QM Agent: Try YC's Harness Free on Agent37

Do not index

An open door to hosted QM: an illustrated key resting on a cloud.

You can try a QM agent for free on Agent37. QM is Y Combinator's open-source agent harness for people and teams, with a web interface and support for Slack. We host it so you can open the app and try a task before deciding whether to run it for your team.

Start at agent37.com/qm. The free option uses your available personal agent seat. It sleeps when idle, wakes when you message it, and has usage limits described below.

We tested the hosted interface with a fictional product launch checklist. This guide includes screenshots from the running instance, the signup steps, and a clear explanation of what free covers.

What is a QM agent?

QM gives people and shared workspaces an AI assistant with context that can survive beyond one conversation. A harness is the software around an AI model that manages the conversation, tools, files, and work environment.

The useful idea is continuity. You can start with a question, refine the result, and keep working in the same environment. For a team, different people and projects can have their own contexts instead of putting every task into one shared chat.

The official QM overview describes its approach to personal and shared work. It also links directly to Agent37 as a third-party hosted version. Y Combinator maintains the upstream project; Agent37 provides this hosting option.

You can inspect the software in the YC Software QM repository. QM is MIT licensed except where otherwise noted. Its design supports different models and agent harnesses, including Pi, OpenCode, Codex, and Claude Code. What you can select in a particular deployment depends on its configuration.

Open hosted QM in your browser

The first stop is the QM page on Agent37. It brings account setup and the launch controls into one place.

Agent37's QM launch page, showing Open QM and Admin panel in an existing workspace.

The screenshot shows our existing workspace. A new visitor sees Get started first; the Open QM button appears once their instance is ready.

Here is the flow:

Open the QM page and sign in. Create an Agent37 account if you need one.

Verify your email if prompted. The free setup requires a verified email address.

Choose Start QM free when your free seat is available. Agent37 sets up the hosted instance.

Click Open QM. This takes you into QM's web interface.

Start a new session and give it one concrete task. A short task with an obvious finished result is easiest to evaluate.

If you already used your free agent allowance, the page explains that you need a plan to run another agent. An existing eligible paid seat can also be used. The free allowance is shared across your personal agents, so switching to QM does not create an extra free seat.

For the first task, describe the output you want: a table, an outline, a draft, or a checklist. Include the information needed to produce it. That gives you a useful result to inspect before adding integrations or a larger workflow.

A real task inside the hosted interface

We opened a fresh session in our running QM instance and sent this prompt:

Create a five-row launch checklist for a fictional team releasing a note-taking app. Columns: task, owner role, done when. Use only made-up information. Keep the answer under 180 words and show the checklist directly in this chat. Do not connect apps, browse, schedule anything, or publish anything.

QM returned a five-row table in the conversation. It covered the app name and icon, core features, a usability review, a launch announcement, and a readiness check. Each row had an owner role and a condition for calling the task complete.

A real QM response to our fictional launch-checklist prompt, with task, owner role, and done-when columns.

Watch a real 28-second hosted QM session: model settings, a fictional ClearNote launch checklist, and a follow-up announcement.

This is a screenshot of the live response. We hid the account sidebar and cropped the view to the prompt and its result. The example uses invented work so readers can see the interface without exposing anyone's private project information.

The result followed the requested structure and stayed short enough to scan. It also gave us something easy to improve: we could ask for clearer acceptance criteria, fewer tasks, or a different set of owner roles.

This particular test demonstrates a working conversation and a structured answer. Publishing an app, delivering files, connecting Slack, and running scheduled work are separate things to test when they matter to your use case.

What the free option includes

You can start with a free hosted QM instance when your free personal agent seat is available. You do not need to pay a hosting subscription just to try that route.

These are the important boundaries:

Item	Free option
Hosting price	$0 for one eligible personal agent
Card for the free seat	No card required
Idle behavior	Sleeps after 15 minutes idle and wakes when you message it
Awake allowance	Up to 10 hours a week
Starting model credit	$0.20 of the default model
Continued model use	Bring your own keys or choose a suitable plan after the starter allowance
Existing free agent	The same allowance cannot fund a second free agent

The free plan is useful for opening QM, asking questions, and deciding whether the workflow fits. Model usage has a cost even when the hosting seat is free. The starter model credit is a small allowance to begin with.

