AGENTIC CONTROL SURFACE

One command surface for the whole platform.

Every action the platform can take, discoverable and callable from one place, by your operators and by the AI agents working for them. Each change is previewed before it runs, allowed only within a credential's permissions, and written to an audit trail.

Start building Book a demo

$t6x whoami --json

identity ops-agent · permissions enforced · tenant acme

$t6x explain refundOrder --check

needs billing:write · you hold billing:read · MISSING billing:write

$t6x api mutate refundOrder --var id=OR-4821 --dry-run

would send refundOrder(id "OR-4821") · nothing was sent

$t6x api mutate refundOrder --var id=OR-4821 --why "SLA credit" --confirm refundOrder

ok · attributed to ops-agent · audit trace appended

ONE SELF-DESCRIBING SURFACE

Discover it, understand it, run it.

The surface describes itself: it lists every available action, shows what each one needs and returns, and runs a fast check of whether your credential's permissions cover it, before you try. That is what makes it safe to hand to an AI agent that has to find its own way, and just as clear for a person at a terminal.

BRING YOUR OWN AI ASSISTANT

Connect your own AI assistant to the surface.

The command surface is not only for your operators and the agents inside the platform. Your own AI assistant can reach it from outside through a standards-based connector, an MCP (Model Context Protocol) endpoint. A coding agent like Claude Code drives it at the command line; a chat assistant connects through the connector. Either way it is the same governed surface, under the same rules as everyone else.

Signs in as itself: The assistant authenticates with your single sign-on, or a scoped key you issue it, and acts under that identity, never a shared secret.
Same rules apply: Preview before running, least-privilege permissions, and confirmation on sensitive changes apply to the assistant exactly as they do to a person.
On the same trail: Every call carries a per-request identity and lands on the same audit trail as the rest of the platform.

GUARDRAILS ON EVERY CHANGE

Preview first. Confirm on purpose.

Anything that modifies data can be previewed first: the surface shows exactly what would be sent and sends nothing. When it is time to run for real, a change in production asks for a business reason and a typed confirmation of the exact action. Nothing bypasses it.

Preview mode: See precisely what a change would do, with nothing sent, before you commit to it.
Business reason: Production changes carry the stated reason they were made, for later review.
Typed confirmation: The most sensitive changes require confirming the exact action by name.

LEAST PRIVILEGE, ENFORCED

Each credential reaches only what it should.

You can hand an agent a narrow credential and trust the boundary. It reaches only the actions it was granted, one tenant's data stays separate from another's, and before acting it can ask a fast question: can this credential run this action, and if not, which permission is missing.

Scoped credentials: A credential reaches only the actions it was granted, and nothing else.
Tenant isolation: One tenant's data stays separate from another's, enforced by the platform.
Permission pre-check: A fast check of whether a credential can run an action, naming the exact permission it lacks.

EVERY RUN, ON THE RECORD

A replayable record of what ran and what came back.

Every command can append a timestamped record of what was run and what returned: an evidence trail you can replay, review, and hand to an auditor. Long and unattended work is handled too, so an agent can keep going without a person watching.

Audit trail: A timestamped, replayable record of each command and its result.
Unattended runs: Sign-in renews on its own, so a long job does not stall waiting for a person.
Wait for state: It can wait until a long job finishes before moving on to the next step.

ATTRIBUTED BY DEFAULT

Nothing an agent runs is anonymous.

A change made through the surface is not a black box. It is attributed to the credential that made it, carries the stated reason it was made, and lands on the same audit trail your people use. You always know who did what, and why.

Attributed: Each change names the credential that ran it, person or agent.
With a reason: Production changes carry the business reason they were made.
One trail: People and agents write to the same record you can review later.

ONE SURFACE, FULLY GOVERNED

Give your operators and agents one surface.

The whole platform, discoverable and callable from one place, with every change previewed, permissioned, and on the record.

Start building Book a demo