Who is Selma for?
A map of the audiences Selma could serve, who else is already in the room, and what those buyers pay today. This is for thinking out loud together. It doesn't try to settle the answer.
A map of the audiences Selma could serve, who else is already in the room, and what those buyers pay today. This is for thinking out loud together. It doesn't try to settle the answer.
Selma has been shaped around Australian not-for-profits: child protection, disability and community services. Pioneer partners are running it on premises, and Charters are signed or ready to sign. Nobody has paid for it yet, and we have never committed to a price.
Two things have changed. People working in international development are interested, including word that DFAT would be, and the idea of AI that works in remote places with no internet has caught people's imagination. And there's a strong hunch that local AI will become a normal part of how every organisation uses AI, which opens up corporate buyers of every size.
This site pulls together research on five audiences, the competitors each one already knows, and the prices they are used to paying. A sixth page looks at the offline and solar question on its own, because it changes the hardware.
Target sectors, the business model, capex and grants, a waiting list, sales and events, dev support, the timeline, investor strategy, brand and a bigger Selma Enterprise. Each with the research under it and room to write the decision.
Open the Commercials Day agendaEach page follows the same shape: who they are and how many, how their money moves, why they would want Selma, who else is in the room, and what price could look like.
Australian human-service charities: community services, child protection, disability, family violence, legal, health, ACCOs.
DFAT and the contractors, NGOs, multilaterals and partner governments that actually spend its money across the Pacific, Timor-Leste and Southeast Asia.
Integrity bodies, ombudsmen, regulators, child protection agencies, councils, ACCHOs, universities and health services.
Law, accounting, advice, medical, allied health and psychology practices, plus IP-heavy firms like engineering and agencies.
Regulated organisations with sensitive functions: mutual banks, insurers, health groups, universities, utilities, remote sites.
What it would really take to run Selma with no internet on a solar panel, and which hardware makes that honest.
A plain-language reading of each audience on the things that shape a sale. These are descriptions, not scores.
| Audience | Realistic buyers | Who pays, and how | Typical first ticket | Sales cycle | How crowded | Why they'd care |
|---|---|---|---|---|---|---|
| Not-for-profits | ~2,500 in Australia | The organisation, from thin overheads and tied grants. Capex-shaped grants help. | A$12K–40K | Weeks to months | Cloud AI is near free for NFPs. No local rival found. | Sensitive client data. 50% name privacy or sovereignty as their main AI barrier. |
| Development | Tens of programmes, contractors and NGOs | Donors through intermediaries. Equipment budgets are normal. | A$40K scoping to A$150K–280K pilot | 6–18 months | Free open-source kits and Starlink plus cloud. No governed offline product found. | Sovereignty, remoteness, data costs, Australian presence in the region. |
| Government | ~900 bodies | Agency budgets. Direct buys possible under A$125K (Cth), A$250K (NSW SME). | A$25K–45K pilot | 3–18 months | GovAI Chat for APS, Copilot, sovereign clouds. On-prem with governance is empty. | Dated rules: OVIC block ends 5 Nov, ADM transparency 10 Dec, DTA registers. |
| SMEs | ~3,700 mid-sized, ~25,000 small practices | The firm, from opex. Compares to ~A$30 per seat. | A$6K–20K setup | Weeks, but many small deals | Very. Copilot, ChatGPT, Claude, local MSPs, cheap 128GB boxes. | Privilege, TPB client consent rule, psychology notes, shadow AI. |
| Enterprise | ~5,400 large, plus department "cells" inside each | Function budgets for pilots, IT for fleets. | A$40K–70K year one per department | 6–18 months | Copilot above, AI factories (A$250K+) below. Nothing local in between. | APRA AI letter, CPS 230, shadow AI, fixed cost. |
Ticket sizes are the ranges explored on each audience page. They are working estimates, not tested prices.
Every Commonwealth agency needs an accountable officer, a register entry and an impact assessment for each AI use case. Chief AI Officers in every agency by July 2026.
Banks and insurers asked to map third and fourth-party AI dependencies, test exit plans and avoid relying on a single provider.
Entering client information into an AI tool can count as disclosure to a third party, which needs client permission.
The Pacific Digital and Artificial Intelligence Technology Principles: the Forum's first collective AI position, endorsed at the 55th leaders' meeting in Koror. A governance framework is due in 2027.
DFFH had to block ChatGPT, Claude, Gemini, Copilot and others for two years. What replaces the block gets decided now.
Organisations must disclose when computer programs make or substantially help make significant decisions using personal information.
Due from the Pacific Islands Forum. PNG's sovereign AI framework and Fiji's data protection and cloud policy are also in train.
Free open-source kits (Project NOMAD, Ollama, Open WebUI) and A$5K–10K boxes with 128GB of memory are easy to buy. What nobody else offers is the Charter, Constitution, citations, audit ledger, measurement and the commissioning around them.
Cloud AI sits at roughly A$12 (NFP) to A$45 per user a month. Any per-seat figure near A$100 fails in every audience. Pricing per site or per organisation, with unlimited users, turns Selma's flat cost into an advantage.
Government, enterprise and development buyers will all ask. Qwen is Chinese-origin, and DeepSeek was banned on federal devices in February 2025. Offering Llama, Mistral, Gemma, Granite or an Australian model, with a documented model list, removes the question.
Harvey, OpenAI with NEXTDC, Microsoft and AWS now process more in Australia. "Your data stays in Australia" is no longer rare. "No third party at all, and a ledger to prove it" still is.
At 5–10 person-days per install, a small team saturates after a few dozen sites. Every audience beyond NFPs points to a channel: MSPs, integrators, managing contractors, peak bodies.
An 8–14B model on a 16GB card will be compared with Copilot and ChatGPT. Retrieval over the organisation's own documents, with citations, is where Selma wins. Pacific languages are thinly supported in any open model.