Early access
Give any AI agent a mandate, not a blank cheque. Accountability the rules now expect.
MandateID is a registry of AI agents, with control. A named, verified person authorises an agent to act for them, for a set period and within set limits. Systems check the mandate before they let the agent act, and every step is recorded.
Built for the rules on AI agents. Singapore's IMDA, CSA and MAS expect clear accountability, set limits and a record of what happened. MandateID is built to provide all three.
Annual plans from S$12,000, or pay monthly from S$1,500 · priced per active agent · See how it maps to the rules →
MandateACTIVE
Agentsupport-agent-07
Authorised by[Accountable person]
MayLook up orders, issue refunds
May notExport customer data
Valid[start] → [end]
CeilingS$200 per refund
Illustrative example. Revoke at any time; every change is recorded.
The gap
Signing in proves who you are. It doesn't say what your agent may do.
No evidence of consent
When an agent acts, there is rarely a clear record that a person agreed to it, or to what.
No limits that travel with the agent
Credentials are handed over whole. Scope, duration and limits are left to each system to guess.
No register of who is acting
Few organisations can list every agent acting in their name, who authorised it, and whether it is still allowed.
How a mandate works
Register, authorise, check, end.
01
Register the agent
Each agent gets an identity in the MandateID registry, tied to the verified person or officer accountable for it.
02
Grant a mandate
The accountable person sets what the agent may do, for how long, and up to what limit, then signs it. That signature is the consent record.
03
Check before acting
Before a system lets the agent act, it asks MandateID for the agent's signed mandate status. Large actions wait for a second approver.
04
Expire or revoke
Mandates end on their date, or sooner when the person pulls them. The agent's status changes at once, and the history stays.
Any agent, any task
If an agent acts for a person, it needs a mandate.
Money is only one of the things an agent can get wrong. A mandate sets what the agent may do, what it may touch and how far it can go, whatever the job.
Customer service agent
Look up orders and issue refunds.
Limit: S$200 per refund, 30 days.
Coding agent
Open pull requests and deploy to staging.
Limit: no production access, 7 days.
Data and analytics agent
Query named datasets and produce reports.
Limit: no export of personal data.
Procurement and payments agent
Place orders and pay approved suppliers.
Limit: S$5,000 per order, approved suppliers only.
Recruiting agent
Screen applications and schedule interviews.
Limit: offers need a human sign-off.
Communications agent
Send email on your behalf to listed contacts.
Limit: no new recipients without approval.
Illustrative mandates. The ceiling can be an amount of money, a number of records or actions, or a risk tier, whichever fits the agent.
Pricing
Priced on risk, not on seats.
You pay for each agent that holds a live mandate, and the price depends on the most that agent is allowed to commit. An agent with a S$500 ceiling costs little to stand behind. One that can commit S$500,000 costs more, because the evidence around it matters more. Agents that don't handle money are placed in the same bands, and we are confirming how with early partners.
Highest ceiling on the agentPer active agent, per month
Up to S$10,000S$20
S$10,000 to S$100,000S$50
S$100,000 to S$1 millionS$150
Over S$1 millionS$400 or custom
List prices, as paid on the monthly plan. The annual plan takes 10% off. Deployments of 50 agents or more: talk to us about terms.
Annual plan · 10% off · minimum
S$12,000 per year
Commit for a year and take 10% off the per-agent prices. Annual contracts include a minimum commitment. Your agent fees draw down against it, and a per-agent charge applies above it. A handful of agents will usually sit within the minimum. With 50 agents or more, the 10% still applies and we will agree terms with you.
Monthly plan · minimum
S$1,500 per month
Prefer not to commit for a year? Pay month to month at the list prices, with 30 days' notice. The monthly minimum is higher than the annual plan's, which is the cost of the flexibility.
Design-partner pilot pricing available
We are working with a small number of organisations first, on a 6 to 12 month pilot, at a reduced rate. Help shape the product and the price.
Ask about the pilotEstimate your price
How many of your agents fall in each band?
Start from an example number of agents, then adjust how many sit in each band. Count each agent once, by the highest ceiling on its mandates. If an agent doesn't handle money, ask us which band fits.
Billing
Up to S$10,000
S$20 per agent per month
10
S$10,000 to S$100,000
S$50 per agent per month
10
S$100,000 to S$1 million
S$150 per agent per month
4
Over S$1 million
S$400 or custom per agent per month
1
Agent fees per month at listS$1,700
Agent fees per year at listS$20,400
Annual discount, 10%−S$2,040
Agent fees per year after discountS$18,360
Annual minimumS$12,000
You would pay
S$18,360 per year
Your agent fees are above the annual minimum, so the per-agent fees apply in full.
Examples on the annual plan, after the 10% discount: 6 agents come to the S$12,000 annual minimum. 25 agents come to S$18,360 a year. 100 agents, weighted to high ceilings, come to S$76,140, and we will agree terms with you for deployments of 50 agents or more. On the monthly plan, at list prices, 6 agents come to the S$1,500 monthly minimum, and 25 agents to S$1,700 a month. For the over-S$1 million band this estimate uses S$400 per agent per month.
How the price is worked out
Per active agent
An active agent is one that holds a live mandate. Retired or expired agents are not billed.
Priced by its highest ceiling
Each agent is priced by the highest authorised ceiling on any of its active mandates. A second mandate on the same agent adds no second fee.
