dataset page: https://huggingface.co/datasets/trace-commons/agent-traces.hermes-agent-reasoning-traces
Hermes Agent Reasoning Traces
Multi-turn tool-calling trajectories for training AI agents using the Hermes Agent harness. Each sample is a real agent conversation with step-by-step reasoning (<think> blocks) and actual tool execution results.
This dataset has two configs, one per source model:
Config
Model
Samples
kimi
Moonshot AI Kimi-K2.5
7,646
glm-5.1
ZhipuAI GLM-5.1-FP8
7,055
Loading
from datasets import load_dataset
# Kimi-K2.5 traces
ds =… See the full description on the dataset page: https://huggingface.co/datasets/lambda/hermes-agent-reasoning-traces.agent-llm-traces
Multi-Benchmark LLM Agent Traces
A comprehensive dataset of OpenTelemetry traces capturing LLM inference behavior across multiple agent frameworks, benchmarks, and model providers. This dataset enables research into LLM performance analysis, agent behavior patterns, and inference optimization.
Collected by Exgentic - A platform for LLM observability and performance optimization.
Dataset Overview
This dataset contains 1,781 execution traces capturing detailed agent… See the full description on the dataset page: https://huggingface.co/datasets/Exgentic/agent-llm-traces.ninja-agent-traces
Tau retired-king tasks and rollouts
This dataset is written by the Tau validator when a challenger becomes king.
tasks contains one viewer-friendly row per generated task.
rollouts contains one viewer-friendly row per terminal qualification or duel
solve and is the default table shown on the dataset page.
events contains one flattened row per redacted proxy-observed LLM call.
payloads contains complete solution diffs plus request and response bodies split
into bounded, ordered… See the full description on the dataset page: https://huggingface.co/datasets/Wejh/ninja-agent-traces.hermes-agent-traces-filtered
Hermes Agent Reasoning Traces - Quality Filtered
A structurally filtered subset of lambda/hermes-agent-reasoning-traces, pruned from 7,646 to 3,679 rows using automated quality analysis targeting reasoning depth, structural integrity, and tool-call validity.
Why This Matters for Agent Training
Most agentic datasets teach models what tool to call but not how to reason about tool selection. The difference matters in production: an agent that dispatches tools without… See the full description on the dataset page: https://huggingface.co/datasets/DJLougen/hermes-agent-traces-filtered.mcphunt-agent-traces
MCPHunt Agent Traces
Agent execution traces from the MCPHunt evaluation framework, measuring
cross-boundary data propagation in multi-server MCP agents.
Contents
main/ — 3,615 traces from 5 models across 147 tasks and 7 environment
variants (risky_v1/v2/v3, benign, hard_neg_v1/v2/v3). One JSON file per model.
mitigation/ — 2,706 traces from the prompt-mitigation study (M0--M3
levels) across 3 models.
live_guard_defense/ — 387 DeepSeek-V4-Flash traces from the… See the full description on the dataset page: https://huggingface.co/datasets/lihaonan0716/mcphunt-agent-traces.swe-rebench-v2-glm-5.1-pi-agent-successful-traces
Agentic coding trajectories of GLM-5.1
We collected 7,777 successful multi-turn trajectories by running
GLM-5.1 with the
PI agent on
nebius/SWE-rebench-V2.
Configuration
Each SWE-rebench-V2 task is an issue of a public repository. An agent has to
implement a fix that is then validated by unit tests.
The model was run on 3,837 issues from Python repositories, with 4 rollouts
per issue. The task list is available as the train split of… See the full description on the dataset page: https://huggingface.co/datasets/whitecircle/swe-rebench-v2-glm-5.1-pi-agent-successful-traces.fable5-gpt5.5-opus4.7-mixed-agent-traces
Fable5 · GPT-5.5 · Opus-4.7 Mixed Agent Traces
A high-density post-training mixture for agentic reasoning, instruction following, code generation, function calling, and tool-use decision making.
