Bring an Android device
Install Talos, set a dashboard password and connect an authorized emulator or phone through ADB. Add an API key for Claude, OpenAI, Gemini or OpenRouter, or a custom model endpoint.
New in 0.6.0 scheduled tasks · phones over Wi-Fi · results with CSV exportDescribe the task.
Talos · AI agent for Android automation
Describe the task.
Let Talos work through Android.
Talos is a desktop automation agent that controls Android devices connected to your computer through ADB. Give it natural-language goals to automate phones and emulators you own or are authorized to control.
Self-hosted · Windows & Linux · Your AI provider
No card required for the licence trial. AI usage billed separately.
Your goal
“In Contacts, add Alex Morgan with the number 5550100. Confirm the saved contact appears in the list.”
- 01Open the contact form
Success: name and phone fields are visible.
- 02Enter the contact details
Success: the fields show Alex Morgan and 5550100.
- 03Save and check the result
Success: Alex Morgan appears in Contacts.
40 iterations15 minutes$0.50 budget threshold
Illustrative task definition, not a recorded run. The final model call can exceed the budget threshold.
Android emulators & physical phonesGoals + visible success criteriaClaude, OpenAI, Gemini or OpenRouterRun history you can inspect
What is Talos?
An AI operator for your Android workflows.
Repeating the same app journey by hand takes attention. Writing a script for every variation takes maintenance. Talos lets you define the outcome and have an agent choose actions from the current screen.
Your dashboard runs on your own machine, behind a password, and controls Android through ADB. Task context and screen observations go to the AI provider you configure.
Explore Android emulator automation →
How it works
Connect. Describe. Run and review.
Start with one app and one observable result.
Tell the agent what success looks like
Select a device and describe the goal. Add context, optional milestones, limits and, if you want, a model for this task. Choose a one-time or continuous task.
Start the task and inspect the result
Watch the device preview and open the run timeline to see what the agent observed, which actions it took, what it cost and how the run ended.
Key features
From a goal to a run you can inspect.
Define the outcome in plain language
Describe the goal and add optional milestones, each with a visible success criterion, or let the AI suggest the stages.
Act on the interface in front of you
The agent reads UI elements and screenshots, then taps, scrolls, types, opens apps and checks what changed. Before a run it wakes the screen and dismisses an unsecured lock screen.
Run once or keep a workflow going
Repeat successful cycles with a pause, optional working hours and a summary carried into the next cycle. Temporary errors are retried instead of stopping the task.
Understand what happened
Each run keeps observations, actions, results, model explanations, tokens, cost and screenshots, with the run’s log entries one click away. The last 10 runs per task are kept.
Set limits and a daily budget
Cap iterations, runtime and cost per run, and set a daily budget per task and for all tasks. When the daily budget is spent, continuous tasks wait for the next day.
Pick the model for each task
Use Claude, OpenAI, Google Gemini, OpenRouter or a custom HTTP endpoint with your own key. Set the model and effort globally or per task; each run records what it used.
Hear about problems without watching
Get a webhook, Discord, Telegram, ntfy or email message when a task fails, hits a limit or its daily budget, loses its phone or finishes.
Take over from the preview
Tap on the live preview, type text and press Back, Home or Recents to unlock a phone or fix its state. Several tasks can share one device and run one after another.
A dashboard you can leave running
Password-protected, with API keys that never return to the browser, a Diagnostics page with a copyable report and an interface in six languages.
Who is it for?
Less repetition for developers, QA and technical operators.
Use it where the work happens inside an Android interface.
Developers
Prepare app state for a demo
Create a test contact or adjust a visible setting through a goal-driven task, then verify the final screen before you begin.
QA & app teams
Exercise a multi-screen journey
Describe an app flow with visible checkpoints. Review where the agent succeeded or got stuck, with a timeline to investigate.
