Is Spokenly Safe? Local, BYOK & Pro Cloud Privacy Verdict (2026)

TL;DR: Spokenly's safety depends entirely on which of its three architectural modes you use. Local Only Mode is genuinely safe — audio runs through OpenAI Whisper Large-v3 and NVIDIA Parakeet locally on Apple Silicon with no network calls during transcription. BYOK cloud mode is only as safe as whichever provider you bring keys for (OpenAI, Deepgram, Groq, Anthropic, or Google). Pro managed cloud routes audio through five named subprocessors per the privacy policy effective March 2, 2026: Cerebras, Fireworks, Groq, Mistral AI, and ElevenLabs.

Three structural caveats apply across all modes:

• No compliance attestations. The privacy policy does not reference SOC 2 Type II, HIPAA BAA, ISO 27001, GDPR, or CCPA. No external audit, no Business Associate Agreement, no certification framework. This disqualifies Spokenly for regulated industries.
• No company entity or jurisdiction. The privacy policy lists only admin@spokenly.app as the contact. Developer Vadim Akhmerov is identified via the App Store listing, not via the privacy policy. Named developer with no disclosed corporate entity.
• iOS keyboard caveat. The developer's published App Store replies recommend switching to online models for iOS keyboard reliability — which defeats the on-device privacy benefit for iOS users who chose Spokenly precisely for local processing.

For users who want a single audit-able privacy guarantee without choosing a mode, alternatives like Voibe simplify the question with two clean modes — a fully on-device mode (Whisper on the Apple Silicon Neural Engine, nothing leaves the Mac) or a private zero-retention cloud that runs only open-source models and is never trained on — with no BYOK and no subprocessor chain to audit in either. Voibe's privacy policy and cloud AI privacy page commit that your audio and text are never stored, never sold, and never used to train any AI model — with a fully on-device mode available when nothing should leave the Mac. Voibe costs $149 lifetime versus $359.64 over 3 years of Spokenly Pro — $210.64 cheaper with a zero-retention posture and no per-token API exposure.

This article walks through each Spokenly mode in detail, the privacy policy's load-bearing claims (and silences), the iOS keyboard documentation gap, a five-question Spokenly Safety Decision Tree, and the alternatives whose two-mode design simplifies the privacy question.

Disclosure: Voibe is our product. We verified Spokenly's privacy policy effective March 2, 2026, the App Store listing (ID 6740315592, v1.7.4 April 7 2026), and the Spokenly homepage feature claims (Local Only Mode, MCP server) on May 26, 2026. Quotes from the privacy policy are verbatim; we cite specific paragraphs where applicable.

The rest of this article walks through each row, gives you the three architectural modes side-by-side, and provides a five-step Spokenly Safety Audit to make your own call.

Unlike most dictation apps that ship one architecture, Spokenly ships three. The mode you select determines where your audio goes, who sees it, and what retention applies. This is the load-bearing structural insight about Spokenly's privacy — most “is X safe?” investigations cover a single architecture; Spokenly's three-mode design forces a more nuanced verdict.

Mode 1: Local Only Mode (Free, On-Device)

• What happens: Audio captured to memory → transcribed by OpenAI Whisper Large-v3 or NVIDIA Parakeet on Apple Silicon's Neural Engine → text written to active field → audio discarded.
• Where audio goes: Nowhere. No network calls during transcription.
• Subprocessors: None.
• Retention: None — audio is in memory only.
• Privacy posture equivalent to: Voibe, VoiceInk, MacWhisper, Apple Dictation (mostly on-device on Apple Silicon).

This is the strongest privacy posture Spokenly offers. For Mac users with Apple Silicon (M1-M4) who can live with Whisper Large-v3 or Parakeet's accuracy without LLM cleanup, this mode is genuinely on-device and architecturally safe.

Mode 2: BYOK Cloud Mode (Free of Spokenly Fee)

• What happens: Audio captured to memory → sent to user-configured API provider (OpenAI, Deepgram, Groq, Anthropic, or Google) → transcript returned → audio discarded by Spokenly.
• Where audio goes: To whichever provider's API key you configured. Each provider has its own data-handling defaults, retention, and security posture.
• Subprocessors: The single provider you selected. Spokenly is a passthrough.
• Retention: Spokenly does not retain. Provider may retain per its own API terms — read each provider's data-handling policy.
• Privacy posture equivalent to: Whichever provider you chose, as if you were calling their API directly.

