Logipeace Responsible AI

Our approach

Logipeace uses AI to take repetitive work out of logistics: reading shipping documents, summarising and filing emails, and answering questions about your shipments and documents. People stay accountable for what the AI does. This page explains how we keep it safe, accurate and under your control.

We follow five principles:

1. 1

   People stay in charge

   AI drafts, suggests and answers. It acts only inside your Logipeace workspace, and only when someone in your team asks it to or has switched an automatic feature on. Anything used in a booking, invoice or customs filing is checked by a person first.

2. 2

   Your data stays yours

   We never use your data, or anything our AI processes, to train, fine-tune or improve any AI model, and our AI provider may not use it to train theirs. AI output generated from your data belongs to you (see our Privacy Policy).

3. 3

   We measure before we trust

   We benchmark our document reading and our AI assistant against reference sets before we change the models or instructions behind them.

4. 4

   We tell you when AI is involved

   AI-created records start as drafts, and assistant replies and email summaries are shown as AI output.

5. 5

   Someone is answerable

   Each AI feature has a named owner, and the AI's actions are recorded so they can be traced.

This page follows Singapore's Model AI Governance Frameworks from IMDA, including the 2026 framework for agentic AI, and the PDPC's guidelines on personal data in AI systems.

Our services

Logipeace runs two platforms. Both are hosted on Google Cloud in Singapore (asia-southeast1) and are reached through one web app, with Docupeace under its own section.

Shipeace platform. A workspace for logistics teams to run their operations together. Teams organise work into voyages and cards, chat in real time, manage contacts and calendars, and connect their Gmail or Outlook mailbox so shipment emails sit next to the work they belong to. AI features: Email AI and the Shipeace AI assistant.

• Web and mobile: Next.js and React web app; Flutter mobile app.
• Backend: NestJS (TypeScript) API with Socket.IO for real-time updates, Temporal workflows.
• Data: PostgreSQL and MongoDB.
• AI services: Python (FastAPI, LangGraph) services for the assistant and email summaries, using Google Gemini 3.8 models, with a separate tool server that limits what the assistant can do in your workspace.

Docupeace platform. Document AI and warehouse operations. Docupeace reads shipping documents and turns them into structured records: Bills of Lading, packing lists, invoices, delivery orders and customs permits. It also handles B/L management, document collections, sea and air tracking, customs clearance, warehouse management (goods received and issued, stock) and accounting (invoicing, receivables, payables, general ledger). AI features: document extraction, the Docupeace AI assistant and the warehouse order assistant.

• Backend: Python with FastAPI.
• Data: PostgreSQL; source documents in Google Cloud Storage.
• AI: Google Gemini 3.8 models through Vertex AI for extraction, classification and the assistants, with Pydantic AI for the agents. Google Document AI is kept as an alternative extractor.
• Operations: OpenTelemetry tracing and logs collected in our own OpenObserve instance.

Where we use AI

AI never makes payments, signs documents, accepts contract terms, deletes your records, or files with customs or other authorities.

Human oversight

We decide how much an AI feature may do on its own by asking two questions: how much harm could a mistake cause, and how easily can it be undone.

To guard against people trusting AI too much (automation bias), we:

• Save AI-extracted Bills of Lading as drafts, so a person checks them before they are used.
• Require a person to approve AI-read customs permit data.
• Show what the assistant did after each action, with a choice to save or discard the drafts it creates.

AI output can be wrong. You remain responsible for checking it before you rely on it, as set out in our Terms of Service.

Safeguards for AI agents

Where AI acts for you, such as the assistants and Email AI auto-save, we limit what it can reach and what it can do.