This is the training-data release behind Qwen3.5-9B-Distill-Agent-Instruct, an Agent Instruct model distilled and post-trained from Qwen3.5-9B-Base. The title highlights three of the mixture's principal model-labelled trajectory families—Claude Fable5, GPT-5.5 Agent, and Claude Opus… See the full description on the dataset page: https://huggingface.co/datasets/lzy510016411/fable5-gpt5.5-opus4.7-mixed-agent-traces.hermes-agent-reasoning-tracescoding_agent_tracesWe release coding agent traces using Claude Code for
Opus
ISL, OSL, ISL_new counts
GPT-oss-120B
ISL, OSL, ISL_new counts and their raw texts
For Opus, only the locally saved files from the harness were used for analysis.
Coding agents take multiple turns to carry out a task from the input prompt. To analyze the token distribution, two models were selected: Anthropic's Opus and OpenAI's gpt-oss-120B. The input sequence length (ISL), output sequence length (OSL) and the uncached, new input… See the full description on the dataset page: https://huggingface.co/datasets/netpreme/coding_agent_traces.nemotron-gym-agent-calendar-qwen3.5-122b-131k-opencode-traces
Agent trace dataset
Decoding the literal token IDs
The prompt_token_ids / completion_token_ids / logprobs columns are the
verbatim tokens the serving engine emitted, stored PER AGENT STEP as a
list-of-lists (one inner list per turn). To turn them back into text you MUST
use the exact tokenizer the model was served with — a generic same-family
tokenizer will decode word tokens to garbage.
Served model / tokenizer source: Qwen/Qwen3.5-122B-A10B-FP8
from transformers… See the full description on the dataset page: https://huggingface.co/datasets/open-athena/nemotron-gym-agent-calendar-qwen3.5-122b-131k-opencode-traces.conda-forge-agent-tracesopen-agent-traces
Open Agent Traces
17,019 LLM-enriched agent trace events across 500 workflow runs in 10 enterprise domains and 3 workflow patterns.
Generated with ocelgen (pip install open-agent-traces) and validated against the OCEL 2.0 standard, PM4Py, and 5 semantic validation layers.
from datasets import load_dataset
ds = load_dataset("juliensimon/open-agent-traces", "incident-response")
for event in ds["train"]:
if event["run_id"] == "run-0000":
print(f"{event['event_type']:25s}… See the full description on the dataset page: https://huggingface.co/datasets/juliensimon/open-agent-traces.swe-bench-lite-agent-traces-v14
AgentBRANE SWE-bench Lite Agent Traces v14
This release contains the 1,890 harness-native agent traces selected by the sealed SWE-bench Lite v14 publication record (1,379/1,890 resolved, 73.0%). It includes Claude Code, Codex, and Pi sessions across seven models and three replicates. No internal research notes are included.
Load the observation table:
from datasets import load_dataset
traces = load_dataset("melissapan/swe-bench-lite-agent-traces-v14", split="train")
Each row… See the full description on the dataset page: https://huggingface.co/datasets/melissapan/swe-bench-lite-agent-traces-v14.read-along-ai-agent-traces
Read-Along AI - Agent Traces
This dataset contains the raw agent traces and conversation logs from the development of Read-Along AI, a submission for the Hugging Face Build Small Hackathon.
Dataset Description
These .jsonl files represent the unedited, behind-the-scenes "agent traces" of the AI coding assistant orchestrating the build of this project.
Sharing these traces fulfills the requirements for the "Sharing is Caring" bonus badge, providing the community… See the full description on the dataset page: https://huggingface.co/datasets/kingkw1/read-along-ai-agent-traces.agent-llm-traces
Multi-Benchmark LLM Agent Traces
A comprehensive dataset of OpenTelemetry traces capturing LLM inference behavior across multiple agent frameworks, benchmarks, and model providers. This dataset enables research into LLM performance analysis, agent behavior patterns, and inference optimization.
Collected by Exgentic - A platform for LLM observability and performance optimization.
Dataset Overview
This dataset contains 1,781 execution traces capturing detailed agent… See the full description on the dataset page: https://huggingface.co/datasets/DiscoPosse/agent-llm-traces.agent-traces
OpenAI-format Agent Traces SFT
A tokenizer-independent, text-only collection of 62,812 deduplicated agent conversations, normalized from publicly accessible Hugging Face agent-trace datasets. This repository is a derived collection, not an original authorship claim. All upstream authors are credited in REFERENCES.md, with pinned revisions, source license labels, retained dataset cards and source license metadata.
