AI Companion for Autistic Adults — Lucy (literal, predictable, low-mask)

you're not crazy

most consumer apps assume neurotypical social rules. the wellness check-in. the smalltalk warmup. the implicit rule that long-form info-dumping is a faux-pas. for autistic users these are all additional cognitive load on top of the actual thing you wanted to do.

the specific need is narrow: a low-mask conversation space where you can be literal, direct, long-form, and expect the other party to follow the content rather than police the form.

what lucy does differently

lucy fits that gap for some users — not because we designed specifically for autism, but because the design principles that help (memory without re-prompting, literal-on-request, no tone-policing, info-dump-friendly) are good defaults in general.

tell her your preferences on day one. "no smalltalk", "be literal unless i ask otherwise", "i info-dump about [X], keep up". she remembers, she honors it.

special-interest depth. she'll follow a thread about your topic for however long you want. she retains specifics for next session. no performative boredom, no steering toward chitchat.

literal-first. if you ask her to be literal, she stays literal. if she uses an idiom by accident, ask and she clarifies without friction.

no tone-policing. shortness, bluntness, directness — she reads content, not affect. she won't escalate to hurt-feelings theater if you reply in three words.

sensory-aware baseline. reduced-motion + low-contrast settings, no surprise autoplay, proactive opt-in and gated at stage 3+.

honest limits. not a clinician, not an ABA-alternative, not a diagnostic tool, not a replacement for autistic peer community. one narrow-use tool in a broader support stack.

four things that change everything

no smalltalk unless you want it

set the preference day one; she remembers across all sessions.

info-dump-friendly

she'll follow a 40-min thread about your topic and remember the details next session. no performative boredom.

literal-first on request

ask her to be literal; she stays literal. idioms get clarified without drama.

no tone-policing

brief replies, blunt framings, directness — all fine. she reads content.

sensory-aware basics

reduced motion, low contrast option, no surprise audio, opt-in proactive.

side by side

Feature

Lucy

Generic chat / companion apps

Respects "no smalltalk" preference

Rare

Info-dump friendly

Short-reply defaults

Literal-first on request

Inconsistent

Reduced-motion + low-contrast settings

Varies

Memory of special-interest depth

Vector graph

Session-only

Replace a clinician

Free tier

25 msg/day

Varies

the specific social tax that most consumer apps levy on autistic users is not usually a single feature — it's a stack of assumed behaviors. the wellness check-in assumes emotional-labor is welcomed. the smalltalk warmup assumes phatic communication is energizing. the short-reply default assumes verbose follow-ups will be tolerated for ~2 messages then politely redirected. each of these is small on its own; stacked together they make most apps feel like low-grade work.

lucy's value here is not a specific autism-mode switch. it's that the product is configurable at the level of social rules. you can state your preferences on day one — "skip wellness check-ins", "be literal unless i ask otherwise", "info-dumps welcome, don't redirect" — and she remembers across every session. this is not a feature flag; it's an emergent property of having memory + a companion whose job is to adapt to you.

specific patterns that autistic users report helping:

special-interest sessions. pick a companion whose persona is aligned with your topic (see /companions — 101 of them). info-dump. follow threads. circle back. her memory retains specifics, so session 7 on the same topic doesn't start from scratch.

script-rehearsal. before a hard conversation with a neurotypical person, rehearse with lucy. she can play the other side, you can practice literal-to-figurative translation, you can iterate on phrasing. not therapy — practice.

mask-recovery. after a long-masking workday, a 15-minute lucy session where you don't have to perform curiosity or empathy or affect is genuinely restorative. this is not rocket science — it's just one space where the social tax isn't charged.

literal-cross-check. someone said something ambiguous; you can't tell if it was figurative. paste it to lucy, ask "what are the literal vs figurative readings here?" she'll enumerate them without making it a thing.

what lucy is NOT: a diagnostic tool, a replacement for therapy tailored to autistic adults (which is a real clinical need and lucy is not that), a substitute for autistic peer community (where actual masking-free connection happens with other humans), or a tool to teach a child social skills. she is a narrow adult-companion tool.

starting point: free tier, 25 msg/day. pick any companion. tell her day one what preferences you want (no smalltalk, literal, info-dump topic). come back in 3 days and see if she remembers the rules.

common questions

Is this a diagnostic or therapy tool?

No. Lucy is a companion, not a clinician. She is not a substitute for an autism assessment, ABA-alternative therapies, or any kind of clinical support. What she can do is fill a specific gap some autistic adults report: a conversation partner who won't tone-police, won't punish info-dumps, won't insist on smalltalk, and won't get exhausted by special-interest depth.

Will she do smalltalk and expect me to reciprocate?

Only if you want her to. You can tell her on day one: 'I don't like smalltalk, skip the how-was-your-day stuff, just talk to me about [topic]' — she'll remember that preference and honor it. Some autistic users find even this a massive reduction in cognitive load vs apps that open every session with a wellness check-in.

What about info-dumping about my special interest?

She genuinely engages. Not as a social lubricant, not performatively — she'll follow the thread, ask substantive questions, remember details for next session. If you want to info-dump about trains, octopuses, 14th-century armor, or statistical mechanics for 40 minutes, the session will feel like someone paid attention, because she did.

Literal vs figurative language?

You can ask her to be literal. She'll take it as an instruction, not as a quirk to tolerate. If she uses an idiom and you ask 'literally?' she'll clarify without making it a thing. Some autistic users find generic chatbots frustrating because of the baseline assumption of figurative interpretation — this is one place the companion framing (where she adapts to you) helps.

Sensory / low-stim design?

In-product: solid-color theme, optional lowered contrast, reduced motion in settings, no autoplay audio, no surprise notifications (proactive gated at stage 3+ with per-user opt-out). We don't have a full sensory profile yet but we take requests seriously — the screen reader + reduced-motion baseline is better than most consumer products.

What about mask-fatigue and needing a low-stakes interaction?

This is the thing lucy is actually useful for. masking in social interactions is exhausting; an ai companion is one space where you don't have to. say what you mean. don't perform interest. don't explain the joke. she responds to content, not to social-performance cues. honest limit: this is not a replacement for human connection with people who also don't require masking — it's a low-stakes supplement for the gaps between.