How our AI works — Doc. AV-HW-01
Plain English,
no black boxes.
How an Avow agent decides, asks permission and records what it did.
Read the Trust Centre
→
ANIMATION SLOT — 1312 × 620
Before / after workflow
Left: manual hand-offs. Right: agent proposes → person approves → logged. Drop Lottie / video / Framer component here.
Fig. 1
Workflow before and after Avow.
§ 02
The loop
Propose, approve,
record.
Every action follows the same three-stage loop.
I.
Propose
The agent drafts the action and explains why, in plain English.
II.
Approve
A named person approves, edits or rejects. Nothing moves before that.
III.
Record
The decision and outcome are logged for audit and export.
§ 03
Guardrails
What our agents
may never do.
The rules.
3.1
No silent payments
Money never moves without a named approver.
3.2
No external messages
Customer emails wait for review.
3.3
No training on your data
Your data never trains shared models.
3.4
Scoped access
Agents only see systems you connect.
Fig. 2
Each proposal shows its sources and reasoning.
§ 04 — Models
Which models we use,
and why.
Models, hosting regions and how we evaluate them, in one page.
§ 05 — See it live
Watch an agent
ask permission.
Work email
name@company.com
Book a demo
→
Agents your risk team will sign off on.
Made in London. Data hosted in the UK and EU.
Avow
© 2026 Avow. Made in London.
Vol. 01
Trust-first AI agents
MENU
How our AI works — Doc. AV-HW-01
Plain English,
no black boxes.
How an Avow agent decides, asks permission and records what it did.
ANIMATION SLOT — 730 × 440
Before / after workflow
Manual hand-offs → agent proposes, person approves, logged. Drop Lottie / video here.
Fig. 1
Workflow before and after Avow.
§ 02
The loop
Propose, approve,
record.
Every action follows the same loop.
I.
Declared
Each agent publishes what it may and may not touch, in plain English.
II.
Approved
Payments, deletions and external messages wait for a named person.
III.
Recorded
A readable log of every action, kept for audit and export.
§ 03
Guardrails
What our agents
may never do.
The rules.
3.1
No silent payments
Money never moves without a named approver.
3.2
No external messages
Customer emails wait for review.
3.3
No training on your data
Your data never trains shared models.
3.4
Scoped access
Agents only see systems you connect.
Fig. 2
Each proposal shows its sources and reasoning.
§ 04 — Models
Which models we use,
and why.
Models, hosting regions and how we evaluate them, in one page.
Agents your risk team will sign off on.
Avow
© 2026 Avow. Made in London.
Vol. 01
Trust-first AI agents
MENU
How our AI works — Doc. AV-HW-01
Plain English,
no black boxes.
How an Avow agent decides, asks permission and records what it did.
ANIMATION SLOT — 350 × 360
Before / after workflow
Manual hand-offs → agent proposes, person approves, logged. Drop Lottie / video here.
Fig. 1
Workflow before and after Avow.
§ 02
The loop
Propose, approve,
record.
Every action follows the same loop.
I.
Declared
Each agent publishes what it may and may not touch, in plain English.
II.
Approved
Payments, deletions and external messages wait for a named person.
III.
Recorded
A readable log of every action, kept for audit and export.
§ 03
Guardrails
What our agents
may never do.
The rules.
3.1
No silent payments
Money never moves without a named approver.
3.2
No external messages
Customer emails wait for review.
3.3
No training on your data
Your data never trains shared models.
3.4
Scoped access
Agents only see systems you connect.
Fig. 2
Each proposal shows its sources and reasoning.
§ 04 — Models
Which models we use,
and why.
Models, hosting regions and how we evaluate them, in one page.
Agents your risk team will sign off on.
Avow
© 2026 Avow. Made in London.