• Same permissions as the user. The assistants act with the signed-in user's permissions, never more, and can only use a fixed set of tools. None of those tools deletes records, makes payments or contacts authorities. Data is retrieved under the same access controls as the rest of the platform, so the AI only sees what that user is allowed to see, and each organisation's data is kept separate.
• No mailbox logins in the AI. Email AI receives email content to summarise, but never your mailbox credentials. Those stay with the Shipeace backend, and the AI has no way to send, forward or delete your email.
• Content is treated as data, not instructions. Emails and documents can contain hidden text that tries to steer an AI (prompt injection). Document and email content is passed to the model separately from its instructions, and it cannot widen the tools or permissions the AI has.
• Limits on failure. The assistant caps how many requests it runs at once and stops calling a model that keeps failing.
• Records. Assistant conversations are stored with the actions taken and the model used, and document extractions are written to the audit log, which Docupeace admins can view.
• A kill switch. We can turn off AI extraction or the assistants across the platform at once if we detect a problem.
• Security. Logipeace is certified to ISO/IEC 27001. Traffic to our platform and to our AI provider is encrypted with TLS 1.2 or higher, and our databases, files and backups are encrypted at rest.

Accuracy, testing and monitoring

AI can make mistakes, so we measure it rather than assume it works.

Benchmarks. We test document extraction against a reference set of Bills of Lading with checked answers, and test the assistant's answers against a set of reference questions.

When models change. Before switching to a new AI model or version, we re-run these benchmarks and compare the results. We switch only if quality is equal or better.

Fairness and reliability. Our AI works on business documents and emails, not decisions about individuals.

Data and privacy

Our Privacy Policy sets out the full detail. In short:

• No training on your data. We never use your data, or anything our AI processes (prompts, files, documents, emails and outputs), to train, fine-tune or improve any AI model. If we ever offer a feature that would, it will be opt-in only, and switched on only with your administrator's written consent. We can also put this commitment in your Data Processing Agreement.
• One AI provider. Our AI features run on Google Gemini 3.8 models through Google Cloud's enterprise platform (Vertex AI). Google processes the data as our sub-processor under terms that do not allow it to train its models on it. We do not use consumer AI services, such as the Gemini app, with customer data.
• Where data is processed. Our applications and databases run on Google Cloud in Singapore. AI requests are sent to Google's AI services through global or regional endpoints, so they may be processed outside Singapore.
• Your output is yours. AI results generated from your data belong to your organisation and are treated as your data.
• Retention and deletion. Saved AI results stay in your workspace until you delete them or your contract ends. Logs of AI requests are kept for up to 90 days. You can ask us to delete specific data, including AI request history, at any time. When your contract ends, you have 30 days to export your data; we then delete it, backups expire within 30 days, and we can issue a deletion certificate.
• Advance notice. We tell your administrators at least 30 days in advance before we add or replace an AI provider, or make a material change to how AI handles your data.
• Staff access. Our staff do not read your content unless you ask us to, or it is strictly needed for security or a legal obligation.
• Internal AI use. Our team may only use company-approved AI tools with these same protections. Using free or consumer AI tools with customer data is prohibited.

Accountability

• Senior ownership. A member of Logipeace's senior leadership is accountable for responsible AI across Logipeace and reports on it to company leadership.
• A named owner per feature. Each AI feature has an owner responsible for its testing, quality and incidents.
• Review before launch. Every new AI feature, or major change to one, goes through a written review covering its purpose, data used, level of automation, risks and safeguards, and test results.
• Incident handling. If an AI feature causes harm or behaves unexpectedly, we pause it if needed, investigate, fix the cause and tell affected customers. If a security incident affects your data, we notify you without undue delay and give you the information you need to meet your own obligations under the PDPA. Where Logipeace itself is responsible for the personal data, we notify the PDPC and affected individuals as the PDPA requires.
• Annual review. We review this page and our AI practices at least once a year, and whenever we add a new kind of AI feature.

Your controls and reporting issues

You can:

• Check, edit or discard any AI draft before it is saved or used.
• Choose whether Email AI auto-saves emails into voyages and cards.
• Choose when to use the assistants. They only act when asked.
• Ask us to switch AI features, or individual AI functions, on or off for your organisation. We usually make the change within 1 business day.
• Ask for an export of your organisation's AI usage records (user, time, AI function and result) in CSV or JSON.

If you see an AI error, unexpected behaviour or a safety concern, email support@logipeace.com. We acknowledge reports within 2 business days.

Alignment with Singapore frameworks

These frameworks are voluntary. We follow them because they reflect good practice and what enterprise customers expect.

Questions about this page: support@logipeace.com.