NGI TALER applicant · Submission June 2026

GNU Taler payments, in reach of every European SME.

TalerSME is an open-source AI assistant that removes the integration barrier preventing European SMEs from adopting GNU Taler payments. The Taler protocol is technically excellent, but its merchant-side requires devops skills small businesses lack — a Berlin cafe, a Lviv freelancer, a Krakow shop cannot deploy a Taler merchant backend on their own. TalerSME bridges that gap with three conversational, mobile-first capabilities.

Star on GitHub How it works ↓

The problem

The merchant-side UX of GNU Taler hasn't caught up to the protocol.

Taler is at a critical adoption inflection — the protocol is stable, the NGI TALER funding cycle is actively producing infrastructure components, and EU Digital Euro discussions have made privacy-respecting payments a policy priority. But the small businesses that should benefit cannot deploy what's available today.

50 pages of devops

The official GNU Taler merchant backend ships with a powerful CLI and REST API — and 50 pages of Linux server setup, key management, and webhook configuration. The non-technical SME owner stops at page 3.

Mobile-only owners

The Berlin cafe owner, the Lviv freelancer, the Krakow shop manager run their business from a phone. Existing Taler tooling is desktop-shell-first. Mobile-friendly onboarding does not exist as an off-the-shelf product.

Surveillance is the default

The proprietary alternatives — Stripe, Mollie, Adyen — have great merchant UX, but their business model leaks transactional data. SMEs that care about customer privacy have no turn-key, privacy-by-default option today.

The solution

Three conversational, mobile-first capabilities.

TalerSME wraps the Taler merchant backend in an AI layer that speaks the merchant's language. The assistant never executes destructive operations on its own — it walks the merchant through every command, with explicit confirmation and a full audit log.

Setup Wizard

AI walks the merchant through Taler backend deployment, key generation, webhook configuration, and bank-account binding in 10 minutes of dialogue (EN/DE/UA/PL minimum) — replacing 50 pages of docs.

10-min onboarding

AI Invoice Generation

Merchant says "invoice Anna for the redesign, €1200, due in 14 days" — TalerSME issues a signed Taler payment request, generates a QR code plus payment link, and tracks payment state through to settlement.

conversational

Customer Q&A Widget

Embeddable JavaScript that answers customer questions about Taler in their native language ("How do I pay? Is my data private?") — onboarding customers into the ecosystem, not just the merchant.

embeddable

A 10-minute setup, by dialogue

A glimpse of the wizard.

The wizard runs locally on the merchant's server. The AI is a guide, not an executor — every destructive operation (key generation, deployment, configuration writes) requires explicit user confirmation with the exact command shown. Full audit log of every step.

Built on open foundations

Open protocol. Open weights. Open license.

We do not invent. We compose proven open primitives so the result is auditable end-to-end and stays free from proprietary lock-in — required for the GNU Taler community's acceptance criteria.

Payment layer
GNU Taler HTTP API

Privacy-respecting digital payment protocol developed under the NGI TALER programme. TalerSME integrates via the official taler-rust crate and pins to a tested protocol version per release.

Application layer
Rust + Axum async

Memory-safe, deployable as a single static binary on commodity hardware. Same Rust + AI delivery pattern the team uses on MyHealth-Europe (parallel NLnet submission), de-risking the stack choice.

AI layer
Phi-3-mini Q4 via Candle

Local-first inference on the merchant's own server. Open-weights model, MIT/Apache-licensed runtime. Cloud LLM is opt-in only, with clear user-facing disclosure. Audit log shows every inference and the model used.

What we deliver

Seven months. Seven milestones. Pilot merchants live by M6.

Concrete, milestone-tracked deliverables under recognised open-source licenses (AGPL-3.0 with a commercial dual license — pattern proven by Mastodon, Nextcloud, Sentry, Open edX). Every release is signed; every interface is documented; an independent security audit covered by NLnet's Radically Open Security partnership lands before the public release.

M1

Design + ICH engagement

UX mockups validated with 3–5 non-developer pilot users. Integration Community Hub design review. Repository skeleton + CI.
M2

Setup-wizard backend

Conversational deployment flow over the Taler merchant backend. Read-only checks executed directly; destructive operations gated by explicit user confirmation.
M3

AI invoice generation

Natural-language issuance, signed Taler payment requests, QR + payment link, settlement-state tracking. UX iteration with M1 designer.
M4

Customer Q&A widget

Embeddable JavaScript widget. Phi-3-mini Q4 via Candle on the merchant's server by default. Cloud LLM fallback opt-in only, with audit log.
M5

Multi-language quality

Taler-specific knowledge base (RAG) seeded from official docs + ICH content. Professional DE + UA terminology review. ≥80% acceptable answers on 20-question test suite per language.
M6

Pilot SMEs live (≥3 EU)

3–5 EU pilot SMEs going live with TalerSME. At least 2 in DE/AT/CH, 1 in CEE (UA/PL). Real-world Q&A logs feed back into iteration.
M7

Documentation + release

Self-hosting guide, deployment recipes, threat model, contributor onboarding. Public v1.0 release. Independent ROS audit findings closed.
∞

Sustainability

AGPL-3.0 anchors community contributions; commercial dual license funds ongoing maintenance. Target ≥3 independent downstream forks or deployments within 12 months post-completion.

