Pre-release · Signup open now
ປັນຍາປະດິດ ພາສາລາວ AI that works in Lao, priced for Laos.
LaoGPT runs on Lao models, hosted in Laos, and improved by Lao contributors. Drafting, document Q&A with citations, translation and voice — on a phone, on a desktop, or through an API you can call from anything. Next: vision, hearing, and a router that picks the right model for you.
Pick a prompt to see how LaoGPT answers.
The product
Three surfaces, one brand.
A mobile app built for a prepaid, 4G, Android-first market. A desktop workspace built for institutions. An API built for everyone else. These are live, interactive previews — click through them.
Mobile — voice-first, offline-capable
Android-first for a prepaid, 4G market. The microphone is the primary control, not a secondary one: Lao keyboard input is slow, and speech reaches users with limited literacy.
Change the network. Watch the app, not an error page.
Most of Laos is prepaid, 4G and Android. The constraints are the product: the app is designed for the connection people actually have, and it keeps working when they have none.
- Full model, full context
- Document Q&A with citations
- Voice in and voice out
- Live translation LO ↔ EN ↔ TH
Every answer is metered in kip and shows the region it was served from.
Replying in Lao.
ຂ້ອຍບໍ່ມີລາຄາຕະຫຼາດແບບສົດ.
I don't have live market prices. I can show the ministry's weekly bulletin, or draft a message to your co-op.
Drafted — 46 words, plain register. Tap to edit before sending.
Tap the microphone — voice is the primary control
The models
One family, six jobs.
A single Lao model is the wrong shape for a country where the same platform has to answer a ministry's 400-page legal corpus and a farmer's question on a phone with no signal. LaoGPT runs a family, and the router picks between them. Sizes and context windows are targets for the pre-release, not benchmark results.
Four ways to say the same thing
This is the part with no off-the-shelf answer. Lao is not one language in practice: Laos recognises around fifty ethnic groups, a large share of the country speaks another language at home and Lao at school and at work, and a model that only knows the written standard cannot hear many of the people it is meant for.
The written standard. What a general-purpose model has seen — when it has seen Lao at all.
Northern Lao, with tones and vocabulary of its own, from the old royal capital.
Champasak and the south — continuous with the Isan spoken across the Mekong, by more people than live in Laos itself.
Khmu, Hmong and dozens more at home, Lao at school and at work. The speaker a model trained only on native text understands worst.
And five ways to be polite about it
Register in Lao is not a tone setting, it is a different vocabulary. Getting it wrong in a letter to a ministry is not a style error, it is a social one — and it is the single thing a foreign model is most reliably bad at.
- 1ລາຊາສັບ · royal and monastic
- 2ພາສາທາງການ · official
- 3ພາສາຂຽນ · written
- 4ສຸພາບ · polite
- 5ພາສາເວົ້າ · spoken
Register is a request parameter, not a prompt instruction — so it is auditable, and a workspace can pin it.
See how the router picksSmart routing
One endpoint. The right model, every request.
Most requests do not need the biggest model, and some must never leave the phone. LaoGPT reads each request before a token is generated, and decides which model answers it, which pipeline it goes through, and which region it is allowed to run in. You call one endpoint and never pick a model again.
How the router worksComing soon Roadmap — not yet shipped
Next: it learns to see, read and hear Lao.
Text was the first sense. LaoLLM-Vision, Scribe and Voice add the others — a model that can be shown a photographed form, a stamped and scanned circular, a handwritten minute, a shelf, a leaf or a prescription label, and answer about it in Lao.
LaoLLM-Vision
Photographs, screenshots, charts, shelves, scenes, forms. Answers about what is in the frame, in Lao, with the region it was served from.
LaoLLM-Scribe
The document specialist: stamped, scanned, skewed, multi-column, part-handwritten Lao paperwork into structured fields you can put in a database.
LaoLLM-Voice
Lao speech recognition and synthesis across the regional varieties, including the common case of a question asked by someone whose first language is not Lao, answered in plain Lao.
Cross-modal
The one that matters: a photograph and a spoken question in the same request, answered against your own documents. This is the capability the separate models exist to make cheap.
The platform
The full force of PrivateGPT-Asia.
LaoGPT inherits a complete private-AI application layer rather than rebuilding one. Everything below is platform capability, available on day one, running in-country.
PrivateGPTASIALaoGPT is a branded deployment of PrivateGPT-Asia, not a rebuild. That is the whole reason a country of seven and a half million people can field a serious AI platform: the hard application engineering is inherited, and the effort goes into the Lao layer that nobody else will build.
privategptasia.comRetrieval with citations
Answers traceable to the page and paragraph they came from. A claim you cannot open is a claim you cannot use.
Document ingestion
PDFs, scans, spreadsheets and office documents — including Lao documents that were typed, printed, then scanned again.