Some tasks also need a separate agent computer to execute commands. The QM setup can offer a card step to unlock those computers with $5 of starter credit. That is a separate Cloud allowance, with subsequent usage charged at the applicable rates. You can evaluate ordinary chat before deciding whether you need it.

Sleeping also matters for unattended work. If your use case depends on a task running while you are away, choose and verify an appropriate always-on setup. Paid personal plans start at $3.99 per month. Review the current personal agent plans and the setup screen before choosing.

Three useful first prompts

A good first task has a small input and a result you can judge. Here are three ways to try QM without setting up company integrations.

Turn rough notes into a work plan

Paste a short list of non-sensitive notes and ask:

Turn these notes into a five-item work plan. For each item, give me the next action, the owner role, and what would count as finished. Mark missing information as a question.

Check whether the output distinguishes the facts you supplied from its suggestions. A polished table is useful only if it represents the underlying work accurately.

Improve a draft

Give it a paragraph you are comfortable using for a trial and ask:

Rewrite this announcement for a reader who has never heard of our product. Keep the factual claims unchanged. Explain the benefit in the first sentence and finish with one clear action.

Compare the revision with your original. Look for missing qualifications, invented benefits, or details it simplified too far. You can give that feedback directly in the same conversation.

Prepare a decision checklist

Start with a choice your team needs to make:

We are choosing between three help-desk tools. Create a comparison template covering shared inboxes, assignment, reporting, migration, and price. Leave unknown facts blank and list what we should verify.

This produces a research structure without pretending that an unresearched answer is current. Add verified information as you collect it, then ask QM to help summarize the tradeoffs.

Moving from a trial to team work

After a useful first session, pick one repeatable job to evaluate. A weekly status summary or a standard research brief is easier to assess than asking an agent to handle an entire department.

Write down the input, the expected output, and who will check the result. Run the same kind of task several times. This helps you see whether the output consistently saves work or still needs substantial rewriting.

QM has controls for files, skills, memory, apps, and scheduled work. Its upstream documentation explains how these pieces fit together. Availability and setup requirements depend on the hosted instance and the tools you authorize.

Slack is supported by QM, but starting a web instance does not automatically connect your Slack workspace. Connecting services and enabling teammate access are additional setup steps. Begin in the browser, then configure the specific connections your workflow needs.

Keep a human review step for customer-facing output while you learn the system. For example, have QM prepare an announcement draft, then review the facts and wording before someone publishes it. That gives the team a practical first workflow and a clear way to assess its value.

For a wider look at hosting choices, see our guide to AI agent hosting.

Hosted QM or your own deployment?

Use Agent37 when your immediate goal is to try QM through a browser and have us run the hosting. The free seat gives you a way to evaluate that experience before choosing a paid setup.

Deploy QM yourself when operating it in your own environment is a requirement and you have someone to maintain it. The official quickstart covers an organization-owned deployment, including cloud infrastructure, administrator access, and verification.

The open-source license lets you use QM without a software license fee. Running your own deployment still involves infrastructure, model usage, configuration, and maintenance. Budget for those separately.

Frequently asked questions

Is this Y Combinator's QM?

Yes. Agent37 hosts the QM project maintained by YC Software. The official QM documentation links to Agent37 as a third-party hosted option.

Can I try QM for free?

Yes, using your available free personal agent seat on Agent37. It sleeps when idle, has a weekly awake allowance, and starts with limited model credit. Extra agent computers and continued model usage have separate requirements.

Do I need a server or a local installation?

You can start through Agent37's QM page and use the hosted web interface. You do not need to deploy the stack locally for that path.

Can I use it with my team immediately?

Start by testing your own session. Teammate sign-in, shared access, and Slack need the relevant configuration. Check those flows before inviting a team to depend on them.

Where do I start?

Open QM on Agent37, sign in, and use Start QM free if your free seat is available. Try the checklist prompt above, inspect the answer, and give QM one follow-up to make it more useful.

QM Agent: Try Y Combinator's Harness Free on Agent37