Skip to main content
​

​
Fixed AML screening criteria

In Edit step engines on the workflow edit screen, you can fix a match threshold (75–100, default 85) and date-of-birth matching (AML Search - Person only, default same year) on an AML engine. Fixed values take precedence over what the agent sends, and the values actually used are kept in agentAuditLog.steps[].mcpcalls[].effective_params and artifacts[]. Read the fixed options from workflowActions[].engines[].options in GET /workflows/:workflowId.

​
Generated Files on the analysis Info tab

Analysis Results → Info now has a Generated Files area. Download the evidence PDF the AML engines retrieved for each match; each file carries a one-line description, subject · threshold · date-of-birth basis · screening result · match count. A cap banner appears with more than 50 matches, and a separate banner appears for PDFs that could not be retrieved even after retries.
​

​
AML evidence file response fields

The GET /analyses/:analysisId response adds artifactFailures[], artifacts[].dobMatching, and artifacts[].dobExcluded. Fields were only added; two existing fields changed what their value is based on.
  • artifacts[].verdict — from per screening to per file (matched party). Files from the same screening can differ from party to party. Analyses run earlier keep the per-screening value.
  • artifacts[].matchCount — the count after removing matches excluded by date-of-birth matching.
When an evidence PDF cannot be retrieved because of a temporary error, the server retries up to 3 times and records what it still could not retrieve in artifactFailures[]. This release also documents what has been in the response since 2026-09-22: artifacts[], artifactOmissions[], the download path, and applied_options · effective_params · overridden_param_keys in agentAuditLog.steps[].mcpcalls[].
​

​
Administrator invitations

Omni v3.0.0On Admin Settings you now add an administrator by email alone. Someone with an Omni account is added right away; someone without one gets an invitation email and lands directly in the project after signing up. An invitation link is valid for 7 days, and right after sending you can see its delivery status (awaiting signup, accepted, expired, send failed, and so on). The name field for adding an administrator is gone.

​
Post-sign-in landing by URL

Omni v3.0.0After sign-in you land on the product whose URL you came in from. Come in through https://omni.argosidentity.com and you land on your Omni project (or the project list if you have several); come in through the ID check URL and you land in ID check. Accounts that use both products follow the same rule.

​
ARGOS design system

Omni v3.0.0The dashboard UI now runs on the ARGOS design system. Colors, type, and common components such as buttons, inputs, tables, dialogs, and date pickers have changed; features and menu locations have not. There are no API changes.
​

​
Reasoning follows the workflow language

Omni v2.2.13reasoning in the extractedData envelope is now written in the language the workflow was written in, whatever the language of the documents or the UI. Previously it was always Korean. This applies to the Report tab, the decision-variable detail on the Summary tab, and the GET /analyses/:analysisId response. reasoning on analyses that already completed does not change.
​

​
Reasoning matches the value

Omni v2.2.12A field value and its reasoning could reach opposite conclusions. That is fixed: reasoning is built from the same grounds used to set the value — the analysis result and the field definition. If a field description says what TRUE and FALSE mean, the reasoning follows that definition, and a field with no grounds found does not get a reasoning for the opposite conclusion.
​

​
List items count in Output Completeness

Omni v2.2.11confidence.components.outputCompleteness now counts list (array) fields by the share that is filled. An empty list scores 0 for that item. The weights (40 · 20 · 40) are unchanged, and the change applies to analyses run after it shipped.
​

​
Upload limits changed

Omni v2.2.10The 5-items-per-folder limit is gone, and a 100 MB cumulative upload limit per profile applies instead (exceeding it returns 409 OMNI_2005). A single file is still limited to 10 MB. Supported formats now follow the document processing engine, and an unsupported format, or a file whose extension does not match its content, is rejected at upload with 400 OMNI_2105.

​
Document identity passed to the analysis

Omni v2.2.9Each document passed to the analysis AI now carries its item ID and name and its folder ID and name. reasoning can point to the document a statement came from, and documents with the same name are told apart by item ID. The metadata set when the item was added is passed along too.
​

​
Analysis report as PDF

Omni v2.2.8A completed analysis can be downloaded as a PDF report. It contains the Analysis Score, Summary, Report, Audit, and Info tab contents and the playbook ID and version used, with the policy text, the submitted documents, and the audit trail as appendices. Korean and English are supported; an analysis that has not completed returns 400 OMNI_2208.

​
Re-analysis no longer reads the previous result

Omni v2.2.8When the same profile is analyzed again, the previous analysis result is no longer injected into the AI prompt. Each analysis judges only from the documents and the workflow definition at that moment, so re-running after you edit the workflow reflects the new rules.
​

​
API response contract corrections and dashboard docs

API response contract corrections — the entries below correct places where the previous documentation did not match the actual response. The server behavior did not change.
  • The error response body carries code and message at the top level. It is not shaped {"error": {...}}. Branch on the OMNI_ prefixed code — API overview.
  • The folder ID prefix is fd_.
  • The GET /profiles/:profileId/items/status response is structured {profileId, isRagReady, summary{}, items[]}. It is not flat.
  • The analysis endpoint is POST /analyses/:profileId/analyze, and the body requires targetItems.
  • Pagination differs per resource. Profiles and folders use nextCursor; the analysis list uses the id of the last item as the cursor.
Dashboard
  • A Credit Usage tab was added to the project settings: usage by period, per-category summaries, transaction history, and CSV export — credit usage.
  • The workflow edit screen can now edit playbook actions (order, name, description, engines) and reference notes. The playbook is not regenerated; only playbookVersion increments — editing a workflow.
  • Copy and Delete are available from the workflow list.
​

​
Verdict grounds and response structure

  • Grounds for every verdict — leaf fields in extractedData come back as an envelope with three keys, value · reasoning · sourceStep, so each value arrives with the reason it came out that way — reading analysis results.
  • Reason codes for missing values — a field that received no value is recorded in details[] with a reasonCode (PARENT_MISSING · EMPTY_VALUE · NOT_IN_FINAL_OUTPUT · TOOL_ERROR) and a human-readable message.
  • Output assembly status — systemMetadata.outputFormatting.status tells you whether extractedData follows the output-schema structure (formatted · validation_failed) or the raw structure (skipped · fallback_raw). Analyses from before 2026-08-14 do not have this field.
  • Playbook freezing — an analysis freezes the playbook and output schema as they were at run time and returns them as playbookSnapshot. Editing the workflow afterwards leaves the grounds of past analyses intact.
  • extractionStatus is deprecated. New integrations should use the envelope’s value and details[].
​

​
First release

  • First release of the ARGOS Omni developer documentation
  • API v1 endpoints for managing projects, workflows, profiles, folders, and items
  • A dashboard for policy configuration, engine selection, and the output schema
  • Four workflow templates: KYB, document compliance, address verification, and income verification