Tools and MCP
The model calls your systems through the Model Context Protocol instead of guessing at their contents.
Skills
Packaged procedures for work your organisation repeats — a circular, a report format, a review checklist.
Text-to-SQL
Questions answered against your own tables, with the query shown so somebody can check it.
Drop-in API compatibility
A Claude-API-shaped interface. An existing client points at a Lao endpoint with a base-URL change.
Runs where you put it
In-country servers, your own premises, or fully air-gapped. The platform does not require an internet round-trip to a foreign vendor.
Nothing leaves
Private by construction rather than by policy: data stays inside the environment the platform is deployed into.
And the part that only exists here
The platform gives LaoGPT everything above on day one. Five endpoints exist because Lao needs them and no general-purpose vendor is going to build them: word segmentation for text written without spaces, normalisation of Buddhist Era dates and kip amounts, translation with register control, transliteration, and speech.
Why it costs less
Lao costs more to say in a foreign model.
Lao is written without spaces between words, is tonal, and carries heavy Pali-Sanskrit vocabulary. General-purpose tokenizers fragment it badly — so a Lao user pays a multiple of the English price per sentence and gets a worse answer back. A Lao-optimised tokenizer attacks both sides of that ratio.
ສະຫຼຸບເອກະສານນີ້ເປັນພາສາລາວ
“summarise this document in Lao”
Lao is written without spaces between words, so a tokenizer trained on spaced languages falls back to fragments. You pay per fragment.
Trained on Lao, it segments closer to real word boundaries — fewer tokens in, more context per request, and a better answer out.
What that looks like at volume
Drag to set how much Lao your office or app actually puts through in a day. Token counts scale from the sentence above.
Late-2025 measurements, DataReportal Digital 2026: Laos and the Digital Watch Observatory Lao PDR profile. Used for product sizing.
Residency and consent
Demonstrable, not asserted.
Anyone can put the word “sovereign” on a slide. These are the four places LaoGPT makes it checkable from outside — in the product, not in a policy document.
The residency badge is one of the three places the silk-weave signature is used at all. It stays in permanent view in the institutional workspace.
"region": "la-vte-1"Each API response carries the region that served it. Residency stops being a claim in a brochure and becomes a field your own systems can assert on.
off unless you turn it onProduct updates and evaluation-data use are two separate decisions. Neither is pre-ticked, both are reversible, and withdrawal propagates into the pipeline.
logged · exportableWorkspace activity is logged and exportable, so an institution can answer its own oversight questions without asking us for a report.
egress: blockedInstitutional workspaces can block egress outright and require approval before anything leaves. On-premise and air-gapped deployments are supported by the platform.
The collaboration
Three organisations, one platform.
LaoGPT is not a re-skin and not a reseller agreement. Each partner owns a distinct layer, and the product only exists where the three meet.
Evergreen AILAOSEvergreen AI Laos brings LaoGPT to market. It owns the relationship with the people and institutions who use it — ministries, universities, banks, schools and the public — and it carries the commercial model: kip-denominated pricing, prepaid credit through local payment rails, and free access for education.
- Market entry, distribution and institutional onboarding
- Kip pricing, prepaid credit and local payment rails
- The Lao contributor programme behind the training loop
XTNDDYNAMICSXTND Dynamics builds the product. Part of XTND Group alongside XTND AI, it takes the PrivateGPT-Asia platform and turns it into three shipping surfaces — an Android-first mobile app, a Windows and web workspace, and a developer API — then engineers the Lao layer that no general-purpose platform provides.
- Mobile, desktop and web applications on the platform
- Lao segmentation, normalisation, transliteration and speech services
- The training platform, consent ledger and evaluation harness
PrivateGPTASIAPrivateGPT-Asia is the platform underneath, and the regional home of PrivateGPT. It is the complete private-AI application layer — retrieval with citations, document ingestion, tools, MCP, skills and text-to-SQL — behind a Claude-API-shaped interface, deployable in-country, on-premise or fully air-gapped, with no data leaving the environment it runs in.
- Retrieval, ingestion and citation-traceable answers
- Tools, MCP, skills and text-to-SQL over your own data
- Drop-in API compatibility and air-gap-capable deployment
Who owns what
Hover a partner above, or a layer below, to see where their work sits. LaoGPT is the commercial platform layer — it deliberately does not own the models or the datacentre.
The application platform is PrivateGPT-Asia — the same private-AI layer deployed across the region, run in-country for Laos.
Pre-release programme
Pick the way you want in.
Three cohorts are open. Choosing one sets up the form below — you can change it there.
Pre-release · Signup open now
Join the pre-release.
Three fields. Tell us who you are and where to write, and we'll email you when your invitation is ready.
- Three fields, nothing you have to look up
- Two consents, separable, neither pre-ticked
- A queue position back immediately
Works on 3G · ພາສາລາວ / English