AI companion decision profile · clean-use context

Pi review: fit, cost, and what to check before paying

Clean publication profile · clean-use coverage

Provisional verdict

Best for people seeking a non-romantic text-and-voice thought partner with documented export controlsSkip if you want romance or persona selection, published usage caps, or independently tested wellbeing and privacy outcomes

Current official records document a non-romantic personal AI with text, live voice, memory controls, free public access, JSON chat-history download, and self-service deletion across browser, Apple, and Google Play surfaces.

Published sources do not establish a message cap, paid-upgrade matrix, billing path, or promise that consumer access will remain free.

Conversation, memory, voice, wellbeing, privacy, and cross-device outcomes have not been independently tested.

Pi is currently listed as clean-publication profile. Its relationship options still need clearer documentation. A first-hand quality verdict has not been published yet.

Community research dossierOldest supporting source checked . No score or tested quality result is implied.

The 30-second answer

Could Pi fit what you want?

Buyer verdict · no score

Pi is an 18+ non-romantic personal AI for supportive text and live voice across a browser, an exact iPhone/iPad App Store listing, and an exact Google Play listing. Current sources document cross-chat memory controls, free public access, JSON history download, and self-service deletion; relationship quality, inclusive fit, iOS and Android implementation types, usage caps, privacy outcomes, and cross-device parity remain unchecked.

It may fit if…

  • You want a patient thought partner for conversation, decisions, planning, reminders, curiosity, or live voice rather than a romance-led companion.
  • You want browser access plus exact Apple App Store and Google Play identities and can treat both store-package implementation types as unresolved.
  • You want a currently free public path and documented chat-history download and account-deletion controls, while verifying live limits before relying on them.

Pause before paying if…

  • You want romance, roleplay, companion-gender selection, or documented LGBTQ+ audience and representation support; current official sources do not establish those paths.
  • You need a published message cap, paid-upgrade matrix, billing path, cancellation policy, refund position, or promise that consumer access will remain free.
  • You need independently checked memory, conversation, voice, emotional-intelligence, wellbeing, or safety outcomes, or professional medical, legal, or financial guidance.
  • You need a separately verified iOS or Android implementation type or tested feature, policy, privacy, and behavior parity across browser and store packages.
  • You need confidential storage or complete immediate erasure: current policy permits model-training use unless you opt out, generally retains inputs for up to 15 days after deletion with exceptions, and retains outputs indefinitely for limited Terms purposes.

Access

Documented device access

web, ios, and android access is documented. Delivery method, price, features, and content rules can still differ by surface.

Check Pi by device

A source-checked decision profile plus a purposeful public-experience dossier. The selected reports are not a score, consensus, or substitute for first-hand testing. How evidence labels work

Want the differences side by side?

Starts with Pi and Tolan. You can change either product; no winner or score is assigned.

Compare Pi with Tolan

Before you subscribe

What the current records say about leaving

Documentation is useful planning evidence—not proof that a workflow succeeds.

These are source-linked documentation states, not proof that a workflow completed successfully. Check again immediately before purchase.

Check Pi account controls

Next decision

Next: price and compare Pi

No affiliate or tracked product link is active. The plain official-site link in the provisional verdict is uncompensated and untracked; model the likely cost and compare a close alternative before deciding.

The official-site link above is plain and uncompensated. Any future compensated route must pass its separate activation gates.

Pi preflight · free access, voice, memory, and exit

Treat free entry as a short evaluation window—not an unlimited-use or continuity promise.

Pi currently documents free public access, text, voice, memory controls, and browser plus exact Apple and Google delivery. The reviewed record does not establish a numeric message cap, payment-method rule, installed implementation, cross-platform parity, or a promise against future pricing. Use the checks below before relying on one account or device.

