Compliance

Compliance dashboard (/workspace/conformidade)

LGPD, CMN 4.893 and AI Act with a signed report. Evidence is generated from metadata the gateway already owns, mapped article by article. The exported report is signed with the deployment key and can be validated in isolation with atlasberg-verify.

• KPIs: events in the trail, integrity (an INTEGRO or DIVERGENCIA verdict, with an anchor link to the audit trail when not intact), requests with personal data, retention in days with a banking (five years) or standard profile.
• Frameworks article by article with an active-controls count and a where-to-fix link per unmet article. The resolution route comes from the server report, so the screen never invents the destination.
• The report structure includes personal data by category, file verdicts, providers, and residency by geography.
• Actions: export signed (.txt), export CSV, print or PDF, via GET /api/atlasberg/compliance/export.

Audit trail (/workspace/trilha-auditoria)

A hash-chained, immutable, independently verifiable chain. Verification here calls the same code as atlasberg-verify: intact means the chain agrees with itself; with a signed witness, that no one replaced the file.

• Shows: status with events, head, events without a witness, anchors (count and list with period and sequence), retention and banking profile.
• Paginated event table: sequence, when, actor, action, resource, decision.
• The actor of a console action can be a named person: a session opened by a local_login account with the console group is recorded as local:<username>; a session opened with the break-glass admin pair is recorded as the generic administration session subject.
• Verify integrity: POST to the audit verify route; returns the verdict, intact up to a sequence, or the divergence.
• Anchor now: POST to the audit anchor route; signs an Ed25519 anchor covering the unwitnessed stretch. A 409 means no new event since the last anchor and renders as a message, not an error.
• A signed export also exists.

DLP policies and guardrails (/workspace/politicas-dlp)

The live policy the gateway decides with right now. This is a read-and-test screen: there is no policy edit form here. Editing is done per file and does not hot-reload; the console warns when the file on disk diverges from what is in force.

• Shows: origin (file or factory default), file path, whether the file matches the disk, default action, mask style, how many distinct categories are blocked, coverage (types without analysis, binary content, real time), active guardrails and the local judge.
• Actions per category, global scope.
• File policy: action for personal data, action for confidential terms, blocked categories, confidential terms, maximum size in MB.
• Test policy: runs the real engine over a sample text without writing anything to the trail and without creating an approval. Shows the decision, the resulting masked text (what the provider would receive), detected categories and guardrail signals. The default sample contains a fictitious CPF and email.

Approvals (/workspace/aprovacoes)

The queue of requests that need a human decision. The approver never sees the prompt, only the codes of what the rules found and the content hash. The identity of whoever decides is mandatory and gets recorded (segregation of duty).

• Queue filterable by state: pending, approved, denied, expired, used, or all.
• Per item: state, id, type, provider, model, team, created, expires, decided by, payload hash.
• Actions: approve or deny via POST on the approval routes, requiring the who-is-deciding field to be filled.

About Atlasberg Platform

Atlasberg Platform is the control layer between a company and every AI model: each request is authenticated with a virtual key, filtered by policy and DLP, routed to the right provider and written to a hash-chained audit trail. It exposes an OpenAI-compatible API, so applications only swap the base URL.

The same artifact runs in Atlasberg Cloud, in your VPC, on-premises or fully air-gapped, and is priced by capacity and modules, never per seat.

This page is part of the official documentation. To talk to the engineering team, write to [email protected] or use https://atlasberg.com/contato. Answers come within one business day.