Technical operators
Repeat a UI-based workflow
Run cycles within working hours, with a daily budget and a notification if the task stops. Keep the task under review and stop it when needed.
Why Talos?
Define intent. Let the agent decide the next action.
Talos can reduce the action scripting needed for goal-driven tasks. Its screen-based decisions also need evaluation on your app. Use it alongside the checks that already work for your team.
| Your workflow | Scripted UI automation | Talos AI agent |
|---|---|---|
| Define the task | Write an action sequence and assertions. | Describe a goal and visible success criteria. |
| Respond to the screen | Use selectors and branches you maintain. | Choose the next action from the current UI and screenshots. |
| Investigate a run | Inspect the logs and captures your framework provides. | Review observations, actions, results, screenshots, tokens and cost in one timeline. |
| Best fit | Exact assertions and repeatable regression gates. | Goal-driven app interaction, setup tasks and exploratory workflow checks. |
| Trade-off | Upfront scripting and maintenance. | Model latency, provider costs and decisions that can vary. |
Before you start
Questions, answered.
What is Talos?
Talos is a self-hosted AI agent for Android emulator and phone automation. You describe a goal, optionally add milestones with visible success criteria, and let the agent observe the interface and choose actions through ADB.
Which emulators and devices can I use?
Talos connects to Android devices exposed through ADB, including Android emulators and USB-debugging-enabled phones. The device must be online and authorized. Before a run, Talos wakes the screen and dismisses a lock screen without a PIN, pattern or password; a secured lock screen has to be unlocked by you. iOS is not supported.
Do I need to write scripts or install Appium?
You can define tasks in natural language without writing an action script. Talos uses ADB, UI observations and screenshots; Appium is not required. Clear goals and observable success criteria still matter, and each workflow needs to be tried on your app.
Which AI models can I use?
Claude (Anthropic), OpenAI, Google Gemini, OpenRouter, or your own HTTP endpoint that implements the Talos protocol. Choose a default model and effort in Settings; each task can use its own model, for example a lower-cost one for a frequent check. OpenRouter models without tool calling are rejected before the first paid call, and models without image input do not receive screenshots.
Is AI usage included in the licence?
No. You use your own API key and the provider bills you directly. Talos estimates costs from the provider’s prices, uses the actual cost reported by OpenRouter and shows today’s total on the dashboard. It does not calculate charges for custom endpoints.
Can I control how long and how much an agent runs?
Set iteration, time and USD budget limits per run, plus a daily budget per task and for all tasks together. The run budget is checked after each model response, so the final call can exceed it. When a daily budget is used up, continuous tasks wait until the next day (server time). You can stop any task from the dashboard.
What data leaves my machine?
The dashboard, task definitions, run history and cost records are stored on your host, behind a dashboard password. The AI provider you choose receives the task context, UI observations and screenshots used for decisions. Licensed installations also contact the licence server, and notifications go to the channels you configure. If you switch on “Help improve Talos” in Settings, Talos also sends a daily anonymous setup report with setup steps and counts, never task or screen content. Self-hosting does not mean all processing happens offline.
Can Talos replace my Android test suite?
Talos can help exercise visible app journeys and collect an inspectable run history. Agent decisions can vary and success is evaluated from the screen. Keep deterministic tests for exact assertions, repeatable regression checks and release gates.
Can it repeat a task automatically?
Yes. Continuous tasks repeat successful cycles after a configurable pause, optionally only within working hours, and pass a summary into the next cycle. Temporary problems are retried up to four times with a growing pause. The summary is model-generated context, not an exact event ledger; review actions that must not be duplicated.
What if the licence server is unreachable?
A running installation keeps working for up to 72 hours after its last successful licence check. Stopping tasks always works, even when the licence is not active.
Start with one workflow
Give your next repetitive Android task to Talos.
Try it on your own emulator or phone. Bring an ADB connection, a clear goal and your model provider.
No card required for the licence trial. AI usage is billed separately.