Updates

What shipped, month by month. Every entry is a change you can see in the product today.

Launching at Black Hat USA

Las Vegas. August 2026.

Sovereign AI goes live on the biggest security stage there is. Governed AI, launched where the people who break things go to find out what holds. Mandalay Bay - come see it running. Booth 8308.

17

Days

:

08

Hours

:

38

Min

:

28

Sec

August 2026

Launch month. Everything is new when the platform is: the first public release, live from Black Hat.

Swarms

NEW

Compose Swarm: prompt-first crew composition

Write the brief and press Compose Swarm: the composer reads the brief plus project context and proposes a crew with a stated reason for every seat, count steppers, and suggested optional roles. Templates remain one click away, reorganized as a size ladder: Solo, Pair, Squad, and specialist multi-phase Workflows.

Swarms

NEW

Loop count on the launch modal

A stepper on the proposed crew sets how many build-review cycles the team gets (1 to 10) before the quality gate renders its verdict.

Swarms

NEW

The role census

A deliberate census of 14 roles, organized by what each does to your delivery: Plan, Build, Analyze, Govern, Operate. Broad roles load specializations per task instead of multiplying cards. Every surface that renders a role reads one shared module, so the census cannot drift.

Flows

NEW

AI Flow Editor

Describe the job in a sentence and the editor drafts the graph; ask for a change and it edits the current one. The node palette carries the full vocabulary: Crew, Agent, Set Variable, Decision, Echo, HTTP Request, Loop, Human, Report, Swarm, Trigger.

Flows

NEW

Run traces that read like receipts

Every run records per-node status, HTTP response detail, the branch each decision took, and who approved each human gate with their note on the record. Runs interrupted by a platform restart are recovered and marked honestly instead of hanging forever.

Chat

NEW

Surface switcher

Chat, Build, Flows, and Data now sit in a pill switcher in the top bar: four surfaces, one governed session, no context lost when you move between them.

Chat

NEW

Syntax highlighting that respects your theme

Fenced code in chat and the Workspace highlights with a palette that passes WCAG contrast on every theme, light through black, and flips live when the theme does. A CI guard walks every token color against every background so it can never regress.

Chat

NEW

An honest context meter

The composer context meter accounts for cached tokens end to end, so prompt caching never makes context look shorter than it is.

Administration

NEW

ERP & Finance provider secrets

Provider Secrets gains an ERP & Finance category: NetSuite token-based auth and SAP S/4HANA Cloud communication users, write-only like every other secret. With NetSuite configured, agents gain read-only SuiteQL and record-lookup tools under the same approval modes as everything else.

Administration

NEW

Connected app status chips

Each catalog app shows one chip carrying state and capability together: Enabled with Read / Write, Enabled with Read-only, Setup required, Not enabled, or Failed. Failures surface as a toast with the reason instead of a silent dead integration.

Models

NEW

Auto routing, priced honestly

Auto resolves each request to the cheapest capable model and prints its verdict as a receipt in the picker. Rows without real catalog cost data show no price rather than a misleading zero; self-hosted models remain the one legitimate $0.