Frequently asked questions

On your phone. HITL is an iPhone and Android app, and your agents, their memory and your chats live there, not on a cloud computer we rent for you. When an agent needs to think, it sends that request to the AI provider you choose. Some Android phones can also run an experimental model on the phone itself.

No. Everything runs in the app, so there's nothing to host, install on a server or keep running at home. You only need a computer if you want your desktop agents, such as coding tools, to reach you on your phone.

Yes, for scheduled agents and workflows. On Android they run in a background service that shows in your notification shade. On iPhone, a Live Activity on your Lock Screen shows progress, with Pause and Cancel. Your phone's operating system can delay background work, and on iPhone, setting a schedule up as an iOS Shortcut makes wake-ups more reliable. Because the agents live on your phone, the phone needs to be on. If you're offline, scheduled runs wait for a connection.

Yes, both. On Android, join the beta through Google Play (Android 10 or later). On iPhone (iOS 17 or later), HITL is available through a TestFlight invitation during the beta. Your agents work the same way on both. Each platform also has its own extras: Apple Shortcuts, Siri and Live Activities on iPhone; notification reading and screen analysis on Android.

Your agents, their memory, your chats and your files are stored on your phone. When an agent works on something, your request, plus anything the agent read to answer it (for example a calendar entry or a photo), goes to the AI provider you've chosen. That's the only way it can reply. The first time this happens, the app tells you what's sent and to whom, and asks your permission. We don't sell your data. Backups are made only when you ask for them.

You choose. Every plan includes managed models in Lite, Standard and Pro tiers, so HITL works the moment you sign in. You can also add your own API key from OpenAI, Anthropic, Google or OpenRouter, connect your own self-hosted server, or (on some Android phones) try an experimental on-device model. You can set a model for each agent.

No. Your plan includes models, and the Free plan is enough to try everything. If you already pay a provider, you can add your own key on any plan, including Free. Your usage on that key is billed by the provider and doesn't count against your HITL allowance.

We don't sell your data, and your chats with in-app agents are stored on your phone, not on our servers. To generate replies, your requests go to the AI provider you use, which processes them under its own terms. HITL's default provider is covered by enterprise cloud terms. With your own key, your agreement with that provider applies.

In your phone's secure storage: the iOS Keychain or the Android Keystore. They aren't synced to our servers. The app sends a key only to its own provider, and only to make the requests you ask for. Never paste a key into a chat. Add it in Model Providers & Secrets instead.

Yes. Agent Data shows what each agent has stored, and you can delete one item or clear a whole area. Delete Account (Profile → Account) removes your account and synced data. On-device data is removed when you delete the app.

No. Texts and emails are drafted for you to send yourself, and calls start only when you confirm. Every tool an agent can use has a trust level:

• Free for read-only lookups;
• Soft Confirm, where the agent pauses with an Approve/Deny card;
• Hard Confirm for sensitive actions such as deleting data or anything involving money.

The agent shows you exactly what it's about to do and waits for your tap.

Because acting for you is a big deal, the default is to ask. Once you trust an agent with a tool, tick Always approve this tool for this agent. You can undo that at any time in Trust & Autonomy. You can also set the trust level of any tool for each agent, so routine steps run without asking and sensitive ones keep a hard stop.

Yes. Profile → Agent Engine & Data → Privacy & Background AI has a master switch that pauses all background processing for every agent immediately. You can also pause a single agent's schedule, or stop a running workflow.

Only what you allow, twice over. First, your phone asks your permission for things like location, calendar, contacts and photos, and you can revoke it in system settings. Second, HITL's trust levels decide whether each tool runs freely, asks first, or needs a hard confirm.

No AI agent today is immune to hidden instructions in content it reads; this is called "prompt injection". HITL's answer is to keep you in the loop. Actions that affect the outside world (sending, deleting, spending) stop for your approval, and you choose which tools each agent can use at all. Review what an installed agent or skill can do before you install it.

The things your day runs on:

• calendar, reminders, contacts and alarms;
• drafting texts and emails, and starting calls you confirm;
• location, including "when I get to…" triggers;
• the camera and text recognition;
• searching your photos;
• browsing the web in an in-app browser;
• Apple Shortcuts, Siri ("Ask HITL") and Health on iPhone;
• reading notifications and analysing your screen on Android.

Each of these needs your permission first.

Yes. Use your phone's share button in any app, choose Human in the Loop, pick an agent or workflow, add an optional instruction, and tap Send. On Android you can share several files at once. On iPhone it's one link, image or file at a time.