Ceiling means total exposure
The most the agent can commit under its mandates in a period: the aggregate cap if there is one, otherwise the per-action limit times the maximum number of actions.
No money involved?
Agents without a dollar ceiling are placed in one of the same bands, based on the sensitivity of the data they can reach and what they can do. We are confirming how with early partners.
Why priced this way
The ceiling is enforced
A mandate's limit is checked every time the agent acts, and anything above it is refused. Understating a ceiling to save money would only restrict your own agent.
Price follows the value of the evidence
A mandate that lets an agent commit S$1 million is the one a compliance team or regulator will ask about, and the record around it matters most.
A predictable bill
The minimum sets your starting cost, and your agent fees draw down against it. You see the price before you sign.
Indicative pricing in Singapore dollars, excluding GST. Figures, the annual discount, the monthly plan, and our approach for agents that don't handle money are starting points that we will confirm with early partners, and may change before general availability. Checking a mandate is free for the system that relies on it, subject to fair-use rate limits. A mandate is authorisation evidence. It is not a guarantee or insurance against what an agent does.
Where MandateID sits
MandateID gives the signed answer. Your systems stay in control of the action.
Is this agent registered? Who answers for it? What may it do, and until when? Any system, whether a payment rail, an application or a data store, can ask MandateID before it lets the agent through.
Person
Verified, accountable, signs the mandate
MandateID
Registers the agent, holds the mandate, records every step
AI agent
Presents its mandate when it acts
Your systems
Check the signed status, then allow or refuse
Designed around open standards for delegation and verifiable credentials, such as W3C Verifiable Credentials and OAuth 2.0 extensions, so mandates can be checked by systems you already run.
Controls and evidence
Built so the record holds up under scrutiny.
Second approver on large actions
Large or high-impact actions need a second person before they are cleared.
Alerts to a verified person
The accountable person is alerted by email or phone, on contact details that have been verified.
Append-only records
Only the server can write a record, and records can be added but never edited or removed.
Checkpoints kept elsewhere
Checkpoints are anchored outside the database, so tampering can be detected.
Kept for 5 years or more
Mandates, approvals and actions are retained for a minimum of five years.
Regulation
What the rules ask, and how MandateID responds.
Regulators are converging on the same few expectations for AI agents: know which agents are acting, name who is accountable, limit what they can do, and keep the evidence. Here is how MandateID is designed against the frameworks that matter most in Singapore.
Model AI Governance Framework for Agentic AI, and CSA Addendum on Securing Agentic AI
IMDA · Jan 2026 · CSA finalised Jun 2026 · VoluntaryWhat it expects
Bound agent risk up front, keep humans accountable, and give agents identity and permissions so their actions can be traced. Agent identity, accountability, a registry and verifiable-credential authentication are named controls.
How MandateID responds
MandateID is the registry. Every agent is bound to an accountable, verified person, and scope, limits and expiry are set in a signed mandate before the agent acts. Credential-based checking is part of the design.
Guidelines on AI Risk Management
MAS · Issued 7 Oct 2026 · Phased from 7 Oct 2027What it expects
Financial institutions stay accountable for AI supplied by third parties. They must keep an inventory of AI use, assess how material each use is, and show human oversight and assurance over vendors. MAS plans to consult on agentic AI in 2027.
How MandateID responds
The registry exports an agent inventory and an oversight log. Scopes and ceilings give a basis for tiering agents by materiality, and signed approvals evidence human oversight. A customer assurance pack, covering security certification, incident notification, sub-processors and data location, is planned.
Safeguards for Agentic Finance at Runtime (SAFR)
MAS white paper · Jul 2026 · Not regulatory guidanceWhat it describes
A runtime checkpoint so that no agent action executes unless it has been declared, authorised and assessed, with records kept.
How MandateID responds
The same model: agents are declared in the registry, authorised by a signed mandate, checked at runtime before they act, and every step is recorded.
Personal Data Protection Act and NRIC guidelines
PDPC · In force · Applies directlyWhat it expects
MandateID handles the identity and contact details of accountable people, so purpose limits, protection, retention limits, breach notification and a data protection officer all apply. NRIC numbers should be collected only where necessary.
How MandateID responds
Privacy by design: collect only what a mandate needs, avoid raw NRIC where a pairwise identifier will do, and keep a breach-response plan. Retention is set deliberately, balancing data-protection limits against the need for evidence.
Electronic Transactions Act
Singapore · In force · Applies directlyWhat it expects
An electronic signature counts if it identifies the signer and their intent and is reliable enough. A secure electronic signature carries a presumption of authenticity. A few things, such as wills, negotiable instruments and dispositions of immovable property, are excluded.
How MandateID responds
Mandate signing is designed to meet the secure-signature test, so a mandate stands as evidence of the person's authority. Excluded instruments can't be covered by a mandate, and we say so.
Also watching: EU AI Act
Record-keeping and human-oversight duties for high-risk systems. A provisional EU deal would move the application dates to December 2027 and August 2028, pending formal adoption.
Also watching: MAS technology and third-party risk guidelines
Financial-institution customers will pass these down by contract: incident-notification windows, sub-processor visibility and data-location commitments.
How to read this. This is our own reading of how MandateID is designed against these frameworks. It is a research summary, not a certification, a regulator's approval or legal advice, and some points are still to be confirmed with counsel. References reviewed October 2026.
Join the MandateID early-access list.
We are working with a small number of organisations first. Tell us what your agents do and who answers for them.
Request early accessinfo@sidian.sg