Aiglade — run several AI agents on Windows with no setup

Download

Windows desktop app, and the Android app that drives the same task queue over your local network. Both are free, and neither needs an account to install.

Windows Desktop installer Windows 10 or later Download ↓ Android Mobile app Drives the same task queue over your local network Download ↓

Both are free. No account needed to install.

What "no setup" actually means

The honest comparison is what each tool asks you to do before the first task runs.

Before your first task	Typical agent tool	Aiglade
Create a provider account	One per model you want to use	None
Paste an API key	One per provider	None
Top up a balance	Per provider, before anything runs	Not required to start
Pick a model	A name you have to research	Free / balanced / high-performance
Install agent CLIs	Each one separately	Included

This is possible because the model supply underneath Aiglade is a gateway, not a vendor. Routing, free-quota selection and failover all happen before you open a settings screen. That layer is FreeModel, and it works on its own too.

What you can do with it

Talk to a team, not a tool

Type @agent do this in a chat window. The request becomes a task, gets dispatched, runs, and the result comes back as a message in the same thread.

Run several agents at the same time

Each agent works in an isolated git worktree, so two of them can edit the same repository without colliding. Their output gets merged when you say so.

Stop at the risky parts

An approval gate holds a task at a decision point. Nothing destructive continues until you approve it, and the waiting task shows exactly what it intends to do.

Survive a crash

Task state is written to disk as it goes. If the app or the machine dies mid-run, the queue is still there when it comes back, and finished steps are not repeated.

Keep the work, not just the chat

Tasks, runs and their evidence stay on your machine. A conversation you had three weeks ago is still searchable, with the tasks it produced attached to it.

Hand it off to your phone

An Android app drives the same queue over your local network. Useful when an approval is the only thing standing between a run and the next step.

What it does not do

It is Windows only. The desktop app ships as a Windows installer. There is no macOS or Linux build.
It is not for building AI apps. If you want a drag-and-drop canvas to assemble a chatbot and ship it as a service, a web platform does that better. Aiglade is for driving existing agents on your own machine.
It is not a team product. State lives on one machine. There is no shared workspace, no roles, no admin console.
It is not a cloud service. Nothing runs when your computer is off, unless you explicitly route through a remote tunnel.

Setting it up yourself

If you would rather understand the pieces before installing, these are the questions that come up most. They are written up on the Aiglade site, in more depth than a product page can carry:

Where model credit comes fromA plan, a key you already hold, automatic selection, or an agent you list — four routes, and how they mix Model tiers, and how to pick oneA tier is a group of models, not a model; what each suits and what to do when you cannot tell Attaching a key you already haveThe four failures that come up most, and why none of their error messages names the cause Four command-line agentsClaude Code, Codex, Hermes, OpenClaw — origin, strengths, and what each needs installed first Checking that it actually finishedThree kinds of false success, all of them silent — "no error" proves nothing When a task stalls or failsRunning, waiting for you, or already dead — and what to check before retrying

Installing it and running a first task

The installer is the whole setup. These are the steps that actually happen, in order.

Step 1: Download the Windows installer and run it. It targets Windows 10 or later, and it does not ask for an account.
Step 2: Open the app. It already points at FreeModel, so there is no provider signup, no API key field and no model name to look up before anything can run.
Step 3: Ask for something in a chat window. Type @agent do this and the message becomes a task: queued, dispatched to an agent, run, and its result returned to the same thread.
Step 4: Answer the approval gate when it appears. A task that reaches a destructive step stops there, and it shows you what it intends to do before you let it continue.
Step 5: Optionally pair the Android app, which drives the same queue over your local network. It is most useful when an approval is the only thing standing between a stalled run and the next step.

Why there is no model to pick

A tier in the settings screen is a group of models, not a model, and a gateway is what makes that honest. The public list is 200+ models: 153 chat models, 77% of the list, and the rest split across image, video, speech, embeddings and rerank. An agent needs a chat model, and the chat group is large enough that the question "which model" can be answered per task, on the server, instead of in a settings screen before the first run.

What that arrangement buys is not speed or a better answer. It is the absence of maintenance. When a provider rate-limits, revokes a key or retires a model, the tier you chose does not change and the task still finishes, because routing moves to the next candidate and logs the skip. That substitution is the same one behind the gateway's modality routes[1], and the app relies on it by default.

Aiglade FAQ

Do I need an API key to use Aiglade?

No. The app ships pointing at FreeModel and works without an account. Keys exist for the case where you have your own provider credit and would rather it be spent than the platform's.

What happens to a run if the app or the machine dies?

Task state is written to disk as it goes, so the queue is still there after a restart and steps that already finished are not repeated. This is the same property that lets several agents work at once: state lives on disk, not in a window.

Can I bring my own provider keys instead?

Yes. Attaching a key you already hold is supported, and it is optional — the app runs without one. The four failures that come up most often when attaching a key are written up on the Aiglade site, because none of their error messages names the actual cause.

Is there a macOS or Linux build?

No. The desktop app ships as a Windows installer, and the second build is an Android app that drives the same task queue over your local network. There is no cross-platform desktop build to wait for.

Do two agents running at once collide in the same repository?

Not by default. Each agent works in an isolated git worktree, so two of them can edit the same repository without overwriting each other, and their output is merged when you say so rather than automatically.

Sources

Every third-party figure quoted on this site was read from the vendor documentation linked below, on the date stated. None of these links are affiliate links and none of these vendors paid to be listed. If a number here disagrees with the page it came from, the page is right — tell us and it gets corrected.

Where the percentages come from

Every share quoted on this site is counted from the live model list, not estimated. Divide the two counts in the last column, round to the nearest whole percent, and you get the number in the first.

Share	Of what	Count
77%	are chat models	153 of 199 listed entries
9%	generate images	18 of 199 listed entries
23%	are not chat models	46 of 199 listed entries
3%	are modality routes	6 of 199 listed entries
17%	carry a context length	34 of 199 listed entries

[1] Counted on 29 September 2026 from GET https://freemodel.online/v1/models, which needs no key and no account. Every row uses all 199 listed entries as the denominator, because the only thing the endpoint states for every entry is its modality. It no longer publishes per-model capability flags such as tool calling or reasoning, so this site quotes no shares for them. Fetch the list and recount whenever you want to check the arithmetic.

Install it, and it already works