Compare to what exists

No existing project combines all three pieces.

The Taler ecosystem today has strong protocol primitives but weak merchant-side UX. Proprietary payment processors have great UX but a surveillance-based business model. TalerSME is the first open-source artefact to put AI-mediated onboarding, conversational invoicing, and a customer-facing widget into one self-hostable package.

Project AI onboarding Mobile-first Open-source Privacy-by-default
TalerSME Yes Yes AGPL-3.0 Yes
Official Taler merchant CLI — — Yes Yes
Taler POS reference apps — Yes Yes Yes (POS only)
WooCommerce / Shopify plugins — Partial Mixed Platform-locked
Stripe / Mollie / Adyen — Yes Proprietary Surveillance-based
Generic LLM business tools Yes (generic) Yes Mixed Cloud-only

Project status

Where we are.

Transparent. Pre-submission stage. We publish status updates here as the project moves.

May 2026 — Week 4
Concept lock + scope separation

TalerSME scoped to AI assistant for GNU Taler merchants. Strict scope separation from MyHealth-Europe (parallel NLnet Commons-13 application by the same team).

28 May 2026
Form-ready submission package

14-field NLnet form-ready package complete and pre-sized against character limits. Public landing live before submission.

31 May 2026
Target submission

Final form submission to NGI TALER (1-day buffer to hard deadline 1 June 2026 12:00 CEST).

Q3 2026
If awarded — project kickoff

7-month implementation window. M1 design + GNU Taler ICH engagement within the first month.

Q1 2027
Pilot SMEs live

3–5 EU pilot merchants going live by M6. ≥2 in DE/AT/CH, ≥1 in CEE (UA/PL).

Q2 2027
v1.0 public release

Documentation, threat model, independent Radically Open Security audit findings closed. AGPL-3.0 + commercial dual license.

The team

Four people. Same team as MyHealth-Europe. Distinct scope.

TalerSME is delivered by the same four-person team operating MyHealth-Europe, bound by an internal cooperation agreement. Roles are deliberately distinct from the parallel MyHealth scope — no work-package overlap, no double-funding.

RH

Ruslan Hryban

Project Lead · Principal Engineer

23+ years systems engineering, 3+ years building MCP servers and AI agents. Operator of HealBot.pro (~2 years, low-thousands MAU). FOP, Ukraine.

OS

Oleksandr Suraiev

Coordination · NLnet liaison

Team management, milestone reporting, regulatory tracking, reporting interface with NLnet. Ukraine.

DM

Dmytro Myroshnykov

Business Development · Taler ICH outreach

EU networking, Taler community engagement via the GNU Taler Integration Community Hub, pilot-SME recruitment. Ukraine.

TH

Tetiana Hryban

Cross-border UX · DE/AT pilots

DE-resident native German speaker. Primary-market UX validation of mockups and DE/UA terminology; introductions to DE/AT pilot merchants. Distinct from her MyHealth domain-adviser role.

Ecosystem alignment

Engaged with the GNU Taler community from M1.

The AI is structurally optional: TalerSME can be operated with the AI features disabled, falling back to step-by-step text guidance. All AI components are Free Software (Phi-3 open-weights, Candle MIT/Apache). No proprietary dependencies in the default path. We engage the community early — at M1 design review, through ICH channels for milestone outcomes, and back into ICH documentation as case studies.

GNU Taler Integration Community Hub NGI TALER consortium GNU Taler / Taler Systems SA Free Software Foundation Europe NLnet · NGI Zero community Berlin / EU Taler merchant community

End-user validation pyramid

  • Tier 1. 3–5 EU pilot SMEs going live by M6 — recruited via ICH, EU network, and NLnet community channels. ≥2 in DE/AT/CH, ≥1 in CEE (UA/PL).
  • Tier 2. 10+ informal merchant interviews during M1–M2 design phase, validating setup-wizard UX hypotheses with non-pilot merchants.
  • Tier 3. Customer-side validation through pilot deployments — the Q&A widget is used by pilot SMEs' actual customers, generating real-world Q&A logs for M5–M6 quality iteration.

Transparency

Generative AI in the writing of the proposal.

Per NLnet's policy, we disclose generative-AI use in the writing of the proposal text itself. All technical, architectural, scope, budget, and ethical decisions in the application were made by the human applicant.

Model: Anthropic Claude (claude-opus-4-7), accessed via the Claude desktop application in Cowork mode.

Use period: 2026-05-28 (concept selection day) through 2026-05-31 (estimated submission cut-off).

Use cases: brainstorming framing options; drafting initial concept architecture and milestone plan; research synthesis on NGI TALER call scope, the GNU Taler merchant ecosystem, and local-LLM viability; editing English text to fit NLnet form character limits.

Generative AI was used as a writing and research assistant, not as a decision-making agent. No external action (submission, messaging, code commits) was performed by the AI. Unedited prompt log is attached to the submission as genai-prompts-log.txt.

A privacy-respecting payments path for European SMEs.

Follow the public repository for release tags, milestone updates, and contributor calls. No mailing list, no tracking — just commits.

View on GitHub Contact the team