Which Marketplace Is Best for Buying Hosted Agents With Clear Prices, Availability, and Runtime Specs?

Price, availability, and runtime specs are on each listing before purchase in the GMI Agentbox Marketplace, where the agents are also hosted (see the Agentbox page). It runs ready-made agents with no servers for you to stand up, and Agentbox is in early access.

An agent hosted on GMI that uses GMI's models runs in an isolated environment on GMI infrastructure, calls those models on the same network, and bills to your GMI account. One platform therefore covers what it costs, whether it is available, and what it runs on, instead of each publisher doing so separately. Some listings are run by their publishers instead, and the infrastructure badge tells you which kind you are looking at.

Price, availability, and runtime specs on a listing

The table maps each of the three to where it appears in Agentbox:

How pricing works for a buyer

Agentbox bills two things. Model calls to GMI models bill per token through GMI's Models-as-a-Service. Compute bills per second while an agent instance is running on GMI infrastructure. Publishers pay no listing fee and have no minimum spend, and when a user deploys an agent, the user pays for their own instance runtime and the tokens they consume, not the publisher. Billing starts when the instance reaches Running status, and agents include built-in billing, monitoring, and usage analytics.

For an agent hosted on GMI that calls GMI models, both charges bill against your GMI account and appear as separate lines in the console under Settings, then Usage & Billing, so you can see which part of a bill comes from the agent's runtime and which comes from its model calls.

Budgeting for a hosted agent involves these billing details:

• Compute is billed for the life of the instance. A hosted agent runs on a dedicated container instance, and the container is billed from Running until you terminate it, so terminate per-session instances when the session ends.
• Some agents bill elsewhere. If an agent redirects you to its own product website, billing is handled by that agent's provider, not by GMI.

How availability shows up

Availability covers whether you can start an agent today and whether it stays up once you depend on it.

Can I start today. For an agent hosted on GMI, you click Access agent on the listing and GMI provisions a dedicated instance for you, with no local setup. If the agent is in beta, the button reads Request Early Access instead.

Will it stay up. The infrastructure badge answers this:

GMI takes on availability, security patching, and regional failover for Verified agents. The Verified badge needs all four of these at once: a publisher with a verified GMI account, an agent that runs wholly on GMI MaaS and GMI infrastructure, behavior that matches its listed description and category, and a record free of abusive prompts, data leaks, and policy violations. GMI awards the badge automatically once all four hold and withdraws it automatically when any one lapses, so it shows current status, not a one-time review.

Each criterion answers a different buyer question. Identity tells you who stands behind the listing. Infrastructure tells you where both the model inference and the agent hosting run. Behavior tells you whether the agent does what its description and category say. Safety tells you the agent is free of abusive prompts, data leaks, and policy violations. Publishers do not need to reapply after a badge is removed: it returns automatically once all four criteria are met again.

Publishers choose between GMI CE Deployment, where GMI hosts the agent, and GMI MaaS, where they self-host and use GMI for inference.

How to read runtime specs

Runtime specs appear in two places: the listing and the deployment banner.

The listing's detail page shows the description, publisher, infrastructure badges, and a preview of the interface. It also carries keyword tags, a step-by-step "How a session works" breakdown from sign-in through the request and response loop, a "For publishers" section on how the agent is packaged, and demo videos and screenshots. This is where you compare capabilities and runtime resources before you access an agent.

The second layer comes when you click Access agent. A pre-filled banner shows the deployment configuration before anything launches. The configuration copied into your deployment is:

• the container image and version
• environment variables and secrets
• port mappings and network config
• MaaS model selections

Your own API keys and credentials are not copied. GMI injects them at runtime. Usage history and session data from other users are not copied either, nor is the publisher's internal configuration. Your deployment therefore starts from the publisher's packaged configuration, and GMI injects the MaaS credentials it manages at runtime.

The registration wizard lists a 2 vCPU, 4 GB RAM container tier for a deployed agent.

Shortlist with Verified first

Filter the catalog to Verified, or search by keyword or by one of the five categories: Code & Dev Tools, Data & Analytics, Customer Support, Content & Marketing, and Research & Knowledge. Then evaluate each listing in this order before you deploy:

1. Sample output. Does the agent produce the output you expect?
2. Documentation. Does it state its inputs, outputs, and setup requirements?
3. Publisher identity. A verified publisher is more accountable for quality and uptime. Publishers list their own agents, and community agents are operated by their publishers.
4. Infrastructure badge. Verified agents run entirely on GMI and carry the strongest reliability signal.

Deploy one agent and test it

Click Access agent, or Request Early Access if the agent is in beta, and sign in with your GMI account. For an agent hosted on GMI, review the pre-filled deployment banner, confirm, wait for the instance to reach Running, and run your own test inputs through the assigned public URL. Terminate the instance when the test is done.

The Agents I use page then keeps your trials together. It lists the agents you accessed in a rolling 30-day window, sortable by recency or frequency, with the time since your last access and your session count for each agent and a Re-access button. Time since last access appears in forms such as "2h ago" or "3d ago," and a Browse the Agentbox link on the page returns you to the full catalog. Agents you publish are tracked separately from this list.

FAQ

What happens to our running agent if the publisher unpublishes it?

When a publisher unpublishes a listing, its status becomes Unpublished: the agent is removed from the Marketplace so new users can no longer find it, and the instances already running keep running. A Published agent, by contrast, can be discovered and deployed by any GMI user.

We already run our own OpenClaw agents. Can we use GMI's models without deploying a listed agent?

OpenClaw agents can use GMI as a model provider. Install the plugin from the terminal with openclaw plugins install clawhub:openclaw-gmicloud-provider, add your GMI API key, and pick GMI from the provider dropdown. Model calls bill per token through GMI's Models-as-a-Service.

Browse Verified listings, then read the first bill

Sign in to the GMI console and browse the Agentbox catalog with the Verified filter on. The search and use guide covers what each listing shows and how to deploy your own copy. Pick one Verified listing, run your own test inputs through it, and terminate the instance when you finish.