The BYOK mode is privacy-neutral from Spokenly's perspective and entirely provider-dependent from yours. OpenAI's API, for example, has different retention defaults for free-tier users vs paid-tier vs zero-retention agreements. Deepgram, Groq, Anthropic, and Google each have their own policies. For sensitive content, the right move is to verify the chosen provider's policy first, then route audio through Spokenly as a thin client.

Mode 3: Pro Managed Cloud ($9.99 / month)

• What happens: Audio captured to memory → sent to Spokenly's managed pipeline → routed through subprocessors → transcript returned → audio discarded per Spokenly's stated policy.
• Where audio goes: Spokenly's five named subprocessors per the privacy policy effective March 2, 2026: Cerebras, Fireworks, Groq, Mistral AI, ElevenLabs.
• Subprocessors: Five distinct entities, each with its own privacy policy and security posture.
• Retention: Spokenly states audio is not stored. Subprocessor-level retention is not detailed in the public policy.
• Privacy posture equivalent to: A managed-pipeline cloud product with a multi-vendor data chain.

This is the most convenient mode (no BYOK setup, one $9.99 / month subscription) and the most complex privacy posture (five subprocessors, undocumented contractual flow-down).

The Spokenly privacy policy effective March 2, 2026 is reasonably substantive — more detailed than Wisprtype's sparse Notion-hosted policy, less detailed than Wispr Flow's published subprocessor list with named regions. Here's what it documents and what it leaves silent.

What the Policy Documents

• Audio recordings are not stored on Spokenly's servers. This is the load-bearing claim that defines Spokenly's data-retention posture. It applies across modes — Local Only, BYOK, and Pro managed cloud.
• Five subprocessors named for Pro managed cloud: Cerebras, Fireworks, Groq, Mistral AI, and ElevenLabs. Each is a separate data-handling entity.
• Analytics collected: button clicks and page views. No personally identifiable information per the policy.
• Contact: admin@spokenly.app.
• Revision date: March 2, 2026 — recent enough to cover the current product surface.

What the Policy Does Not Document

• No SOC 2 / HIPAA BAA / ISO 27001 / GDPR / CCPA. No external audit framework is referenced. No Business Associate Agreement path. No regional data-residency commitments.
• No company entity or jurisdiction. The policy does not name a legal entity, a country of registration, or a corporate parent. Developer Vadim Akhmerov is identified only via the App Store listing.
• No subprocessor-level retention details. The five subprocessors are listed by name but the policy does not detail which workload runs at which subprocessor, what retention applies at each layer, or what contractual flow-down requires.
• No BYOK provider data-handling specifics. The policy treats BYOK as out-of-scope for Spokenly's retention — your audio is governed by the chosen provider's terms, which the policy correctly notes but does not elaborate.
• No iOS keyboard data-handling clarification. The keyboard reliability fix (switch to online models) shifts iOS privacy posture from local-only to cloud-by-default, but the privacy policy does not address this product reality.
• No retention windows in days or months. “Not stored” is the framing, but processing windows, log retention, and analytics retention are not specified.

What the Documentation Gaps Mean in Practice

The combination of a substantive privacy policy without an external audit, no named entity in the policy text, and the iOS keyboard reliability caveat produces a documentation posture that is:

• Adequate for general consumer dictation — drafts, emails, notes, AI prompts.
• Insufficient for regulated or compliance-audited work — HIPAA, attorney-client privilege, NDA-bound source code, SOC 2-required procurement.
• Better than the most anonymous indie peers (Wisprtype, parts of the Whisper-wrapper ecosystem) but less than VC-backed peers (Wispr Flow's published subprocessor list, Willow Voice's YC company page disclosure).

Three structural caveats apply to Spokenly regardless of which mode you choose. These are not mode-specific privacy issues — they're cross-cutting concerns about the product and its documentation.

Caveat 1: No SOC 2 / HIPAA / ISO 27001 / GDPR / CCPA Attestations

The Spokenly privacy policy does not reference any external compliance framework, audit, or certification. No SOC 2 Type II attestation, no HIPAA Business Associate Agreement, no ISO 27001 certification, no GDPR or CCPA processor agreement. This is consistent with most consumer Mac dictation tools — Voibe, VoiceInk, Superwhisper, and MacWhisper similarly do not carry these attestations — but it disqualifies Spokenly for any regulated workflow where compliance documentation is required.