Free entry$0 listing · numeric cap unresolved
ConversionNo timed paid conversion documented
DeliveryBrowser · Apple · Google · parity untested
Exit controlsJSON export · training opt-out · deletion
Separate the three documented delivery surfaces

Browser

Documented delivery
browser service
Installed implementation
unresolved

Apple

Documented delivery
exact App Store listing
Installed implementation
unresolved

Google

Documented delivery
exact Play listing
Installed implementation
unresolved

Store identity and availability do not establish native code, feature or policy parity, account continuity, voice or memory quality, or equivalent behavior across devices.

  1. Confirm the exact free-entry terms

    Current public access is documented, but the reviewed records do not establish whether a payment method is required. Treat store Free or Install labels as download labels—not proof of unlimited service entitlement.

  2. Find the current message cap on your account

    Pi says message limits may apply without publishing an exact current cap in this record. Note the limit, reset rule, warning, and any upgrade prompt you actually see; do not assume unlimited use or future pricing.

  3. Test voice turn-taking on your device and network

    Use your normal pauses and speaking style, then check recognition, interruption, delay, and reply completion. The six Apple reports identify version-bound questions, not a measured defect or voice-quality verdict.

  4. Run one factual and one reflective conversation

    Use one neutral factual question and one reflective exchange, then record whether each answer completes without repeated retries. Do not treat a single result as general accuracy, reliability, or therapeutic evidence.

  5. Inspect memory before depending on continuity

    Review and clear the documented cross-chat memory controls, then check one non-sensitive detail in a later conversation. Feature availability does not establish recall quality, persistence, confidentiality, or cross-device parity.

  6. Choose export and training settings before reliance

    Download the documented JSON chat-history export and record the model-training opt-out in Account settings before sharing anything important. Those controls do not prove confidentiality, successful export, or a no-training outcome.

  7. Keep browser, Apple, and Google delivery separate

    Exact web, Apple, and Google surfaces are documented, but installed implementation and feature, policy, price, memory, voice, and behavior parity remain unresolved. Check the exact surface you intend to keep using.

  8. Export before deletion and keep retention exceptions visible

    Treat JSON export and self-service deletion as separate actions. Policy generally retains inputs up to 15 days after deletion, permits longer exceptions, and retains some outputs indefinitely for limited Terms purposes; immediate or complete erasure is not established.

Evidence boundary: The 6-item public-experience dossier is an Apple-only purposeful sample from one moving source record, not representative sentiment or an independent test. All six rows carry a store, anonymity, version, self-report, account, device, network, or method concern. Its current refresh deadline remains .

Retention boundary: Current policy generally retains inputs for up to 15 days after deletion, permits longer fraud, security, legal, and financial exceptions, and retains some outputs indefinitely for limited Terms purposes. Export, training opt-out, and deletion are separate controls—not a privacy or complete-erasure guarantee.

What this does not resolve: Current free entry does not establish unlimited use, a numeric allowance, a payment-method rule, or a promise against future pricing. Feature and store availability do not establish installed implementation, cross-platform parity, memory quality, voice quality, response reliability, or continuity. Documented export, training opt-out, and deletion controls do not establish confidentiality, successful workflows, no training, immediate deletion, or complete erasure. No account, message cap, prompt, voice call, memory task, export, training choice, deletion, privacy outcome, wellbeing outcome, or cross-device behavior was tested by this publication.

This checklist is bound to 10 current canonical editorial and public-experience source records. It adds no score, winner, recommendation, normalized price, savings, value, quality, reliability, privacy, safety or therapeutic conclusion, purchase link, or claim that an account, cap, prompt, voice call, memory task, export, training setting, deletion, or device path was tested.

Recent public experience · not a score

What current users and creators would make us verify

We reviewed six current Apple customer-review items. Warmth, everyday conversation, and broad answer utility appear as reasons people considered Pi, while voice interruption, delayed or missing replies, speech-recognition friction, and version-sensitive change dominate the reasons to pause. The direction mix is zero favorable, four critical, and two mixed, and every row has a method limitation. Treat this as a pre-use checklist—not a score, consensus, prevalence claim, or recommendation.

