Poke
Poke
AI AgentActive

Poke

Poke is a proactive personal AI assistant available through Apple Messages, Telegram, WhatsApp and RCS. This independent guide examines its recipes, account connections, privacy terms, confirmation boundaries, reliability, task selection, alternatives and a safe two-week trial.

188

Views

0

Likes

May 2026

Added

poke.com

Website

Tags

Pokepersonal AI assistantAI agentiMessageWhatsApp

Product Preview

A quick visual look at Poke before you visit the official site.

Published 5/23/2026
Poke screenshot

Editorial Review

About Poke

Poke is a consumer AI assistant designed to live inside messaging rather than a separate chatbot workspace. Its official documentation currently lists Apple Messages, Telegram, WhatsApp and RCS. A user can send text or voice messages, request reminders, connect services and install “Recipes” that package repeatable tasks. The product’s central advantage is low interaction cost: delegating something feels like texting a contact.

That convenience is also the main evaluation risk. A friendly conversation can obscure the difference between drafting a suggestion, reading connected data and taking an external action. Poke should be judged task by task: what information it receives, which account can be changed, whether confirmation is required, how failure is surfaced and whether the user can reverse the result.

Official Poke messaging assistant product preview
Poke’s messaging interface reduces setup friction. The conversation is an instruction surface, however, so users should keep approvals explicit and avoid sending secrets or high-risk requests.

What the messaging-first design changes

Design choicePractical benefitHidden cost to check
Lives in messagingNo new interface to learn; easy follow-upImportant permissions can feel as casual as ordinary chat
Proactive remindersInformation arrives at the useful momentNotification fatigue, incorrect timing and sensitive previews
Connected accountsCan act with personal contextBroader data exposure and external side effects
RecipesReusable automation without building a workflowRecipe behavior, publisher and permissions may change
Text and voice inputFast capture while mobileAmbiguity, transcription errors and bystander privacy
Model routingService can select a model for the taskUsers may not know which provider receives particular content

A hand-drawn trust map

 you
  |
  | text / voice / photo
  v
 [messaging network] ---> notification preview
  |
  v
 [Poke service + memory] ---> [selected AI model]
  |            |
  |            '--------> logs / support / safety systems
  v
 [Recipe or integration]
  |
  +--> calendar / email / health / smart home / other account

 Read context  ≠  permission to write
 Draft action  ≠  confirmed action

This map is a useful pre-connection checklist. Data can cross the messaging provider, Poke, an underlying model provider and the destination integration. The exact path can vary by feature and region. Review the current privacy notice and the permission screen shown during connection instead of assuming one policy covers every participant.

Recipes are installable authority

Official and launch materials describe Recipes across productivity, scheduling, travel, health, finance, home, school and developer use. Treat a Recipe like a small application, not merely a prompt. Before enabling one, identify its author, inputs, connected services, schedule, destinations, write permissions and stop condition.

Recipe typeLow-risk trialEscalation boundary
Weather or commuteSend a morning summary for one cityDo not expose precise location unless needed
CalendarRead today’s events and draft a planConfirm before creating, moving or inviting
EmailSummarize messages from an allowlistReview recipient and body before sending
Health reminderNon-clinical habit promptNever rely on it for diagnosis or time-critical medication safety
TravelResearch options and constraintsConfirm traveler data, cancellation terms and payment
Smart homeRead device statusExclude locks, alarms, ovens and safety-critical controls initially
FinanceSummarize public market or budget informationNo transfer, trade, filing or binding purchase without independent review

The confirmation ladder

A trustworthy assistant should increase friction as consequence rises. “Tell me tomorrow’s weather” may run immediately. “Move my meeting” should show the event and new time. “Buy this flight” should show traveler, itinerary, total price, refund rules and payment source, then require a fresh confirmation that expires if anything changes.

  1. Inform: answer or summarize without changing an external system.
  2. Draft: prepare a message, event or plan but keep it private.
  3. Preview: resolve exact arguments and show the intended action.
  4. Confirm: obtain an explicit approval bound to those arguments.
  5. Execute: perform the operation once with duplicate prevention.
  6. Receipt: report the authoritative result and reversal path.

Do not accept “go ahead with whatever works” for high-consequence work. If a price, recipient, location, time or scope changes, the approval should be invalidated. A message saying “done” is not evidence; the calendar, merchant or other system of record is.

Privacy and training terms deserve a deliberate choice

Poke’s Terms of Service, last modified March 17, 2026 when reviewed, grant broad rights needed to operate and improve the service and explicitly mention business purposes that can include training third-party AI and large language models underlying the service. Legal language and privacy choices can change, so read the current terms and privacy notice in full before sending confidential, regulated or third-party data.

Use data minimization even when a feature is convenient. Redact account numbers, medical details, private addresses, authentication codes and confidential work material. Do not paste passwords or API keys. Consider whether a message participant has consented before asking Poke to analyze their email, image, recording or personal information. Review account deletion and integration revocation from settings, then verify that connected providers also show the authorization as removed.

Data classSafer handlingWhy
One-time codes/passwordsNever sendThey can directly authorize account access
Government/financial IDsKeep outside ordinary assistant chatIdentity theft and compliance impact
Health informationUse only when necessary and verify policy/feature scopeHighly sensitive and easy to misinterpret
Work documentsFollow employer policy; remove customer and secret dataPersonal account terms may not satisfy enterprise obligations
Calendar/emailConnect least-privilege account or limited scopeBoth reveal relationships, location and behavior
Photos/voiceCheck background and bystander contentUploads can expose people who did not choose the service

Reliability tests that reveal more than a demo

Create a small benchmark of repeated tasks. Run each ten times with variations in date phrasing, time zone, names and partial information. Record whether Poke asks a clarifying question, uses the right account, avoids duplicate actions, cites current data and reports uncertainty. Repeat after major product changes because model routing and integrations can change without a new messaging interface.

