Part 13: Setting up five trial environments: what you need before you build

Part 13 · One Agent, Five Platforms

Nobody writes about the part before the build: accounts, entitlements, regions, model access and trial limits. It's also where a comparison can quietly go wrong, because a trial that leaves out the agent product isn't a fair test of the platform.

This part is the checklist I'm using for all five platforms. Like Part 11, it's documentation-only. Everything below comes from each vendor's own trial, pricing or getting-started pages as I read them on 1 November 2026, linked in the text. Trial terms change often, so check them on the day you sign up.

The question that matters most

For each platform, the first question isn't "is there a free trial?" It's "does the trial include the agent product?" A platform trial and an agent-product trial are often not the same thing.

That turned out to be the most useful thing to check on every platform below.

MuleSoft

The Agent Fabric getting-started page lists what you need:

• An Anypoint Platform account, with Agent Fabric enabled for your organisation by an administrator.
• Omni Gateway, CloudHub 2.0 (every organisation has a shared space) and Exchange.
• Anypoint Code Builder, plus Java 17 and Maven for custom MCP servers or A2A apps.
• API keys for your LLM providers, and an identity provider that issues JWTs for some features.

Anypoint has a free-trial sign-up. The page doesn't say whether Agent Fabric is included in a trial, and says to contact your administrator or MuleSoft account team if it isn't available. Check that first.

Boomi

The Boomi AI overview says Agentstudio is generally available to direct customers, and its Base Edition comes with every Boomi Enterprise Platform edition. The only trial it mentions is a 30-day free trial of Personal Edition. It doesn't spell out which Agentstudio features that edition includes. Use of Boomi AI is covered by Boomi's AI acceptable use policy.

SAP

SAP has two relevant trials, and they cover different things:

• The generative AI hub trial is 30 days. It gives access to SAP AI Launchpad and SAP AI Core, with models from several providers, including Anthropic, Google and OpenAI. Prompt optimisation isn't included.
• The SAP Build trial is a 30-day basic trial on a shared tenant with sample data. It includes a workbook for creating Joule skills with Joule Studio, but it's "a selected subset" of SAP Build, and Joule skills created there can't be deployed to production. The page doesn't name the agent builder.

For this series, the generative AI hub trial covers model access. Whether the agent-building side can be tested properly in a trial needs checking.

Azure

The Foundry Agent Service pricing page says there's no additional charge for creating or running Foundry-native agents with prompts and workflows. You pay for the models, the tools (such as file search and code interpreter) and, for hosted agents, the container compute. Azure's general free account offers $200 of credit for 30 days. That's an Azure-wide offer, not specific to Foundry.

AWS

The AWS Free Tier FAQ says new customers get up to $200 in credits on a free plan that lasts up to six months or until the credits run out. When it ends, the account is suspended, data is kept for 90 days, and the account is erased if you don't upgrade. The FAQ doesn't list Bedrock among the credit-eligible services, so confirm that on the Free Tier page.

AgentCore pricing is consumption-based, with separate charges for runtime, gateway, identity, memory, evaluations and more. There's no AgentCore-specific free trial; some features are free during preview.

The checklist I'm using

The same eight checks, for every platform:

1. Does the trial include the agent product? Not just the platform.
2. How long does it last, and what happens at the end? Some trials stop; AWS suspends and later deletes.
3. Model access. Bring your own API key, or use the platform's models? Which models?
4. Region. Pick one region and keep it for every component.
5. MCP and A2A support in the trial. The tool contract has to be testable.
6. Spending limits. Set a budget alert before the first model call. Credits run out faster than you'd expect with agents.
7. Terms on publishing results. Some trial terms limit publishing benchmarks. As Part 5 said, where they do, I'll score from documentation and build experience only.
8. Data. Only the fictional golden set from Part 10. No real invoices, no client data, ever.

Two rules on top: personal trial accounts, not employer ones, and nothing that connects to a real system of record.

What this changes

The builds stay documentation-only for now, as with MuleSoft. Where a trial does include the agent product, a hands-on build can follow, and the scores will be updated with a note saying what changed.

Boomi is next: hosting the tool contract.

Which platform was easiest to start on for you?

This series describes a reference build on a fictional company. Trial and pricing details come from vendor pages as of 1 November 2026 and change often; check current terms before you sign up.