evijit's picture
evijit HF Staff
README: update included / not included
2096126 verified
|
Raw History Blame Contribute Delete
2.47 kB
metadata
title: State of OSS - Spaces and agents
emoji: 🤗
colorFrom: yellow
colorTo: blue
sdk: static
app_file: index.html
pinned: false
short_description: Open models in Spaces and agent usage on the Hub, Oct 2026

State of open source on the Hub: Spaces and agents (Oct 2026)

index.html is the report, with two tabs:

  1. Open model usage via Spaces: what people build with open models, from Space creation, linked models, tasks, app categories and traffic.
  2. Agents on the Hub: how much Hub activity is agentic (the agent/<name> tag in huggingface_hub's User-Agent), what agents build, whether agent-made or human-made Spaces get more traffic, and which coding agents people develop with (config files in repos).

What's here

  • index.html: the self-contained report (Chart.js, HF design-system styling). Every chart has a "Show data" table.
  • analysis/: the pipeline that builds it (run order and definitions in analysis/README.md)
    • 01_classify_hubstats_spaces.py: purpose labels for public Spaces (hub-stats)
    • 02_fetch_hub_api.py: runtime, hardware and linked models for 2026 Spaces (Hub API)
    • 03_agent_config_files.py: coding-agent config files in models, Spaces and a sample of datasets
    • 04_prepare_chart_data.py: everything above + the internal CSVs → analysis/cache/chart_data.json
    • 05_build_report.py: inlines the data into analysis/report_template.html → the report
  • *.csv: aggregated internal extracts from the backend (Moonbot). Sources and start dates are in table.tsv.

Also included (team-only Space)

  • spaces_by_creator_client_*.csv: Moonbot's per-Space export (every Space created Apr–Sep 2026, incl. private and deleted ones, with creating client, agent name, visits and likes). This drives the per-Space agent charts.
  • moonbot.txt: the Slack thread with Moonbot's original numbers and caveats. state_of_oss_report.html: Moonbot's first report.
  • analysis/cache/chart_data.json: all aggregated chart data, so python3 analysis/05_build_report.py rebuilds the page directly. Run logs are alongside.

Not included

  • Parquet files in analysis/cache/ (hub-stats spaces.parquet, classified Spaces, Hub API pull, config-file scan stages). They're all built from public data (hub-stats and the Hub API), and steps 1–3 recreate them.
  • The Hugging Face Design System zip the report is styled with. The tokens it uses are inlined in analysis/report_template.html.