Merciful, Wherever You Work

New clients, a developer API, and a small widget for your own corner of the web.

October 21, 2026 · updated November 11 · by the Merciful team

The app is in beta. Expect some rough edges. Use Feedback in the app or send us a report about what worked and what needs attention.

A chat box is a useful place to start. You can untangle an idea, draft a difficult email, or ask why a piece of code keeps breaking. Eventually, though, the work needs somewhere to go: into a document, a project folder, a website, or a tool you are building yourself.

Today we are opening up more of those paths, with three pieces you can use independently:

Our hosted client uses the API, and the widget is another way to use that same conversation service. Each has its own starting point, so you can choose the part that is useful to you.

The idea running through all of it is simple: useful software should fit into the work you already do, on the machine you already have. You should be able to understand what it is doing and decide what happens next.

From a conversation to something you can use

In our post about Vesa, we described the tool loop we use to maintain Merciful: inspect a file, make a change, read the result, and try again when the first idea fails. Giving a model that feedback makes a substantial difference. It can work through a task instead of leaving you to copy commands between windows.

The packaged Merciful clients bring that approach into a workspace. Ask the desktop or mobile client to make a small Pong game with a score counter and keyboard controls. It can create the files, inspect them, and revise the result when you ask for a slower ball or larger text. You can read the source and try the game yourself. The model still needs your judgment about whether the thing it made is any good.

Or start with a folder of project documents: ask for a summary of the decisions, open questions, and next steps, with references to the source files. Read those references. Correct the interpretation. Keep the useful parts. A summary earns its place by helping you understand the work.

Use the machine you have

The new download site brings the clients together, with a clear page for each way of using Merciful:

The legacy build comes from the same impulse we discussed in Fairness and AI. Access depends on ordinary things: the laptop available at home, a school computer, the money left after the bills. Extending the useful life of a machine matters. The legacy client uses an old runtime, so its download page includes installation notes and system requirements.

You can also keep using Merciful in a browser, including on a phone or tablet. The web version keeps the familiar chat experience; the desktop, mobile and Chrome clients add a file workspace. Account-backed rooms and device-local files have different homes: signing into another client does not automatically move your local workspace across with you.

Your workspace, your choice of model

Use Merciful's hosted models, or explore the beta's BYOM settings for a local Ollama instance or an OpenAI-compatible endpoint. The workspace gives you a consistent place to work while you choose the backend that suits your needs.

That choice also determines where model requests go. Files stored locally can still become part of a request when you ask the assistant to read them. A hosted provider receives the context sent to it; a local endpoint has its own configuration and storage practices. We want those boundaries to be understandable before you put sensitive work into a conversation.

The beta provides a set of file tools and JavaScript commands for working inside the app's workspace. Review the changes, open the files, and try the result. Chrome and legacy clients keep their files in local browser storage; desktop and mobile clients also have their own local workspace folders.

Agent workspace rooms stay tied to their workspace, while shared rooms remain ordinary chat. Reading a transcript on another device does not give that device permission to run its commands or access the original files.

An API for the things you build

The developer documentation opens up Merciful's conversation service to your own applications. Create a room, submit a prompt as a job, and retrieve the reply when it is ready. Room history provides conversational continuity; your integration supplies the task instructions and reference material it needs.

This gives you room to make something small and specific: a writing exercise, a study tool, a custom interface for your team. Server-side integrations can use account API keys. Public browser embeds can start anonymously, without exposing an account key in a page's source.

The API provides queued model responses. The client workspace supplies its own file tools and command loop. An API request does not automatically run commands or gain access to somebody's local files. Managed document collections and shared context mounts remain future work; include the reference material you need in the request.

A small widget for your own corner of the web

If you want to add a conversation to a blog or personal site, the Merciful widget gives you a starting point: a stylesheet, a script, and a few settings for the title, placement, and colors. It uses the same API described in the docs, so you can read how it works and build your own interface if you prefer.

You can seed a fresh conversation with a short, visible first message, such as your opening hours or a few useful facts about the site. It uses the same service instructions and allowance as a visitor's own message. Simply loading the script does not request a model response.

Visitors can talk in the widget and follow its Merciful link to continue in the full app. A useful conversation should survive the little box it began in. The welcome message also makes the service behind the widget visible to the person using it.

Adding the widget does not automatically teach it your website or connect it to your private workspace. A title saying “Ask about my articles” needs actual article context behind it. We would rather make that requirement clear than let a convincing answer stand in for knowledge the model was never given.

Keep it useful. Keep it sustainable.

More ways to use Merciful also mean more requests reaching real hardware. Free access operates within shared capacity and fair-use limits. Our Pro subscriptions help pay the running costs, as we explained in How Merciful Runs. Opening an API does not make compute free to provide.

Our purpose remains the same: build a practical tool that helps people think, learn, and make things. Human relationships belong with people. Decisions about your work belong with you. A successful session can be a short one that leaves you ready to get on with the rest of your day.

Find your client at Merciful.app, or start building with the developer docs. Make something small. Read what it produced. Tell us where the tools get in your way. That is how this gets better.

— The Merciful Team

← What We're Building Next Back to Blog