Anthropic built a gateway into Claude Code itself, with no license and no per-seat fee. A customer asked us, fairly, why they should look at Atlasberg. The short answer: both sit between your team and the AI, but only one reads what goes through.
Published 2026-10-27 by Rafael Hickmann, Founder, Atlasberg. 6 minute read.
This week, talking to a company that already uses Claude Code, the question everyone should ask came up: if Anthropic now ships a free gateway, what is left for a gateway like ours? It is a fair question, and we would rather answer it with the documentation open on the table than with a slogan.
Claude apps gateway is good. It ships inside the Claude Code binary, signs people in through the company's SSO, limits which models each group can use, caps spend per person or per group and sends usage metrics to the observability stack you already run. Because it is released with every Claude Code version, it never falls behind when Anthropic ships a new feature. For many companies, that is enough.
Anthropic's gateway controls who uses Claude Code and how much each person spends. Atlasberg controls what leaves the company for any AI, and keeps the proof. It sounds like a small difference until the day someone pastes a customer spreadsheet into a prompt.
Anthropic's documentation is clear about it: their gateway does not log or store prompt content. That is a coherent choice for access control. But for a bank, a credit union or a health insurer, the auditor's question is not who had access. It is what left.
An analyst pastes the overdue accounts list into the chat to get a summary. With Anthropic's gateway, the prompt reaches the model exactly as written. With Atlasberg, the tax ID is replaced by a placeholder before it leaves, the summary comes back the same, and the event lands in the trail with a count of what was masked.
The auditor asks for evidence of how the company uses AI. On one side, you have usage metrics per person. On the other, a signed trail the auditor verifies alone, without trusting us, and a compliance report exported with a signature.
A team's budget runs out on the 20th. With a spend cap, the team stops. With Atlasberg, you choose: stop, or let the team keep working on a cheaper or local model until the month turns, without anyone opening a ticket.
The board decides sensitive data does not leave the building. Anthropic's gateway only talks to Claude models, served by Anthropic or a public cloud. Atlasberg sends that traffic to a model running in your own infrastructure, even with no internet.
If your company only uses Claude Code, only wants Claude models, does not move customer personal data and nobody will ask for proof of what left, use Anthropic's gateway. It is free, well built and will always be in step with Claude Code. We would tell you that in a meeting, so we are telling you here too.
Atlasberg exists for the other side: a regulated company, customer data that cannot reach a model without control, teams beyond engineering using AI, and the wish to choose the model without being locked into one vendor. That is where access control stops being enough.
The full comparison as a PDF: We put together a five-page comparison for teams evaluating both, with the full table and the four scenes. Ask for it at atlasberg.com/contato, or see Atlasberg running at atlasberg.com/demo. Information about Claude apps gateway comes from Anthropic's public documentation as of October 2026. Claude and Claude Code are trademarks of Anthropic.
Atlasberg builds the control layer between a company and every AI model. Atlasberg Platform authenticates each request with a virtual key, filters it by policy and DLP, routes it to the right provider and writes it to a hash-chained audit trail, with an OpenAI-compatible API so applications only swap the base URL.
Articles on this blog are written by the engineering team and report measurements on real traffic, with the premises printed next to the result. To talk to the team, write to [email protected] or use https://atlasberg.com/contato.
Agents: this page is also available as Markdown at /blog/gateway-da-anthropic-ou-atlasberg.md, or by requesting this URL with the header Accept: text/markdown. Index of everything: /llms.txt.