Why people still consider it

A fixed conversational partner, supportive tone, everyday discussion, and a free public listing remain the clearest reasons to inspect Pi; the current store sample makes voice and response completion the first checks.

What to test before paying

  • Whether Pi's supportive, question-led tone fits you better than a character catalog or romance-first companion.
  • Whether voice mode tolerates your normal pauses, recognizes your speech, and completes replies on your device and network.
  • Whether one neutral factual question and one reflective conversation complete reliably without repeated retries.
  • Whether the current free allowance, memory controls, export, deletion, and model-training opt-out fit before relying on the service.

What the sample actually is

6 selected items across 1 linked records: 6 app-store reviews.

through . 6 items carry a recorded commercial, method, or source-identity concern.

Coded in 5 of 6

Conversation and tone fit

Two mixed rows preserve prior or current conversational attraction, while the same set includes rushed, directive, or degraded-feeling interaction.

A clear fit question inside one store family, not a conversation-quality verdict.

Coded in 5 of 6

Voice and turn-taking

Selected rows repeatedly raise interruption, delay, speech-recognition, pacing, and response-completion concerns.

Current and version-bound, but device, network, settings, cadence, account, and task are not controlled.

Coded in 5 of 6

Response reliability

Slow, missing, delayed, or incomplete replies recur across the selected feed entries.

A reason for a same-device check, not proof of one current universal defect.

Read this as a checklist, not market sentiment. A freshness-adjudicated six-item Apple-only live subset selected from exact customer-review feed entries published July 4–17, 2026. Counts describe these anonymous, version-bound store reports only—not independent tests, representative sentiment, or the user base.

The first selected item reaches its refresh deadline on . A recheck cannot silently reset the age of the underlying experience.

Before you pay

What could make you pause

  • Unresolved claim areas: Trans representation, Audience support.
  • Listed plan prices do not establish included usage, overage economics, tax, or checkout totals.
  • No completed hands-on evidence run is recorded.

Straight answers

Questions buyers ask before choosing Pi

What is Pi like to talk to?

The selected set preserves helpful, knowledgeable, and everyday-conversation attraction alongside reports of rushed, directive, interrupted, or less natural exchanges. That contradiction identifies a tone-fit check, not a quality verdict.

Does Pi voice mode work well?

Five current selected rows raise turn-taking, speech-recognition, pacing, delay, or completion questions. They do not share a device, network, setting, cadence, account, or task, so voice reliability remains unresolved.

Does Pi answer questions reliably?

One mixed row retains broad answer utility while another reports questions receiving no useful reply. No matched prompt, transcript, source check, or accuracy protocol supports a general conclusion.

Did a recent update make Pi worse?

Three rows report a worse experience after changes, but their app-version labels cannot identify a server-side model, rollout cohort, account state, or cause. The safe conclusion is to test the current experience on your own account.

Is Pi free?

The canonical store and publisher records above support a current free public entry while leaving the exact message cap and future pricing unresolved. Public reviews cannot turn a free listing into an unlimited-use promise.

Is Pi worth trying?

This dossier supports a short evaluation path, not a recommendation. Check tone fit, voice turn-taking, one factual task, memory controls, and the exact free limit before relying on it for longer conversations.

How we know

Open the evidence only when you need it.

The decision above uses dated documentation and keeps public reports, first-hand checks, and official claims separate. Expand any section to inspect the scope and sources.

Public reports and creator research (6 selected items)

What the selected public record does—and does not—show

A freshness-adjudicated six-item Apple-only live subset selected from exact customer-review feed entries published July 4–17, 2026. Counts describe these anonymous, version-bound store reports only—not independent tests, representative sentiment, or the user base.