Contents
Split
Conversations
Train
62… See the full description on the dataset page: https://huggingface.co/datasets/beomi/agent-traces.glm52-datagen-r11-17-agent-calendar-tracespi-agent-traces-sft
pi-agent-traces-sft
Merged supervised-fine-tuning dataset of coding-agent session traces (the "pi"
harness format), converted to conversational format for Unsloth SFT of Qwen3-family
models.
Files
pi_sft_qwen3_24k.jsonl — recommended training file. Every session fits in
24,576 tokens (Qwen3 tokenizer). 5,717 sessions, max length 24,564 tokens.
pi_sft_qwen3.jsonl — full unfiltered set (6,929 sessions, up to ~632k tokens).
Sources (merged)… See the full description on the dataset page: https://huggingface.co/datasets/vojtavlas2/pi-agent-traces-sft.agent-traces-flashmaniaswival.dev agent traces for single-shot "Reimplement trackmania" prompt using Gemini-3.5-flash.
Generated code: https://github.com/dip-proto/flashmania
mcphunt-agent-traces
MCPHunt Agent Traces
Agent execution traces from the MCPHunt evaluation framework, measuring
cross-boundary data propagation in multi-server MCP agents.
Contents
main/ — 3,615 traces from 5 models across 147 tasks and 7 environment
variants (risky_v1/v2/v3, benign, hard_neg_v1/v2/v3). One JSON file per model.
mitigation/ — 2,706 traces from the prompt-mitigation study (M0--M3
levels) across 3 models.
live_guard_defense/ — 387 DeepSeek-V4-Flash traces from the… See the full description on the dataset page: https://huggingface.co/datasets/mcphunt-benchmark/mcphunt-agent-traces.agent-spaces-traceshermes-agent-reasoning-traces
Hermes Agent Reasoning Traces
Multi-turn tool-calling trajectories for training AI agents using the Hermes Agent harness. Each sample is a real agent conversation with step-by-step reasoning (<think> blocks) and actual tool execution results.
This dataset has two configs, one per source model:
Config
Model
Samples
kimi
Moonshot AI Kimi-K2.5
7,646
glm-5.1
ZhipuAI GLM-5.1-FP8
7,055
Loading
from datasets import load_dataset
# Kimi-K2.5 traces
ds =… See the full description on the dataset page: https://huggingface.co/datasets/insym/hermes-agent-reasoning-traces.neulab-nebius-swe-agent-trajectories-sandboxes-traces-terminus-2agent-race-traces
Agent Race Traces
Coding-agent session traces from running the same task and the same one-line prompt through multiple harnesses and driver models. Useful for comparing how different agent harnesses + models approach an identical real-world ML task.
Blog post: danielvanstrien.xyz/posts/2026/agent-race
The task
Fine-tune a model on biglam/on_the_books to identify Jim Crow laws, train via HF Jobs, push the trained model to the Hub.
The exact one-line prompt was… See the full description on the dataset page: https://huggingface.co/datasets/davanstrien/agent-race-traces.soc-agent-traces-10k
SOC-Agent-Traces-10K
Multi-step SOC investigation agent traces in session-trace format. Each
record is a complete investigation session: an alert arrives, an analyst agent
gathers evidence through nine read-only tools, and closes with a structured
JSON triage report.
Instead of single-turn alert → answer pairs, every record captures the full
reasoning trajectory:
alert → get_surrounding_events → get_process_tree → lookup_attack
→ search_sigma → get_asset_context →… See the full description on the dataset page: https://huggingface.co/datasets/alirezaaminzadeh/soc-agent-traces-10k.claude-fable-5-agent-tracesagent-traces-exampletutordesk-agent-traces
TutorDesk AI — Agent Traces
Sharing is Caring badge dataset for the HuggingFace Build Small Hackathon 2026.
Contents
build_sessions/ — Claude Code Build Sessions (native trace viewer)
7 raw Claude Code (Sonnet 4.6) JSONL sessions covering the full build of
TutorDesk AI — from blank repo through Phase 6 completion. Natively rendered by
HF Data Studio's agent trace viewer.
Sessions cover: repo scaffolding, Modal serving architecture, 5-agent pipeline… See the full description on the dataset page: https://huggingface.co/datasets/naazimsnh02/tutordesk-agent-traces.