Skip to content
Agent Labora

By Cloud Granite Ltd

The platform

Build → Connect → Permit → Deploy → Observe.

Five phases, 5 route subtrees and 30 capability pages. The order below is the order the application's own navigation uses, so what you learn here you recognise after signing in.

1

Build

Describe the job. Review what it becomes.

2

Connect

Give it capabilities, not credentials.

3

Permit

Allow, approve or never. Default never.

4

Deploy

It runs, inside its limits.

5

Observe

See what it did, and why.

The nine areas

Everything an agent needs, and everything that bounds it.

Roughly half of this list is capability and half is constraint, which is an accurate reflection of where the work is.

Agent Studio

Describe a job; get a configured agent.

Explore

Tools

Named capabilities, not credentials.

Explore

Memory

Context scoped and time-limited.

Explore

Permissions

Allow, approve or never. Default never.

Explore

Identity

Every agent acts as itself.

Explore

Approvals

Humans keep the decisions that matter.

Explore

Agent Runtime

Isolated, limited, retried, resumable.

Explore

Triggers

Events, schedules, people, other agents.

Explore

Observability

Every run, readable in twenty seconds.

Explore

The argument

An AI agent should not have unlimited access to your company.

Two ways to give a model the ability to act. The one on the left is four steps, and it is what most of this category currently does.

The shortcut

An API key with full access, pasted into a prompt.

AI
Database credentials
Admin API
Everything
  • Nobody can tell which agent acted, or whether it was an agent at all.
  • There is no capability it cannot reach, so there is nothing to review.
  • Revoking it breaks every other thing using the same key.
  • The audit log records the key, not the decision.

How this works instead

Eight steps, and the agent cannot skip one. Each is a place you can look, change or revoke.

  1. 1Agent
  2. 2Identity
  3. 3Permissions
  4. 4Approved tools
  5. 5Policy
  6. 6Human approval where required
  7. 7Action
  8. 8Audit trail

In depth

The five phases.

Each is a hub with its own pages — the detail a security review or a finance team will actually ask for.

Agent Studio

Turn a described job into a configured agent.

6 pages

Tools

Named capabilities with defined inputs, outputs and risk.

6 pages

Permissions

Three states per capability, and the default is never.

6 pages

Agent Runtime

Isolated execution, resource limits, retries and durable state.

6 pages

Observability

Run timelines, reasons, metrics and cost — in business terms.

6 pages

On their own

The ideas that make the rest believable.

12 pages, each arguing something a capability page could not. Identity, approvals and the emergency stop are the three a cautious buyer reads first — and every one of them states where it stops.

Agent identityEach agent acts as itself, never as a shared key.Human approvalsAgents ask a named person, with the evidence attached.Emergency stopPause one agent or halt everything, immediately.Agent memoryFive scopes, each permissioned and each with a retention period.Business knowledgeApproved claims only. When it does not know, it says so.TriggersEvents, schedules, people and other agents.Testing and simulationSandbox, dry run, and simulation against real history.Audit trailActor, version, trigger, inputs, policy and result — on every action.Agent directoryOne list of everything acting on your behalf.Humans and agentsSplit responsibility explicitly, not "AI replaces the team".SecurityControls enforced outside the model, not inside the prompt.GovernanceWho granted what, when, and why — kept as you work.

Give AI a job.

Build an agent, choose what it can access, define what it may do, and deploy it to work safely across your business.

Build an agent Build with the API

No code required · Full SDKs and API · Stop any agent instantly · [email protected]