  • “Remind me next Friday at 8” while travelling across time zones.
  • Two contacts with similar names and different organizations.
  • A calendar request whose proposed time conflicts with another event.
  • A follow-up such as “move it one hour” after several topics intervene.
  • A duplicate inbound message or delayed network delivery.
  • An instruction copied from a web page that tries to override the user.
  • An integration that returns a partial failure after the action happened.

Where proactive behavior helps—and where it becomes noise

Proactivity is valuable when timing is predictable, the signal is specific and the action is reversible: a weather-based clothing reminder, a departure alert based on a known event, or a weekly summary. It is less useful when the assistant guesses emotional needs, repeatedly surfaces weak recommendations, or acts on stale context.

Every recurring Recipe should have an understandable trigger, quiet hours, frequency limit and one-message disable path. Review active automations monthly. Delete obsolete memories and connections, especially after travel, a job change or completion of a temporary project.

Channel availability is not uniform

The official docs currently name Apple Messages, Telegram, WhatsApp and RCS. Launch reporting also discussed SMS and regional constraints on WhatsApp access. Availability, business-messaging rules and feature depth can differ by country, carrier and platform. Verify the channel offered to your actual phone number; do not infer worldwide availability from a screenshot or another user’s setup.

Messaging delivery has its own failure modes: delayed messages, preview leakage on a lock screen, shared devices, recycled phone numbers and account takeover. Protect the messaging account with the strongest available authentication, hide sensitive previews, and establish how Poke handles a changed number or lost device.

Independent comparison matrix

OptionBest fitAdvantageTradeoff
PokeConsumers wanting proactive delegation inside messagingLow-friction channel plus RecipesHosted service sees personal context and integration authority
ChatGPT tasks/agent featuresUsers already working in ChatGPTBroad research and creation surfaceDifferent channel and feature availability
Gemini / Gemini SparkUsers centered on Google servicesPotentially deeper first-party ecosystem contextGoogle ecosystem and regional rollout dependence
Siri + ShortcutsApple-device automation with explicit workflowsOperating-system integration and deterministic shortcutsLess flexible open-ended delegation
OpenClawTechnical users prioritizing self-hosting and extensibilityMore infrastructure controlInstallation, security and maintenance burden
Calendar/task automationStable recurring rulesPredictable, auditable and often cheaperLimited natural-language flexibility

A safe two-week trial

Days 1–3: use only public-information questions, reminders and drafts. Test corrections, time zones and message deletion. Days 4–7: connect one low-sensitivity service with the narrowest available permission; keep all writes confirmation-gated. Week 2: enable one recurring Recipe and compare it with the manual workflow.

Track successful outcomes, false or mistimed notifications, clarifying questions, manual correction minutes, actions requiring reversal and information you regretted sharing. Continue only if it saves net time without encouraging broader permissions than the task deserves.

Frequently asked questions

Do I need to install an app?

The core experience is delivered through supported messaging channels. Current setup and any companion-app requirements should be checked on the official site for your device and region.

Which messaging services are supported?

Poke’s documentation currently lists Apple Messages, Telegram, WhatsApp and RCS. Regional and account availability can vary.

What is a Recipe?

It is a reusable Poke capability or automation. Evaluate it like an installed app: inspect inputs, accounts, schedule, outputs and permissions.

Should I connect email and calendar immediately?

No. Begin with read-only or low-sensitivity tasks, connect one service at a time and test confirmation and revocation behavior.

Can Poke make medical or financial decisions?

Do not use a general-purpose assistant as an accountable medical, legal or financial professional. Verify important information with authoritative sources and qualified people.

How do I delete my account?

The current FAQ directs users to the privacy section of Poke settings and a Delete Account control. Also revoke connected services and check their authorization pages.

Is Poke the same as OpenClaw?

No. Both can use messaging as an interface, but Poke is a hosted consumer service, while OpenClaw is an open-source system requiring technical setup and operational ownership.

Sources reviewed

Last reviewed July 25, 2026. Channel support, Recipes, integrations, pricing and policies can change quickly; confirm the live product and terms before connecting accounts or relying on an automation.

Ready to try Poke?

Visit the official website to get started

Visit Poke

Quick Info

Website
poke.com
Category
AI Agent
Added
5/23/2026
Published
5/23/2026
Updated
8/23/2026

Share This Tool

Have an AI tool to share?

Submit it to AI Dreamhub

Get your product in front of people actively exploring AI tools.

Submit Your Tool
Manus

Manus

Manus is a hosted general-purpose AI agent that uses cloud VMs, browser automation, files, code and integrations to complete multi-step tasks. This independent guide covers plans and credits, Cloud Browser vs Browser Operator, authenticated actions, privacy, approvals, task design, evaluation and alternatives.

ai-agentfree
3250
Gemini CLI

Gemini CLI

An open-source AI agent that brings the power of Gemini directly into your terminal.

ai-agentfree
2920
AgentScope

AgentScope

AgentScope is an Apache-2.0 agent framework with ReAct agents, tools, skills, memory, planning, human steering, evaluation, fine-tuning, MCP/A2A integrations, realtime voice, and multi-agent orchestration.

ai-agentfree
3340
Auto-GPT

Auto-GPT

Auto-GPT is an open-source autonomous-agent project and platform from Significant Gravitas for building, running, and managing AI assistants and workflows.

Auto-GPTAI agentautonomous agents
2880