Personal Content Moderation Tool / Open source · 2026
Your space.
Your rules.
Platforms must moderate for millions. You should be able to moderate for yourself.
A portable moderation policy for the people who live with the comments.
Status: Public policy alpha; live integrations experimental.
001 / The problem
At platform scale, a wrong decision can silence someone.
A global platform has to weigh precision against recall, fairness against safety, free expression against harm — across every language, culture and context, at a volume no team can read.
Every tool it has is imperfect. Classifiers misread. Word lists and regular expressions miss what they weren’t written for. Reports are noisy; human review is slow. Coded language mutates weekly, slang cuts both ways, and communities disagree on where the line sits.
So platforms are cautious, and rationally so. A mistaken removal at that scale can erase legitimate speech for many people at once.
002 / The premise
The scale changes. So can the threshold.
A creator moderating replies under their own posts faces a different trade-off. One may reasonably prefer to catch more abuse and accept some false positives. Another may want a wider margin for argument. Both are choices about one space — not rules imposed on the platform or on anyone else’s voice.
A global moderation policy has to serve everyone. A personal moderation policy only has to serve you.
003 / The shift
From platform enforcement to personal control.
PCMT combines familiar methods — rules, lexicons, regular expressions, human judgement — with language-model analysis that can read context, recognise spelling variations and, when you authorize it, act on your behalf.
AI is not infallible. Every proposed decision should come with a reason you can read. The default is audit first; hiding or deleting is optional and only where platform APIs allow it.
The policy itself is portable Markdown and YAML that moves with you across your own Instagram and Threads presence. Portable does not mean every API can act on it.
004 / The experience
01
Choose a threshold
Balanced, Aggressive or Maximum — described in plain language, not scores.
02
Review your rules
Turn categories on or off. Add your own terms, variants and candidate patterns.
03
Export and instruct
Download Markdown and YAML, then hand them to your Muse agent with an audit-first prompt.
What you produce is a policy document. Connecting it to a social account requires separate, authorized access — this website does not do that.
Make your policy →005 / Proof of concept
Same words. Different context. Your threshold.
Synthetic example — no live account action
- “I hate you.”ReviewPersonal hostility; the owner can decide if it crosses their line.
- “You're fat.”ReviewAppearance-based insult; review in context.
- “The movie character says “I hate you”.”KeepQuoted dialogue, not abuse directed at the creator.
- “I hate rainy Mondays.”KeepOpinion about the weather, not harassment; no always-hide match.
- “Check my page for a discount, link in bio.”ReviewPossibly unsolicited promotion; ambiguous, so it's yours to judge.
Illustrative, hand-written dispositions — not output from a live model. Real interpretation varies, and every threshold produces some false positives and misses.
006 / The community
Language moves faster than lists.
Help track emerging coded hate, slurs and spam phrasing. A good submission includes the term, its variants, the context where it bites, and a benign example of where it doesn’t.
Every submission is read by a person. Accepted entries ship in versioned GitHub releases. Nothing crowdsourced changes anyone’s policy automatically.
007 / Open source
Read it. Fork it. Argue with it.
- Code
- MIT License
- Lexicon & policy text
- CC BY 4.0
- Current release
- v0.1.0-alpha · policy v1.2 · 193 candidate lexicon entries
License terms as stated in the repository; the LICENSE files there are authoritative.