For regulated alternatives:

• HIPAA-covered dictation: Dragon Medical One ($79-99/user/month with BAA), Sonix Enterprise (HIPAA on Enterprise), or a dedicated medical-scribe product like Suki AI or Heidi Health. See our best dictation software for doctors guide.
• Attorney-client privileged work: Local-only dictation eliminates the cloud disclosure surface. See our best dictation software for lawyers guide.
• SOC 2-required procurement: Wispr Flow Enterprise has SOC 2 Type II, ISO 27001:2022, and HIPAA BAA available across plans. See our is Wispr Flow safe? investigation.

Caveat 2: No Disclosed Corporate Entity

The Spokenly privacy policy lists only admin@spokenly.app as the contact. There is no:

• Company name (e.g., “Spokenly, Inc.” or “Akhmerov Software LLC”)
• Country of registration (US, EU, UK, etc.)
• Registered address or office location
• Team page or about page disclosing personnel

Developer Vadim Akhmerov is identified as the developer on the App Store listing. Apple's App Store developer disclosure requirements provide this name, but the privacy policy does not separately confirm jurisdiction or corporate structure.

For procurement-driven privacy reviews — particularly Enterprise IT and legal teams that require a named legal counterparty — this is a documentation gap. The right mitigation: contact admin@spokenly.app and request entity and jurisdiction disclosure in writing before Enterprise deployment. Or use a product with a published entity (Wispr Flow's Wispr, Inc., or Willow Voice's YC company page).

Caveat 3: The iOS Keyboard Reliability Trade-off

App Store reviews on Spokenly's iOS app document keyboard reliability issues — unexpected app switches, recording-start failures, device-performance limitations with local models. The developer's published replies in the App Store recommend switching to online models for keyboard reliability.

The structural implication: if iOS keyboard users follow the recommended fix, the iOS keyboard posture shifts from local-only to cloud-by-default. The on-device privacy claim that draws users to Spokenly in the first place no longer applies to the iOS keyboard surface under the recommended configuration.

The pragmatic Mac-only buyer can largely ignore this caveat — the Mac app does not suffer the same issues. The pragmatic iOS-keyboard-priority buyer should weigh Willow Voice's iOS voice keyboard or Wispr Flow's iOS keyboard as alternatives that ship cloud-first by design and don't trade away an architectural claim to get reliability.

Use the Spokenly Safety Decision Tree to decide which mode is safe enough for your specific situation. The five questions, in order, take you from the lowest-risk use case to the highest. Stop at the first question where you cannot accept the answer Spokenly currently provides.

1. Are you dictating only general content (drafts, emails, notes, AI prompts, casual messages) on your Mac? If yes — Spokenly Local Only Mode is reasonable. Whisper Large-v3 or Parakeet on Apple Silicon, no cloud route, no subprocessor chain. Continue to question 2 only if you need cloud accuracy.
2. Do you need cloud accuracy and are comfortable managing API keys at multiple providers? If yes — Spokenly BYOK cloud mode works, but the privacy posture is the provider's posture. Read OpenAI / Deepgram / Groq / Anthropic / Google's API data-handling policies before routing sensitive content. Continue to question 3 if you don't want BYOK setup.
3. Want managed cloud without BYOK and willing to accept a 5-subprocessor data chain at $9.99/month? If yes — Spokenly Pro is the convenience path. Cerebras, Fireworks, Groq, Mistral AI, and ElevenLabs each carry their own data postures; the contractual flow-down is not documented in the public privacy policy. Continue to question 4.
4. Is the content covered by HIPAA, attorney-client privilege, NDA, or compliance regulation? If no — Spokenly across any mode is a reasonable consumer product. If yes — Spokenly is disqualified across all three modes (no SOC 2, no HIPAA BAA, no ISO 27001). Skip to question 5 to evaluate the architectural alternative, or use Dragon Medical One / Sonix Enterprise / a dedicated medical-scribe product.
5. Want a simple two-mode design with one durable privacy promise? If yes — dictation tools like Voibe simplify the choice: a fully on-device mode (Whisper on the Apple Silicon Neural Engine, nothing leaves the Mac) or a private zero-retention cloud that runs only open-source models and is never trained on. There is no BYOK to manage and no subprocessor chain to audit in either mode; the durable promise is the same — your audio and text are never stored, never sold, and never used to train any AI model. Voibe at $149 lifetime versus 3 years of Spokenly Pro at $359.64 saves $210.64 (59%) with one durable privacy posture.

The pattern: the further you progress through the tree, the more on-device architecture wins. For the first three questions, Spokenly offers viable answers — different ones depending on which mode you select. By question 4, the absence of compliance attestations becomes a structural blocker for regulated work across all modes. By question 5, the architectural answer beats the policy answer.