Yes. Every agent can hold a voice conversation. You can turn on the "Hey HITL" wake word, or on iPhone say "Ask HITL" to Siri.

Partly. Agents can open apps that support links into them (deep links) and get things ready there. For example, an agent can prepare a WhatsApp reply to a specific contact, which you then send. On iPhone, agents can also build and run Apple Shortcuts, and on Android they can read your notifications and analyse your screen when you tap. They don't control other apps or tap through their screens for you. Built-in system assistants go further here, because they're part of the operating system.

One assistant juggling every instruction gets confused. With HITL, each agent has one job, such as a morning brief, a receipts tracker or a research helper, with its own instructions, tools, model and schedule. Your Companion helps you build and manage them, and agents can team up in a Squad for bigger jobs.

No coding needed. Tell your Companion what you want handled, in your own words. It checks it understood you, shows a summary to confirm, builds the agent, and offers to run it. If you prefer to do it yourself, Manual Setup gives you full control. Workflows and Quick Actions are in beta.

Yes. The Marketplace has ready-made agents, skills and workflows. Open a listing to see what it does before you tap Install. If you already have a matching agent, you can install a new copy, replace yours, or update it while keeping your own settings.

Yes. Add HITL as an MCP server in Claude Code, Codex CLI, Gemini CLI, Cursor, Windsurf, Zed or any MCP client, or connect an OpenClaw instance. Its questions, notifications and approval requests then arrive in your HITL Inbox. You answer with a tap, by voice or with a file, from anywhere. Optional end-to-end encryption is available.

No. Questions, notifications and approvals that external agents (such as Claude Code, an MCP client or OpenClaw) send to your phone aren't capped by plan during the beta. Plans meter only the model usage of your in-app agents, and active schedules.

Open your Companion (the round button in the middle of the tab bar) and describe something you'd like handled, for example "every weekday at 8, give me my calendar and the top news". It builds the agent, tests it and can put it on a schedule. The Morning Brief is a ready-made example you can set up during onboarding.

Yes. The Free plan includes agents, up to 4 active schedules and a monthly model allowance. Agents are unlimited during the beta. Pro and Max add more model usage and schedules, and their prices are shown in the app for your country. Bringing your own API key works on every plan, including Free.

Lite models cover fast summarization and daily briefs; Standard models power agent planning and tool selection; Pro models handle deep multi-step reasoning. If you bring your own API key, HITL token limits do not apply.

HITL is in public beta: the features are live and ready for daily use, and we're shipping improvements every week. Paid plans are live too. iPhone purchases through TestFlight use Apple's test sandbox and aren't charged; Android purchases through Google Play are real.

HITL is in public beta on Google Play, and on iPhone through TestFlight invitation. Which countries can install it depends on each store's review, so availability can differ by country. If you can't get it where you are, tell us.

Siri is built into your iPhone, and Siri AI (in beta) can act inside apps that support it. HITL works alongside Siri: you can say "Ask HITL" to reach your agents. What HITL adds:

• several specialised agents you create or install;
• agents that run on a schedule on your phone;
• approval levels you set for each tool;
• your choice of AI provider, including your own key or server;
• a place where your other AI agents can ask you questions.

Google is replacing Google Assistant with Gemini on most phones. On Android, Gemini is the built-in assistant and can control phone settings and some apps. Gemini Spark is Google's 24/7 agent; it runs in Google's cloud and needs a paid plan. HITL is different in four ways:

• it's an app where you build a team of agents that run on your phone;
• it works the same way on iPhone and Android;
• it lets you bring your own key or server instead of being tied to one company's models;
• it lets your other AI agents reach you on your phone.

dots and Muse are always-on agents that work on their own computer in the company's cloud, and your phone is where you chat with them. HITL's agents live on your phone. They use your phone's own features, run your schedules there, and keep your chats on your device. You get several specialised agents rather than one, and you choose the AI provider, including your own key. The trade-off is that cloud agents keep working when your phone is off.

Grok Bot gives you several named bots that work on a shared cloud computer, managed from desktop and phone apps. It comes with a Cursor plan, or with a linked SuperGrok or X Premium+ account. HITL's agents run on your phone and use its features directly. You can pick the AI provider or bring your own key; Grok Bot's models are chosen for you.

Hermes Agent and OpenClaw are powerful open-source agents that you host on your own computer or server and chat with through messaging apps. HITL gives you the same freedom over models and multiple agents with no setup or hosting. Everything runs in an app on your phone, with touch-first approvals. Already running OpenClaw or another agent? Connect it to HITL, and its questions and approvals come to your phone.