Method
How we research our comparisons.
This page explains where our facts come from, how we cite them, and how we keep them up to date. The same rules apply to every comparison, including the ones with Blocks.ai.
Sources
Everything we say about another product comes from a public source. We use:
- The vendor’s documentation, pricing pages, trust center, and release notes
- The vendor’s press releases and blog
- The vendor’s terms, service level agreements, and compliance pages
- Independent news, only for dated events such as a launch
- G2 and Gartner Peer Insights, only on a product’s profile page and never in a comparison table
We don’t use analyst reports, forums, social media posts, or other vendors’ comparison pages.
Footnotes
Each claim about another product has a numbered footnote. The sources list at the end of the page gives each source’s title, publisher, link, and the date we checked it.
We save the exact quote that supports each claim and check it against the source before the page is published.
Claims about Blocks.ai are footnoted the same way, with links to our docs or other pages on blocks.ai.
When we can’t find an answer
If no public source answers a question, the page says “Not publicly documented” and gives the date we checked. We don’t treat a missing answer as a “No.” We say a product doesn’t do something only when the vendor’s own sources say so, and the footnote links to that page.
Shipped features only
We only count features that have shipped. If a feature is in preview or has only been announced, we label it “preview” or “announced,” or leave it out. This applies to Blocks.ai too.
Comparisons with and without Blocks.ai
We make Blocks.ai, one of the 16 options these comparisons cover.
Comparisons between two other products open with a note that we publish them. They don’t pick a winner or give scores. They say when to choose each product, based on the sourced differences, and a short section near the end says where Blocks.ai fits.
Comparisons that include Blocks.ai make our case, and always say when the other product is the better fit.
Alternatives pages, which list the other products by category, follow the same rules.
The 18 criteria
Every comparison table has these 18 rows, in this order.
- What it is and who it’s for
- What the vendor says the product is, and who it’s for.
- Agent registry and discovery
- Whether there is a registry or catalog of agents, what gets listed, and who can discover them.
- Identity and access control
- How an agent is identified, and how it’s decided who may call which agent.
- Ownership, policy, and revocation
- Owners, admin policies, and whether an agent can be disabled or have access revoked.
- Audit log and observability
- Audit logs, tracing, monitoring, and export.
- How agents connect
- Inbound endpoints or outbound only, private networking, and network requirements.
- Agents across organizations
- Whether agents owned by another organization can be brought in or reached, with access the other side controls.
- Frameworks, models, and clouds supported
- Which frameworks, models, and clouds agents can use, and whether agents built elsewhere can be managed.
- Protocol support
- MCP, A2A, and other agent protocols.
- Deployment options and data residency
- SaaS, self-hosted, or hybrid, regions, and data residency.
- Compliance attestations
- Attestations and certifications the source says cover this product.
- Pricing model and public prices
- Pricing model, public list prices, and free tier.
- Time and effort to get running
- Prerequisites, required licenses, and setup steps, as the vendor describes them.
- Maturity
- Generally available, preview, or announced, with dates.
- Support and SLA
- Support plans and uptime commitments.
- Agent building tools
- No-code, low-code, and SDK tools for building agents.
- Model access
- Which AI models are included or supported.
- Integrations and ecosystem
- Connectors, integrations, and marketplace.
Where each one sits
Each comparison opens with a diagram that puts the two products side by side across six layers:
- Build agents
- Host and run agents
- Identity and access
- Registry and governance
- Traffic between agents, tools, and models
- Agents across organizations
In each layer, each product is marked with one of the following, usually with a short label and a footnote to its source:
- “Offered”: the vendor’s own sources say the product does this, and it has shipped.
- “Preview”: the vendor’s own sources list it as preview or beta.
- “Not included”: the vendor’s own documentation says the product doesn’t do this. The label says what to use instead, and the footnote links to that page.
- “Not publicly documented”: we couldn’t find a public source either way, so there’s no label or footnote.
- “You build it”: only for building it yourself (DIY). The label names a piece you build or wire up.
The diagram has no totals and no scores.
Updates and corrections
We re-check every fact at least every 90 days, and each page shows the date it was last checked.
Something wrong or out of date? Tell us and we’ll correct it.