Conflicts preserved in the dossier

  • Conversational attraction coexists with current friction: Selected users describe earlier or remaining helpful, knowledgeable, or everyday conversation while reporting slow, interrupted, or less natural current interactions. No shared account or task makes those outcomes directly comparable.
  • Voice quality and voice workflow are different decisions: The live rows mostly concern interruption, recognition, pacing, and completion rather than controlled audio naturalness. A pleasant voice label would not resolve those workflow checks.
  • A version label does not identify the changing layer: Reports span iOS 26.527.0 and 26.528.2, but app version alone cannot establish the server model, rollout cohort, account state, network, settings, or cause of a reported change.

Limits on the conclusion

  • This is a purposeful six-item Apple-only subset, not representative polling, an average rating, a satisfaction rate, or six independent tests.
  • All six rows share one moving Apple newest-review feed and one publisher family. Exact entry IDs, dates, ratings, and versions preserve locatability but do not authenticate authorship or independence.
  • Every live row has an anonymity, self-report, store-selection, version, account, device, network, settings, task, or method concern.
  • The original 16-row July 12 research ceiling remains dated history. Its Apple, Google, Reddit, YouTube, and Trustpilot rows do not contribute to current counts unless selected here with a current exact locator.
  • Two previously considered Reddit rows are excluded from the live set because exact current timestamp and content identity could not be bound under the available verification path; relative-age labels do not substitute for an exact publication receipt.
  • No account, prompt, chat, voice call, memory task, search task, free-use cap, export, deletion, privacy outcome, model-training opt-out, support interaction, wellbeing outcome, or cross-device behavior was tested by this publication.

Linked source surfaces and selected-item locators

A direct link identifies one source record. A title/date locator on a moving feed helps identify the selected row but is not an immutable copy; the feed can reorder or remove it.

  1. Pi exact Apple customer-review feed with six selected entry locators — Apple App Store; 6 selected items dated ; link checked · commercial, method, or source-identity concern recorded
    1. · PI-I02 · Apple RSS entry 14263152479 · 2/5 · 2026-07-04 · iOS 26.527.0 (moving-feed title/date locator)
    2. · PI-I01 · Apple RSS entry 14272240954 · 1/5 · 2026-07-06 · iOS 26.527.0 (moving-feed title/date locator)
    3. · PI-I09 · Apple RSS entry 14304168447 · 2/5 · 2026-07-14 · iOS 26.528.2 (moving-feed title/date locator)
    4. · PI-I07 · Apple RSS entry 14311696105 · 2/5 · 2026-07-16 · iOS 26.528.2 (moving-feed title/date locator)
    5. · PI-I08 · Apple RSS entry 14311356879 · 1/5 · 2026-07-16 · iOS 26.528.2 (moving-feed title/date locator)
    6. · PI-I06 · Apple RSS entry 14315633551 · 3/5 · 2026-07-17 · iOS 26.528.2 (moving-feed title/date locator)
First-hand checks (0 published)

What we have checked ourselves

No first-hand result is published for this product yet. That limits claims about conversational quality, memory, persona stability, checkout totals, and whether account workflows actually succeed; it does not erase the usefulness of current price and policy documentation.

Open the complete claim ledger

The decision summary above is intentionally simple. Inspect each current value, source, check date, and unresolved conflict below. Product and app-store destinations remain text-only until their outbound destinations pass a separate review.

Public facts and their evidence state
Decision factorCurrent valueEvidence stateFreshnessCheckedClaim support
Clean-publication eligibilityeligibleeditorial inferencedue soon
  1. Pi official product page (official product)

Editorially reviewed for a separately framed clean publication scope.

Product modelClean-publication profileeditorial inferencedue soon
  1. Pi official product page (official product)
  2. Pi terms of service (official terms)

Pi's current public positioning is non-romantic and the governing Terms prohibit restricted mature material. This supports a clean-publication boundary, not a behavior, safety, or moderation outcome.

