Concept: Babelshift Cloud + monetization (licensing, managed LLM tier)

[robflowk]

Goal

A monetization path for Babel Shift built on two separate axes that compose into a tier model, plus "Babelshift Cloud" as the managed AI tier. Design-first: agree on the model here, then split into implementation issues when the prerequisites (#8, #6) are in place.

The two axes (different mechanics — don't conflate them)

Axis 1: Commercial-use licensing (license-based)

The BabelEdit/Sublime model: free for personal/OSS use, paid license for commercial use.

• Needs: a source-available license (PolyForm Noncommercial, FSL, BUSL — NOT MIT), license key issuance (Stripe + keygen, no accounts required), a license-state indicator in the app.
• Enforcement is honor-system with a nag — standard and accepted for dev tools.
• ⚠️ Hard dependency: the license must be decided BEFORE the first public release — this blocks Distribution: package and publish for Linux (AUR/Flatpak/AppImage), macOS (Homebrew), Windows (winget) #6 (distribution). Relicensing after going public under a permissive license is not realistically reversible.

Axis 2: Babelshift Cloud (usage-based)

Managed LLM translations without bring-your-own-key, plus (later) team features.

• Architecturally just another provider in the Concept: AI translations with a learned per-project glossary and leading language #8 AI-SDK layer: a preconfigured entry pointing at our endpoint, authenticated by account login (device-code flow) instead of a keychain API key. Suggestion pipeline, glossary, drafts: unchanged.
• Backend: thin OpenAI-compatible gateway (metering, rate limits, abuse control) in front of the LLM(s); margin via credit packs or subscription. Zero pipeline lock-in — BYOK and local (One-click local LLM setup for AI translations (managed Ollama sidecar) #9) remain first-class.
• The real product value: zero-key onboarding for non-developer users (translators, PMs won't create Anthropic accounts), and later cloud-only team features: shared glossary sync, translation memory across projects, server-side batch jobs.

Tier model (proposal)

Tier	Price mechanic	Gets
Free	—	Personal/OSS use, BYOK + local LLM (#9)
Pro	one-time or yearly license	Commercial use, same features
Cloud	subscription/credits	Managed LLM (no key), team features as they land

Constraints & risks

• GDPR/DPA: the Cloud proxy processes customer translation content. Needs a zero-retention posture and ideally EU-region inference for German business customers. The local/BYOK story from Concept: AI translations with a learned per-project glossary and leading language #8/One-click local LLM setup for AI translations (managed Ollama sidecar) #9 is the privacy counterpart and stays the escape hatch.
• Sequencing: do not build cloud infra before the AI feature proves itself locally. Order: Concept: AI translations with a learned per-project glossary and leading language #8 core (BYOK/local) → license decision (with Distribution: package and publish for Linux (AUR/Flatpak/AppImage), macOS (Homebrew), Windows (winget) #6) → Cloud tier once there's real usage.
• Metering must be credits/caps, not unlimited flat — LLM COGS.

Open questions

• License choice for the repo (PolyForm NC vs FSL vs BUSL) and what exactly "commercial use" means for a translation tool
• Pricing shape for Pro (one-time + upgrades vs yearly) and Cloud (credits vs seats)
• Account system scope for Cloud v1 (email + device code vs full org/teams)
• EU inference options (Anthropic EU / Bedrock eu-central / Vertex europe-west) and their quality/cost trade-offs
• Does the desktop app phone home for license checks at all, or purely offline keys?

Dependencies

• Concept: AI translations with a learned per-project glossary and leading language #8 provider layer (Cloud = one more provider entry) — blocked by its concept sign-off
• Distribution: package and publish for Linux (AUR/Flatpak/AppImage), macOS (Homebrew), Windows (winget) #6 distribution is blocked by the license decision from this issue (axis 1)
• One-click local LLM setup for AI translations (managed Ollama sidecar) #9 local LLM setup is the privacy-tier counterpart

[robflowk]

Decisions recorded from today (Robin):

1. Honor system, hard commitment: no license keys, no activation, no phone-home, no telemetry — ever. Payment will always be voluntary/honesty-based. This narrows axis 1 to pure legal text + a future payment option, and rules out any enforcement infrastructure.
2. No monetization yet: the only near-term goal was not foreclosing future options. That's done — Add PolyForm Noncommercial license + honest licensing README section #15 shipped PolyForm Noncommercial 1.0.0 (d72410a), which reserves commercial rights so a paid commercial license can be added later without relicensing. The repo had been public with NO license (all-rights-reserved by default, worse for users).
3. Axis 2 (Cloud) stays future work, gated on Concept: AI translations with a learned per-project glossary and leading language #8 shipping and proving the AI translations locally first.

Remaining in this concept: pricing shape for the commercial license, the Cloud tier design, and the GDPR/EU-inference questions — none of them urgent.

[richardchen874-sys]

The managed LLM tier is the part I would design very carefully before monetization, because it decides whether Babelshift Cloud sells convenience only or also absorbs provider complexity for users.

I would put usage metering, tenant budgets, provider/model provenance, and fallback behavior into the cloud layer early. That gives you room to offer a simple managed tier while still controlling cost and reliability behind the scenes.

I am testing an OpenAI-compatible multi-model API layer around official Chinese models, so managed LLM routing and cost control are directly relevant. Do you imagine the cloud tier being priced mainly by subscription, usage, or a hybrid of both?