DAYA FOR SECURITY

See and limit what every person and agent can do

Every AI agent works inside a readable permission matrix, a spend limit and an approval a person holds. Each action is on the record, and any run can be stopped, or any agent switched off, at any time.

ACCESS

One set of rules for people and AI agents

For every person and every agent, set what they may create, read, update and delete in each part of the workspace. Set it per person, role, agent, base, channel or app. Anything not allowed is out of reach.

SUGGESTED SETUP
Authority matrix
Per person, role, agent, base, channel and app.
Limited by its parent
A sub-agent never has more access than its parent.
Skills and tools
Granted to each agent, and revoked the same way.
More about Agents & authority →
Authority · Refund tracker
Per base
WHO
CREATE
READ
UPDATE
DELETE
Kate Sullivan
owner
System agent
whole workspace
Finance agent
invoices and refunds
Research agent
read-only
Delegation only flows down
APPROVALS

Nothing an agent builds goes live without a person

A new agent stays in draft until a person approves it. A workflow or skill an agent writes is reviewed by the workspace's system agent, then approved by someone whose role allows it. Changing the hierarchy is itself a proposal, and escalation always ends at a person.

SUGGESTED SETUP
Draft until approved
A new agent does nothing until a person signs it off.
Review before release
Workflows and skills are checked by the system agent, then by a person.
Approval roles
The matrix decides who may approve what.
More about Agents & authority →
Waiting for approval
Agents
3
waiting
12
approved this month
2
sent back for changes
New agent · Visit scheduler
Draft · proposed by the system agent
Waiting for Kate
Workflow · Follow up with silent customers
Built by the support agent · reviewed by the system agent
Waiting
Skill · reconcile.md
Changed by the finance agent · reviewed by the system agent
Approved
Hierarchy · Research agent reports to the system agent
A proposed change
Waiting
An agent cannot release what it builds
SPEND

No agent can outspend its limit

Each agent has a daily and a weekly limit, so a runaway loop stops at the cap. Agent work runs on prepaid credit: at zero it stops, and auto top-up carries a monthly ceiling and a cooldown between charges. Choose which model providers answer, and bring your own keys if needed.

SUGGESTED SETUP
Limits by agent
A daily and a weekly cap for each one.
Prepaid credit
At zero, agent work stops. Nothing is billed later.
Model providers
Hold several at once, with your own keys.
More about Agents & authority →
AI usage and limits
Today
$4,280
credit left
26 days
at the current rate
$3,000
monthly top-up ceiling
System agent
Daily limit 400 credits · weekly 2,000
62%
Finance agent
Daily limit 250 credits · weekly 1,200
38%
Research agent
Daily limit 250 credits · weekly 1,200
12%
Support agent
Daily limit 200 credits · weekly 900
81%
At its limit an agent stops. Nothing is billed later.
AUDIT

See every step an agent took, and what it cost

Account events are recorded once and cannot be edited. Every agent run is saved step by step, with what each step was given, what it returned, which model version it used and what it cost. Anyone in the stream can stop a run, and the steps it had finished stay on the record.

SUGGESTED SETUP
Audit log
Every action, handoff and approval.
Run history
Each step of an agent's work, with its cost.
Stop and pause
Stop a run midway, or switch an agent off.
More about Agents & authority →
Audit log
Today
14:02
Refund £48.00
Finance agent
Allowed
14:05
Edit customer record
Research agent
Blocked
14:09
Approve refund £112.50
Kate Sullivan
Approved
14:12
Send reply
Support agent
Waiting
Written once and never edited
Run stopped by Kate at step 3
The finished steps are still on record. The finance agent stays switched off until Kate turns it back on.
24 CONTROLS

Every control over AI, in one place

Grouped by the question each one answers.

What it can touch

Authority matrix
Create, read, update and delete, in every part of the workspace.
Limited by its parent
A sub-agent never has more access than its parent.
Per base, app and source
Rights are set for each base, mini app and warehouse source.
Skills and tools
Granted to each agent, and revoked the same way.

What it knows and says

Policies
Documents every agent must follow.
Knowledge bases
The files it answers from, with its source named.
Instructions and voice
Its brief and how it should sound, set for each agent.
Silent unless asked
An agent can be told to speak only when a stream asks for it by name.

Who approves

Draft until approved
A new agent does nothing until a person signs it off.
Review before release
Workflows and skills it builds are reviewed, then approved by a person.
Draft and approve
Messages to customers are drafts unless you allow more.
Escalation ends at a person
However far a task is handed up, it stops with a human.

What it can spend

Limits by agent
A daily and a weekly cap, so a runaway loop stops.
Limits by person
What each user may ask for in a cycle.
Prepaid credit
At zero, agent work stops. Nothing is billed later.
Top-up ceiling
A monthly cap and a cooldown between charges.

What is recorded

Audit log
Account events, written once and never edited.
Run history
Every step, with its input, output, model version and cost.
Stop mid-run
Anyone in the stream can stop it, and the record survives.
Pause switch
An agent that is off still has its instructions, but answers nothing.

Where data is stored

Residency
EU, US or APAC, per workspace on Advanced.
Model providers
Choose who answers, and bring your own keys.
Encryption
In transit and at rest.
No training on your content
Workspace content is not used to train models offered to other customers.
START HERE

Four things security teams set up first

BASES

Risk register

Risks, owners and review dates as records.

BASES

Vendor management

Each vendor, what it holds and when it renews.

TEAM COLLABORATION

Security incident stream

One thread per incident, with an owner.

MINI APPS

Access review app

A regular check of who can reach what.

Pick which model providers answer

Hold several at once and bring your own keys.

ClaudeClaude GPTGPT GeminiGemini DeepSeekDeepSeek KimiKimi GLMGLM

Common questions

What stops an agent going too far?

Three things: its permissions limit what it can touch, its spending limit caps what it can cost, and anything that cannot be undone needs approval first.

Does our data leave the workspace?

Only to the places that are connected. Sources sync in, nothing goes out unless that is set up, and on Advanced the region is chosen for each workspace.

Is workspace content used to train models?

No. Workspace content is not used to train models offered to other customers.

OTHER DEPARTMENTS
Sales Marketing Customer service Operations Procurement Supply chain Finance Human resources Project management Engineering IT All departments

Start with security

Free for up to twenty people. Set up this team now, and add the others as needed.