Your custom private AI agent

Dedicated software and private compute, built around you. Xinan connects personal knowledge, task workflows and Mirror boxes, workstations or compatible devices for tailored financial research, health administration, work and everyday assistance.

Discuss my private agent

AI-generated application concepts. Confirm features, models, hardware compatibility, costs and timing in the project scope and acceptance agreement.

AI built around the way you work

Your knowledge, connected: Connect documents, notes and authorized business material with traceable answers and less repeated searching.
Your preferences, remembered: Tailor language, tone, roles and task preferences; inspect, edit and delete stored memories.
Your tools, working together: Connect calendars, email, knowledge and business systems into a continuous search, drafting, planning and execution workflow.
Your data and actions, controlled: Choose local or private deployment by data sensitivity. Configure access, transport and storage encryption; preview and approve critical actions.

Three areas, one personal system

These are customization options, scoped by available interfaces, permissions and acceptance tests. The product name does not imply unlimited or unsupervised autonomy.

Financial research agent

Organize news, filings and authorized holdings into sourced research, watchlists and review records with configured alerts.

No return guarantees. Trading requires separate interface, permission, limit and human-approval assessment; it is not enabled by default.

Health and appointment agent

Organize authorized health records, checkups and questions for clinicians; coordinate appointments, follow-ups and care reminders with approved family access.

Information and administration support, not diagnosis, prescriptions or emergency care. Assess sensitive data and healthcare integrations separately under applicable requirements.

Work and everyday agent

Organize email, meetings and tasks; assist purchasing comparisons, order support, travel, family schedules, draft replies and authorized device alerts.

Approve payments, bookings and important replies under agreed rules. Alerts depend on integrations and connectivity and do not replace public emergency services.

Designed for your actual use cases

AI-generated scene: a night study with two monitors of market curves and analysis charts beside a local compute tower

Research with traceable sources

Start with a watchlist and authorized material, summarize filings and open questions, and prepare a research draft. Investment and trading decisions remain yours.

Organize family healthcare visits

Turn authorized checkup records into a visit summary, clinician questions and follow-up reminders. Confirm family access before sharing; clinicians make clinical judgments.

Turn meetings into follow-through

Extract tasks, owners and dates from authorized meeting notes. Draft email and calendar updates, then approve changes through connected tools and record results.

Compare first, approve before paying

Collect requirements, budgets and supplier information into purchasing lists and support tracking. Verify items, totals, addresses and platform conditions before ordering.

Keep travel plans together

Compare itineraries against schedules and preferences, prepare meeting material and travel reminders. Approve costs and consequences before bookings, cancellations or changes.

Alerts with a human fallback

Receive authorized device alerts, notify agreed contacts and show delivery or failure. Escalate undelivered notices for human handling, without an emergency-response guarantee.

From a request to a traceable result

Understand
Check access
Retrieve and plan
Approve actions
Execute and track
Report results

Insufficient access, platform errors or cancellation stop the affected action with a clear explanation. Purchases, payments, bookings and important replies require agreed approvals; retries must prevent duplicates.

Connect software with private compute

Deploy on your existing equipment, a Mirror compute box or a private AI workstation, with private-cloud and hybrid options. Evaluate models, data volume, responsiveness and budget before selecting NVIDIA, AMD or workstation configurations.

NVIDIA platform box: Select by models, software ecosystem and actual device; not every brand is an original DGX Spark.
AMD platform box: An alternative private deployment path; validate platform, device model, memory and tool compatibility separately.
Private AI workstation: Size for professional loads, VRAM, cooling and noise, prioritizing workstation GPUs.
Existing devices and private cloud: Reuse compatible equipment or assess private-cloud and hybrid options, stating external dependencies and offline limits.

Agree hardware, software customization, third-party subscriptions and maintenance separately; not every feature is bundled. Compatible existing equipment can receive software-only customization. Confirm models, memory, compatibility and timing after validation.

How your data and compute connect

Private deployment boundary

Authorized knowledge: Documents, preferences and retention
Private agents: Retrieval, plans and approvals
Private compute: Box, workstation or compatible device

Authorized external services

Local deployment does not mean always offline. External models, search, email or notifications may send data out; document destinations, costs and consent. Transport and storage encryption depend on configuration; end-to-end claims require verified links and key control.

DCP · DCP participation needs separate consent. Private data and spare compute are not shared by default; agree isolation, fees and exit rules without promised returns.

Permissions before execution

Read and retrieve: Answer only within authorized data and cite sources.
Draft only: Prepare suggestions and replies without sending externally.
Approve then act: Review recipients, amounts and impact before agreed actions.
Bounded automation: Separately agree tasks, limits and revocation with audit records.

The foundation from understanding to action

RAG retrieval and citations: Parse and chunk documents, combine vector and keyword search, rerank and cite results, and filter by access scope.
Tool use and workflow orchestration: Use approved APIs with input validation, approval steps, idempotency, retries and logs so tasks can be traced and stopped.
Manageable memory and personalization: Separate sessions, lasting preferences and business facts; retrieve relevant memory with retention, versioning and deletion controls.
Model adaptation and evaluation: Select models, prompts and inference settings by task, fine-tune when needed, and evaluate quality, cost, latency and recovery.

Model development and training

From a requirement to your personal AI system

Requirements and tasks
Map goals, data, tools and acceptance tasks; agree priorities.
Knowledge and connectors
Prepare authorized data, knowledge structure, access and connectors.
Prototype and experience
Test retrieval, memory, interaction and execution with representative tasks.
Deploy, validate and train
Deploy to equipment or private environments with test records, documentation and user training.
Operate and improve
Provide agreed maintenance, version updates and new tasks, reviewing quality and cost.

What you receive

Hardware and software scope, third-party licenses and maintenance terms
Data flows, retention and deletion, access revocation and backup recovery
Custom roles, interface and task configuration
Knowledge, data connectors and memory management
Tool connectors, access and approval rules
Test records, deployment guides and recovery plan
User training and ongoing upgrade plan

Validate with real tasks

Agree a task set and measure citation accuracy, task completion, response time, resource use, access isolation and recovery. Set targets from prototype measurements and track version changes and running costs.

Learn and explore

Understand the fundamentals, explore governance and energy, then move to technical practice.

YouTube explainers

International organizations

Courses, security and deployment

External learning resources, not partnerships, certifications or product endorsements. Captions, translations and access depend on the provider. No autoplay or republished third-party videos, thumbnails or reports. 2026-10-13

Tell us what you need

Which tasks, documents and tools should your agent handle? Which actions need your approval? Share equipment, users, languages, budget and timing so we can scope a prototype and proposal.

Do not submit passwords, trading keys, complete medical records or payment details in an inquiry. Start with tasks and budget; assess data through an agreed secure process.

Discuss my private agent

Your custom private AI agent

Discuss my private agent

AI-generated application concepts. Confirm features, models, hardware compatibility, costs and timing in the project scope and acceptance agreement.

AI built around the way you work

Your knowledge, connected: Connect documents, notes and authorized business material with traceable answers and less repeated searching.
Your preferences, remembered: Tailor language, tone, roles and task preferences; inspect, edit and delete stored memories.
Your tools, working together: Connect calendars, email, knowledge and business systems into a continuous search, drafting, planning and execution workflow.
Your data and actions, controlled: Choose local or private deployment by data sensitivity. Configure access, transport and storage encryption; preview and approve critical actions.

Three areas, one personal system

These are customization options, scoped by available interfaces, permissions and acceptance tests. The product name does not imply unlimited or unsupervised autonomy.