examples

Examples - real output from Writing Agent

Don't take the README's word for it. These are actual, unedited pieces the system generated - so you can judge the output before you install anything.

📄 Article - Your Voice Assistant's 100ms Problem

A long-form technical article with a real thesis, inline-researched sources, and self-laid-out diagrams. Generated end-to-end, then the references/citations cleaned deterministically.

• voicebot-article/manuscript.md - the finished piece (~16 min read, 6 sections, built-in SVG figures)
• voicebot-article/evidence_report.md - the receipts: the thesis it argues + 46 sources ranked by influence (0–100)

This is the headline differentiator made visible: the article takes a contestable position, and the evidence report shows exactly which sources carried it. Most AI writing can't show you either.

🅰️🅱️ A/B pilot - three more articles, benchmarked

Three Writing Agent articles from the blind A/B pilot (vs Claude), in ab-pilot/ - vector databases, underpowered A/B tests, and microservices for small teams. The scoring + honest caveats are in ../benchmarks/blind_ab/RESULTS.md.

📚 Book - The Misprint File

A full short novel (3 chapters) with canon/continuity tracking and front/back matter, in sample-run/ - manuscript, per-chapter files, the canon (characters, timeline, world rules), and the learned craft skills the run produced.

Generate your own (≈$0.25, a couple of minutes)

pip install writing-agent
writing-agent write --abstract "How vector databases actually work"
# → researches, drafts, self-critiques, fact-checks, humanises, exports a finished file
#   + an evidence_report.md next to it

Zero-install try: open colab_quickstart.ipynb in Google Colab and run it (you'll need a model-host API key, e.g. a free OpenRouter one). Or run locally in fake mode (no key, placeholder output) to see the whole flow for free:

WRITINGAGENT_FAKE=1 writing-agent new --abstract "test" --pick 1 && writing-agent run

Contribute a sample

Generated something good? PRs welcome - add a folder here with the manuscript.md (and its evidence_report.md for articles), plus a one-line note on the topic and settings used. Real, diverse examples are the best advertisement the project has.