Join us
AIport Passport  ·  now in early access

Your AI memory. Finally yours.

AIport Ltd is a London-based tech startup, founded in July 2025, that builds artificial intelligence and brings it to the masses. Our flagship, AIport Passport, is a portable, encrypted memory that lives on your device. Any AI model can board with it, but only with your permission, only for the context you choose, and only for as long as you allow.

Meet AIport Passport Explore the project
Departures
AP 001 AIport Passport Your portable, encrypted AI memory. Any model, your rules Boarding
AP 101 Model development Proprietary and open AI models for practical use Boarding
AP 204 AI assistants Conversational helpers for people and businesses En route
AP 318 AI academy Clear, jargon-free education for every generation Soon
AP 427 Open infrastructure Tools and APIs anyone can build on Planned
Flagship product · Early access

AIport Passport: the AI memory you own.

Today every assistant either forgets you or remembers you inside someone else's cloud. Passport is a portable, encrypted memory and identity layer that lives on your device. Any AI model can “board” with it, but only for the slice of context you choose, only for as long as you allow.

Request early access How it works

Switch models, keep yourself

Move between assistants without re-explaining who you are, what you work on or how you like your answers.

Share context, not data

Customs pseudonymizes names, numbers and addresses on your device. The model sees PERSON_1, never you.

Revoke in one click

Every pass expires on its own. The flight log shows what was read, when, and by which model.

How a Passport flight works

Four steps between your memory and any model.

  1. Step 1

    Vault

    Your preferences, notes and documents are encrypted on your device with keys only you hold. Sync is optional and end-to-end encrypted.

    XChaCha20-Poly1305Argon2idlocal-first
  2. Step 2

    Customs

    Before anything leaves the device, a small local model detects personal data and swaps it for reversible pseudonyms.

    on-device NERpseudonymspolicy rules
  3. Step 3

    Gate

    You issue a boarding pass: which memories, for which model, for how long. The model receives that slice and nothing more.

    scoped tokensTTLrevocable
  4. Step 4

    Flight log

    The answer returns, pseudonyms are restored locally, and every read is written to a hash-chained log you can inspect.

    hash chainre-hydrationopen export
issue-boarding-pass.shPOST /v1/passport/passes
# Issue a 15-minute, read-only pass for one model
curl https://api.aiport.uk/v1/passport/passes \
  -H "Authorization: Bearer $AIPORT_API_KEY" \
  -d '{
    "audience": "model:any",
    "scope": ["preferences", "projects/aiport"],
    "customs": {
      "redact": ["name", "email", "iban"],
      "mode": "pseudonymize"
    },
    "access": "read-only",
    "ttl": "15m"
  }'

# 201 Created
# { "pass_id": "bp_7f3a91", "expires_at": "...",
#   "memory_endpoint": "mcp://passport.aiport.uk/bp_7f3a91" }
Encryption
End-to-end, keys on your deviceXChaCha20-Poly1305, keys derived with Argon2id
Customs
On-device redactionLocal NER model; pseudonyms are re-inserted into replies locally
Interop
Model-agnosticMemory exposed through MCP-compatible tools and a plain REST API
Access
Scoped, expiring passesCapability tokens with TTL from 1 minute; revoke at any time
Audit
Tamper-evident flight logHash-chained record of every read, stored in your vault
Portability
Open export, no lock-inJSON-LD archive you can take anywhere, any time

// AIport Passport is in development. Specifications describe design targets for early access.

Mission

The port where AI meets people.

Artificial intelligence is changing how we work, learn and create, yet too often it stays within reach of a few. AIport is built on the opposite belief: the greatest value emerges when as many people as possible can use it.

That's why we don't just build technology. We make sure you understand it, that it's available to you, and that it genuinely helps.

Accessible

Simple tools with no barriers and no technical background required.

Responsible

Transparency, data security and a human at the centre of every decision.

Practical

We build solutions that solve real problems, not just demos.

Focus areas

Four directions, one goal.

From research to everyday use, we cover the entire journey of artificial intelligence.

AI development

We research, train and adapt models for specific tasks and languages.

Products for people

Assistants and apps with a clear interface that anyone can use.

Education

Workshops, guides and content that explain AI without the jargon.

Open community

Open tools and collaboration, because progress happens together.

Technology

How AIport works under the hood.

From raw data to an answer in the chat window: five stages we repeat with every model release.

  1. Stage 1

    Data

    Corpus collection, cleaning and deduplication, removal of personal data, and balanced language coverage.

    dedupPII filtertokenization
  2. Stage 2

    Adaptation

    Open models are adapted to the task with parameter-efficient fine-tuning and alignment with human preferences.

    LoRA / QLoRASFTDPO
  3. Stage 3

    Evaluation

    Automated test suites, measurement of hallucination and bias, and manual red-teaming before every release.

    benchmarksred-teaming
  4. Stage 4

    Serving

    Quantization reduces memory requirements, while batching and caching speed up responses.

    INT8 / INT4KV-cachebatching
  5. Stage 5

    Interface

    The same model is available through web chat, a REST API and client libraries, with answers streamed in real time.

    RESTSSE streamSDK
example-request.shPOST /v1/chat
# Chat with the model, the answer streams token by token
curl https://api.aiport.uk/v1/chat \
  -H "Authorization: Bearer $AIPORT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "aiport-base",
    "stream": true,
    "rag": { "collection": "documents", "top_k": 5 },
    "messages": [
      { "role": "user",
        "content": "Summarize the attached contract." }
    ]
  }'

# event: token  data: {"delta":"The contract"}
# event: token  data: {"delta":" governs"}
# event: done   data: {"tokens":412}
Architecture
Transformer with open weightsAdaptable to your own data and languages
Context
Window of up to 128k tokensLong documents without manual trimming
Retrieval (RAG)
Vector search + rerankerAnswers grounded in your sources, with citations
Latency
Target: first token under 300 msStreaming via server-sent events
Privacy
No training on your dataUnless you explicitly opt in; encrypted in transit
Protection
Content filtering and rate limitingPer-project API keys with scoped permissions

// Figures shown are target capabilities of the platform currently in development.

Our journey

From idea to everyday use.

Every project passes through three stops before it lands with users.

  1. Stop 1

    We build

    We research and build advanced models and test them on real-world problems.

  2. Stop 2

    We simplify

    We wrap complex technology into tools that are clear and pleasant to use.

  3. Stop 3

    We bring it to people

    Through education and open solutions, we bring AI to the masses, not just to experts.

Contact

Board with us.

Interested in collaborating, partnering or contributing to the project? Drop us a line.

info@aiport.uk

Registered office

AIport Ltd Building 4, Chiswick Park
566 Chiswick High Road
London W4 5YA
United Kingdom

Get in touch

Hours
Mon–Fri, 09:00–17:30

Company details

Legal name
AIport Ltd
Incorporated
4 July 2025
Registered in
England & Wales
Company No.
14398275
VAT No.
GB 429 7361 27
ICO reg.
ZA478592