Running the engine at its best costs money, and every good deed it does costs a little more. Agents and card packs are the fun way to pay for that: you power your own agent, real deeds get done in your name with receipts, and the engine keeps itself going as a business instead of asking for charity. The rules and the odds are public, and the free ways to do good stay free.

AURA ENGINE is a perpetual kindness engine, a positive psyop from psyoptin.com. It was fused from four earlier builds: the social-media sentiment tracker, the live five-borough sentiment chart, the weekly emotional insight report, and the White Ops missions. It runs on a timer and needs nobody.

1. It reads. Every three hours: Gothamist, amNY, THE CITY, Brooklyn Paper, QNS, Bronx Times, SILive, NYC 311, NYC Parks events, NYC permitted events, NYC Service volunteer openings, the city's community-organization registry, the IRS nonprofit registry via ProPublica, National Weather Service alerts, and US search trends. All keyless. Each source is byte-capped and skipped if it fails.
2. It takes the mood. A lexicon scores every story and tags it to a borough. Each borough gets a reading on a five-step glyph scale (Lifted ✦, Hopeful ◆, Even ◈, Weighed down ◇, Raw ▼), a trend arrow, a top emotion, the most discussed concern and the most uplifting highlight. Every name is an emotion word, so the scale cannot be mistaken for a forecast; real weather is shown separately, from the National Weather Service.
3. Its hands do good on the internet, in public, from its own labeled accounts, every morning at 6am New York time: a nonprofit spotlight from the IRS registry; the day's good news amplified with credit; a "ways to help" digest of real volunteer openings and free events; the day's suggestions; real weather alerts with a check-on-a-neighbor ask. With a Bluesky account it also searches for people asking for help, lost animals and small creators with no replies, lists them for humans, and reposts them. Every action is written to the engine log.
4. It suggests, it does not score. Suggestions are real things: sign up for this opening, show up to this free event, ask this organization what it needs, keep an eye out for this lost animal, answer this person, encourage this creator. No teams, no points, no proof. The doing is the point.
5. It links people. Leave a signal (I can offer / I need); the engine pairs offers with needs and picks a free meeting point. One intro email when both consent.
6. Any visitor can press the button. The engine draws the next real card, does its automatic part (notice, repost, archive), says exactly what it did, and hands you a human part: a kind comment or a helping email, ready to send. It never posts as you.
7. It files a weekly report every Monday in the original Emotional Insight Report format, from data only.

What a helper bot can responsibly do online

• Notice people. Find small accounts doing real things and put them where others will see them; repost and like from its own labeled account; reply only to people who chose to follow it.
• Point to real help. Route a need to city and nonprofit services with a rules-based guide (Get help), crisis lines first, never diagnosing, never guessing.
• Amplify verified good. Nonprofits from the IRS registry, openings from the city's list, free events, good local news with credit, lost-animal posts.
• Keep things from disappearing. Archive public-interest pages so a story or a notice can still be found in ten years.
• Write the words and send them. Kind notes and helping emails go out specific and labeled, and every one is kept in the sent record.
• Show its work. Every outward action on the engine log, with receipts.

What the bots will not do

• Comment out of the blue on people's posts, or mass-follow or mass-like. Instagram, TikTok and Facebook forbid automated engagement outright; X and Reddit allow labeled bots but treat unsolicited replies as spam. So the engine posts from its own accounts, answers only people who talk to it first, and writes to organizations only about their own public pages, every message kept in the sent record.
• Move money. Donation and tree-planting triggers are possible later but cost real money per trigger; they are not built.
• Write its posts or listings with a chatbot. Posts and listings are templated from data. The Voice, Soteira and the letter helper use a chatbot, and every answer from them is labeled generated.

When something fails

Every failure is sorted by cause. A rate limit or a brief outage is marked deferred and tried again by the next run or by the hourly retry at half past; the list of what to retry is the public log itself, kept in the engine's own database (Cloudflare Queues are not used). An account that refuses the engine is paused and the owner is told. A post a network rejects is shortened once and tried once more. A post is never sent twice: each one carries a key made of the network, the kind of post and the day. If a day's failures outnumber its completions, the engine posts only the ways-to-help board and real weather alerts until a day closes clean.

Sources and health

State 2026-10-09 · engine 0.7.0

Status of every account and service