SICA Multi-Agentic Systems Limited · Hong Kong
We are building one — you'll be able to watch it run.
Trust the architecture, not the agent.
Runtime toolkits govern what runs. SICA governs who decides.
Organisations are deploying AI agents that act: research agents, sales agents, customer-facing assistants, and systems that take irreversible actions. Agents fail in a specific way: confident, well-formed, wrong answers that lead to irreversible actions. The missing layer is not another dashboard; it is control over who decides.
A Canadian civil tribunal held a company liable when its chatbot misrepresented a discount its policy did not allow (Moffatt v. Air Canada, 2024 BCCRT 149). In Hong Kong, a business that misdescribes goods or services in the course of trade — whose medium-neutral drafting is capable of covering statements made via automated or chatbot channels — faces criminal liability under the Trade Descriptions Ordinance (Cap 362).
The EU AI Act requires providers and deployers of AI systems to take measures supporting AI literacy among staff and other persons involved in operating those systems (Article 4, applicable since 2 February 2025). From 2 August 2026, Article 50 adds transparency obligations for certain AI systems and uses — including disclosure of AI interaction and labelling of AI-generated content. For organisations that deploy action-taking agents in scope, governance is increasingly a compliance obligation, not a discretionary tooling choice.
Gartner expects over 40% of agentic AI projects to be canceled by the end of 2027 (Gartner press release, 25 June 2025).
One governed architecture — sold as services, built into a product, taught as a discipline.
System architecture services: assessments and workshops for organisations deploying governed agents. The revenue engine and paid customer discovery.
Decision-process governance: policy-as-code gates, approval and veto routing with separation of duties, council-style voting, fail-closed escalation, and an immutable audit log. Framework-native (MCP/LangGraph).
Governed-agent engineering workshops and cohorts for engineering and risk teams — built on an 11-year teaching pedigree.
LIVE reference deployment · public artifacts next
We run a live internal reference deployment of the governed architecture — council-governed, cryptographically signed, with full audit trails. Published decision records (one approved path and one vetoed path) will appear on The System when redacted for public release. Until then, the mechanics are shown; the case files stay internal.
We sell what we run; we teach what we operate.
SICA is pre-traction by design: services first, buyer conversations beginning in months 0–3. If you are deploying agents that act, start with a conversation.
LIVE = our internal reference deployment (dogfood). The governance platform (Gatekeeper) is IN BUILD; the hosted tier is DESIGNED.
The operational domain is confidential. The governance mechanics — gate logic, vote records, cryptographic signing, audit trail format — are shown without redaction on The System page.
Pre-traction by design. Services come first, and buyer conversations begin in months 0–3.
SICA is a Hong Kong engineering firm for governed multi-agent AI systems. Separate proposer, approver, and action roles; deterministic gates; fail-closed defaults; audit trails of who decided. We sell what we run.
Our market deep-dives are research (August 2026). Conversations begin in months 0–3.
A structured review of how your agents propose, approve, and act — where authority sits, what is logged, what fails closed — with a remediation roadmap.
An append-only record of every decision: who proposed, who approved, who could have vetoed, and why — signed and tamper-evident.
