Someone asked me a direct question a few days ago.
You say the software is free, so how do you make money? Are you going to charge per device later?
I understand the question. Free has burned people before: a crippled free tier that nags you to upgrade, or three months free followed by a subscription.
So this piece avoids the vague version and separates two things clearly: how far free reaches, and where money does get spent.
What free actually covers
The iEasyRun software itself is completely free. No paywall, no charge per device, no charge per task. Install it locally following the installation selection guide and you have the full feature set.
How it runs is what makes that possible. The software sits on your own computer, consuming none of our server capacity, so there is no monthly bill to justify. Task records, screenshots and assets stay on your machine, so there is no storage fee either.
The only thing that costs money is the thing we do not provide: the model.
Which operations need a model
Add and save a model, and only then can conversation drive the phone properly. That line appears in the installation tutorial and it is a hard prerequisite.
Two things need it.
AI conversation. When you describe a task in plain language, something has to understand what you mean, work out the order of steps, and decide how each one is done. That comprehension and planning goes through the model.
Having AI write a flow. Asking AI to generate or edit steps on the workflow canvas also goes through the conversation model.
Nothing else does.
Which operations do not
This part matters more.
Running saved workflows does not. Per the official technical documentation, trial runs and saved workflow runs use a local step engine. They do not call the conversation model, consume no conversation tokens, and need no model API connected.
Scheduled tasks do not either. A schedule triggers the same execution chain as a manual run, with no conversation tokens consumed.
That rule decides your cost profile. Think of it as paying once to teach the flow, then running it for free. The sooner and more completely you freeze flows into templates, the lower the running cost.
Conversely, if you drive phones through conversation every day and re-describe the task each time, you are paying model fees on every run and waiting for planning each time.
Two model types, completely different jobs
On the capabilities and models page, models fall into two categories.
| Type | Job | When it runs |
|---|---|---|
| Conversation model | Understands your words, plans tasks, generates steps | AI conversation, AI-generated or AI-edited flows |
| VLM | Reads screen captures to decide where to tap | Whenever the next step depends on reading the screen |
The conversation model thinks, the VLM looks. They can come from different providers, and the interface offers models such as qwen. You supply the API key and endpoint you have registered.
One note: model API costs are paid to the provider you choose, not to us. You are billed for what you use, at rates the provider sets.
Three ways to bring the cost down
Turn anything repeatable into a workflow. This is the most effective one. Once a flow is a template, reruns generate no conversation fees. Running the same task three times in conversation today versus building the template once and running it thirty times is a difference of tens of times.
Talk complex tasks through before running them. Trial and error is where the money actually goes. One vague instruction that goes wrong, a re-description, another wrong run, another edit: those rounds add up to more than thinking it through once. Refinement mode is the cheaper path.
Match the model to the task. Simple tasks are fine on a cheaper small model, and only complex planning needs a large one. Not every job needs the most expensive option.
So what does a month cost
Honestly, there is no single number, because it depends entirely on usage. Two reference points help.
If you turn your daily tasks into workflows, day-to-day execution produces almost no conversation cost. You run workflows every day without a single conversation, so the model bill approaches zero. You only spend a little occasionally when writing a new flow, and per-use that amount is small.
If you drive phones heavily through conversation every day, the cost depends on two things: your model price and how complex the tasks are. The same task can differ several times over between models.
Which is why my suggestion is simple: do not calculate it, use it. Take one task you do daily, run it once in conversation, and look at what it consumed. Then turn it into a workflow and run it ten times, and look again. Comparing those two numbers gives you your own answer.
Back to the opening question. Free is not a marketing line here. It follows from how the product runs, since a local setup puts no server cost on us. The only place money goes is the model, and that switch is in your hands. Turn flows into templates and you can mostly leave it off.
Install it and try one small task and the number will be more accurate than any cost estimate you read.
Frequently asked questions
- Is the software genuinely free, with no hidden charges?
- The iEasyRun software itself is completely free, with no paywall and no charge per device or per task. It runs on your own computer, so there is no trial period, and no threshold after which you have to upgrade.
- So where does money go?
- To the language model. iEasyRun does not sell models, so you supply your own API key and pay your chosen provider directly for inference. That is the only recurring cost, and it is usage-based.
- Can I use it without configuring a model?
- It depends on the task. Running saved workflows needs no model and no model API call. But using Chinese conversation to have AI plan a task, or having AI generate and edit flows, requires a model configured on the capabilities page first.
- Do workflows consume tokens?
- Per the official technical documentation, trial runs and saved workflow runs go through a local step engine. They do not call the conversation model, do not consume conversation tokens, and do not require a model API. Only AI conversation and AI-generated or AI-edited flows use the conversation model.
- Are the conversation model and the VLM two different things?
- Yes, and they do different jobs. The conversation model understands what you say, plans the task and generates steps. The VLM reads screen captures and helps decide where to tap. They can come from different providers and are configured in the model section of the capabilities page.
- How do I choose a cheaper model?
- Two approaches. Turn recurring work into workflows, since running a workflow consumes no tokens and you pay once when creating it. And for complex tasks, talk the requirement through in refinement mode before executing, which cuts the trial-and-error cycles where the real spend happens.
- What does a month cost?
- There is no single number, because usage varies so much. If you turn everything into workflows, daily execution costs almost nothing and you only spend a small amount occasionally when writing a new flow. If you drive phones through conversation heavily every day, it depends on your model price and task complexity.
- Do screenshots add cost?
- It depends on the task. Anything requiring a VLM to read the screen calls a vision model, and that is billed by image. Tasks that do not depend on reading the screen, such as pure tapping and typing, incur no model cost from screenshots.
Work out the numbers first
The software is free, install it and see what you use
Runs locally with no server to buy. Bring your own model API key and pay only for what you use.