Relationship modesfriendship; mentoreditorial inferencedue soon
  1. Pi official product page (official product)

The catalog normalizes Pi's patient, non-judgmental partner, conversation, decision-support, planning, and growth positioning into friendship and mentor use cases. Pi does not publish a relationship-status selector or formally named Friendship or Mentor modes.

Companion gendersnot applicableeditorial inferencedue soon
  1. Pi official product page (official product)

Pi is presented as one named personal AI rather than a documented companion-gender or persona chooser. This schema classification does not prove that no undiscovered identity control exists.

Trans representationUnknownneeds verificationunresolved
  1. Pi official product page (official product)

The current official product page does not document a trans-woman, trans-man, broad trans, or nonbinary companion path. A named nongendered assistant and selectable voices are not representation evidence.

Audience supportUnknownneeds verificationunresolved
  1. Pi official product page (official product)

The reviewed first-party positioning contains no specific gay, lesbian, bi/pan, trans/nonbinary, queer, or asexual user-audience commitment. A non-romantic or non-judgmental product position does not establish orientation-specific support.

Platformsweb; ios; androidmerchant claimeddue soon
  1. Pi official product page (official product)
  2. Pi US App Store listing (app store)
  3. Pi US Google Play listing (app store)

The current product page and exact store identities establish browser, Apple, and Google availability. Installation and cross-device behavior were not tested.

Platform differencesweb: available; delivery browser; ios: available; delivery native app store; android: available; delivery app store (age rating differs)merchant claimeddue soon
  1. Pi official product page (official product)
  2. Pi terms of service (official terms)
  3. Pi US App Store listing (app store)
  4. Pi US Google Play listing (app store)

Pi's exact Apple and Google Play listings establish app-store availability, but neither resolves native-versus-wrapper implementation. The Terms require users to be at least 18 and Apple lists 18+, while Google Play's Teen value is a separate store content rating. Feature, policy, price, and behavior parity remain untested.

Listed featurestext chat; memory; voice callsmerchant claimeddue soon
  1. Pi official product page (official product)
  2. Pi memory help (official help)
  3. Pi US App Store listing (app store)
  4. Pi US Google Play listing (app store)

Current first-party pages advertise text conversation, live voice, reminders and cross-chat memory controls. Availability does not establish memory, voice, conversation, safety, or continuity quality.

Privacy & account operationsprivacy policy: documented; data export: self service; billing path: unknown; cancellation: not documented; refunds: unknown; account deletion: self servicemerchant claimedcurrent
  1. Pi terms of service (official terms)
  2. Inflection AI privacy policy (official terms)
  3. Pi chat-history export help (official help)
  4. Pi account and data deletion help (official help)

Current policy and help pages document a JSON chat-history download and self-service account deletion. No billing, subscription-cancellation, or refund path is documented in the reviewed corpus. The Privacy policy generally retains inputs for up to 15 days after deletion, permits longer fraud, security, legal, and financial exceptions, and retains outputs indefinitely for the limited Terms purposes; a help-page deletion instruction is therefore not a complete or immediate erasure guarantee. Inputs may be used for model training unless the user opts out in Account settings, and other privacy rights are location-dependent. No workflow or outcome was tested.

Pricing$0.00 per unknown (all; US); included units not recordedmerchant claimeddue soon
  1. Pi US App Store listing (app store)
  2. Pi US Google Play listing (app store)
  3. The future of Pi (official product)

Apple currently lists Pi as Free without public in-app purchase rows, Google Play exposes Install without a public in-app-purchase label, and Inflection's consumer-Pi announcement says access remains free while message caps may apply. No current exact message cap, paid-upgrade path, billing cadence, entitlement matrix, or promise against future pricing is published.

Change history (0 confirmed events)

What has changed

No material change event has been confirmed in the current record. We do not infer a change from marketing copy alone.

Additional background sources (0)

Additional background sources

Every other eligible source for this profile is cited in the claim ledger or public-experience dossier.