Spokenly's three modes sit at very different points on the privacy spectrum. Here's how each one compares against the major peer postures we've investigated in this series.

Spokenly Local Only Mode is structurally comparable to Voibe's on-device architecture. Spokenly Pro managed cloud has more subprocessors than most peers and no compliance attestation — placing it weaker than Wispr Flow Pro (audited stack) and Willow Voice (default opt-out + AI Mode anonymization commitment) on the cloud-comparison axis.

Spokenly's three-mode design illustrates a deeper category lesson: there is a difference between architectural privacy and audited privacy. Local Only Mode is architectural — audio processing happens on the user's Apple Silicon chip and never crosses the network boundary. Pro managed cloud is audited-by-policy — Spokenly states audio is not stored, but the user trusts this through the privacy policy rather than through a third-party audit.

Five things architectural privacy delivers that audited privacy cannot:

• Survives a policy change. A privacy policy can be updated with notice. Audio that never crosses your network boundary cannot be re-classified by a future policy revision. Spokenly's privacy policy is two months old at publication; the next revision could materially alter the framework.
• Survives a subprocessor incident. Five subprocessors (Cerebras, Fireworks, Groq, Mistral AI, ElevenLabs) each represent a separate breach surface. On-device processing has zero subprocessors for dictation data.
• Survives an acquisition. Indie solo-developer products like Spokenly can change hands. New ownership may bring new data postures. On-device data has nothing to transfer.
• Survives a documentation gap. The current Spokenly privacy policy does not document HIPAA BAA, subprocessor flow-down, retention windows, or company entity. Decisions made under documentation uncertainty depend on the gap remaining benign. On-device dictation has nothing to document.
• Survives legal compulsion. A subpoena or national security letter can compel a vendor to preserve data normally discarded. On-device processing removes the vector — there is no preserved data, and the vendor cannot produce what it never had.

None of this means cloud dictation is unusable — Spokenly Pro and the BYOK path are reasonable for general content. It means cloud dictation is contract-driven privacy, and the contract is only as strong as the documentation, the auditor, and the policy's continuity. For confidential, privileged, regulated, or compliance-audited work, architecture is the stronger guarantee. For a deeper treatment of this distinction, see our cloud vs local dictation guide and the AI Privacy Tracker.

Run this five-step audit before committing Spokenly for any work where data handling matters. Each step takes 2-10 minutes.

1. Confirm your mode and read the corresponding privacy section. Open Spokenly settings, identify whether you're using Local Only Mode / BYOK cloud / Pro managed cloud. Each mode has a different privacy posture — Local Only is on-device, BYOK is provider-dependent, Pro is five-subprocessor managed cloud. Don't make safety decisions until you know which mode you're operating in.
2. If using BYOK cloud, read each provider's API data policy. Your audio is governed by OpenAI / Deepgram / Groq / Anthropic / Google's API terms — not by Spokenly's privacy policy. Open the chosen provider's API data-handling page. Verify retention defaults, zero-retention options, and any compliance attestations relevant to your work.
3. If using Pro managed cloud, verify each subprocessor matches your risk tolerance. Cerebras, Fireworks, Groq, Mistral AI, and ElevenLabs each have separate privacy policies. Review whichever subprocessor handles the workload you care about. If you're not sure which subprocessor handles which workload, contact admin@spokenly.app and ask in writing.
4. Check your work against the regulated-content disqualifier. If your dictation includes HIPAA-covered content, attorney-client privileged work, NDA-bound source code, or compliance-audited material, Spokenly is disqualified across all three modes (no SOC 2, no HIPAA BAA, no ISO 27001). Use Dragon Medical One, Sonix Enterprise, or a dedicated compliance-attested product instead — or use Spokenly Local Only Mode in a separately-attested environment (your Mac under documented MDM with FileVault, etc.).
5. Run an outbound-traffic monitor during a Local Only Mode session. Install Little Snitch or another macOS network monitor. Start a Local Only Mode dictation session. Outbound traffic from Spokenly during transcription should be zero. If you see network calls, you're not in Local Only Mode or there's a configuration issue worth investigating.

If any of the steps fail or feel uncomfortable, on-device dictation tools like Voibe eliminate the mode question — there is no BYOK to manage, no subprocessor chain to audit, no Pro tier to subscribe to. The architectural answer beats the policy answer for sensitive work.

