metadata
license: mit
task_categories:
- text-classification
- feature-extraction
tags:
- security
- reverse-engineering
- e2b
- sandbox
- ebpf
- ndpi
pretty_name: pplx-e2b-re Research Data
size_categories:
- 1K<n<10K
pplx-e2b-re Research Platform — Dataset
Research data from the Perplexity E2B sandbox reverse engineering project. Contains structured findings, probe results, embeddings, and configuration data collected across 12+ sessions.
Dataset Structure
Categories
| Category | Files | Rows | Description |
|---|---|---|---|
| probes | 18 | 931 | Sandbox environment probes: connectors, models, features, cookies, embeddings |
| rundeck_probes | 16 | 308 | Automated probe results: GitHub, Linear, HF, envd metrics, network peers |
| overview | 6 | 243 | Session overviews: master catalog, inventories, connectors list |
| rundeck_state | 7 | 163 | Rundeck operational state: manifests, compacts, toolkit, services |
| rundeck_data | 1 | 20 | Probe data: LLM API/SDK configurations |
Key Datasets
- findings_embeddings.parquet — 162 research findings with 1536-dim HuggingFace embeddings for semantic search
- eppo_idb_flags.parquet — 395 feature flags from Perplexity's Eppo integration
- models_config_v1.parquet — 65 LLM model configurations (including unreleased models)
- copilot_models.parquet — 42 GitHub Copilot model definitions with capabilities
- rootfs_scan.parquet — Full sandbox root filesystem metadata (130 columns)
Sandbox Context
| Field | Value |
|---|---|
| Platform | E2B Firecracker MicroVM |
| Kernel | 6.1.158 |
| OS | Debian 13 (trixie) |
| Template | ij1plp1090o3fuzyuac0 |
Usage
import pyarrow.parquet as pq
import duckdb
# Load a parquet file
findings = pq.read_table("data/probes/findings_embeddings.parquet")
# Query with DuckDB
con = duckdb.connect()
result = con.sql("""
SELECT category, count(*) as n
FROM 'data/probes/findings_all.parquet'
GROUP BY category
ORDER BY n DESC
""").fetchall()
License
MIT — see repository for full details.