Voibe is a Mac-native dictation app built around a durable promise: your audio and text are never stored, never sold, and never used to train any AI model. Voibe gives you two user-selectable modes. In on-device mode, Voibe runs OpenAI Whisper on Apple Silicon's Neural Engine — audio is captured into memory, transcribed locally, written into the active text field, and discarded, so nothing leaves the Mac (this mode requires an Apple Silicon Mac, M1 or later). In private cloud mode, audio goes over an encrypted connection to Voibe's own infrastructure, runs only open-weight models, and is deleted the moment transcription completes — no proprietary models from the major labs. There is no BYOK option and no subprocessor chain to audit in either mode.

Mapped against the Spokenly privacy questions raised above:

• Architecture. Voibe runs on-device on Apple Silicon's Neural Engine, or through a private zero-retention cloud that uses only open-weight models — your choice. In on-device mode there are no cloud servers and no third-party LLM providers in the dictation path.
• Modes. Two clean modes — fully on-device, or private open-source cloud. No BYOK setup to manage, no Pro tier to subscribe to, and no subprocessor chain to audit in either.
• Subprocessor list. In on-device mode nothing is transmitted, so there is nothing to list; in private cloud mode audio is deleted the moment transcription completes and is never used to train any AI model.
• Regulated content. For PHI and other regulated dictation, Voibe's on-device mode keeps audio on the clinical device so nothing leaves the Mac; the durable promise across both modes is that your audio and text are never stored, never sold, or used to train any AI model. See our HIPAA dictation guide for the clinical pathway.
• Entity. Voibe is developed by a disclosed team with a published privacy policy at getvoibe.com/privacy.
• Permissions. Voibe requests microphone access and macOS accessibility permission — the minimum surface required to capture audio and paste text into the active field. No screen recording, no camera, no full-disk access.
• Network monitor. Run Little Snitch during a Voibe on-device dictation session. Outbound traffic from Voibe during transcription is zero.
• Account. Voibe does not require an account to dictate.

Pricing: $7.50/month, $59/year, or $149 lifetime for unlimited dictation. Voibe runs on all Macs; on-device mode requires an Apple Silicon Mac (M1 or later). Voibe also includes Developer Mode for VS Code and Cursor with file/folder name resolution — useful for technical workflows where Spokenly's MCP server is the configurable alternative. Over 3 years, Voibe lifetime at $149 is $210.64 (59%) cheaper than Spokenly Pro at $359.64.

Try Voibe for Free — install, grant microphone and accessibility permissions, dictate. No account, no credit card, no BYOK setup, no subprocessor chain to audit, no Pro tier to subscribe to.

• 8 Best Spokenly Alternatives for Mac (2026) — Full alternatives roundup with pricing, architecture, and decision tree.
• Spokenly Review (2026) — Hands-on review of the product with accuracy testing, setup walkthrough, and MCP server analysis.
• Spokenly Pricing (2026) — Free + BYOK math, Pro $9.99/mo, 3-year TCO breakdown.
• Voibe vs Spokenly — Head-to-head comparison with Different Fits verdict.
• Is Wispr Flow Safe? — Sibling investigation for the cloud-first peer.
• Is Superwhisper Safe? — Sibling investigation for the on-device + BYOK peer.
• Is Aqua Voice Safe? — Sibling investigation for the cloud-only peer.
• Is Willow Voice Safe? — Sibling investigation for the privacy-default-protected cloud peer.
• Is Otter Safe? — Sibling investigation for the meeting-transcription peer.
• Is Dragon Safe? — Sibling investigation for the legacy enterprise peer.
• Is Claude Code Safe? — Sibling investigation for the developer-confused consumer-vs-commercial peer.
• Is Blip AI Safe? — Sibling investigation for the young indie cloud peer with strong privacy claims and thin third-party verification.
• Is VoiceDash Safe? — Sibling investigation for the OpenAI-routed cloud peer with a two-perimeter trust model.
• Is Voicy Safe? — Sibling investigation for the Groq-routed cloud peer (no-training promise on marketing pages only).
• Is Wisprtype Safe? — Sibling investigation for the local-by-default, closed-source peer (telemetry shipped on despite the policy).
• Is VoiceInk Safe? — Sibling investigation for the open-source GPL v3 on-device peer (zero telemetry, verified in source).
• Is Handy Safe? — Sibling investigation for the free MIT-licensed local tool (no cloud transcription path at all).
• AI Privacy Tracker — Cross-tool privacy posture comparison across 30 AI tools.
• Cloud vs Local Dictation — Architectural framing for the privacy question.
• Voice Data Privacy — Pillar with deeper privacy frameworks.