Summary
Bring the product suite together as a coherent offering: four products users can subscribe to all-in or pick a-la-carte, sold on a client website with Stripe, and hosted on Phala as two lean CVMs (transcription alone; the three translation products together).
This issue is the umbrella that ties together the existing product issues (#8, #9) and adds the commercial surface (website + payments) plus a clearer third translation mode (Parallel Translation).
Business strategy
Offer an assortment of interoperable products instead of funneling every group into a single “what we think you need” mode.
- Users choose what fits: all-access subscription or a-la-carte products.
- Products compose: e.g. pair voice transcription with any one of the translation options.
- Same website / billing surface for discovery, signup, and payment — not four disconnected tools.
The four products
| # |
Product |
Where it runs |
One-liner |
| 1 |
Voice transcription |
Own CVM |
Speech → text inside Signal (Whisper pipeline) |
| 2 |
In-chat (group) translation |
Shared translation CVM |
Bot translates messages in the main Signal group thread |
| 3 |
Parallel Translation |
Shared translation CVM |
Main thread in one language; bot relays a parallel thread in another (bidirectional) |
| 4 |
Language Threads |
Shared translation CVM |
Multiple language sidecar threads branch off a bilingual/trilingual main group |
1. Voice transcription
Standalone product (see #8). Heavy Whisper work should not share RAM with translation fan-out.
2. In-chat / group translation
Classic V1 shape: the bot translates (or dual-posts) inside a single Signal group, in the main thread. Simple mental model for smaller groups. Related framing in #9.
3. Parallel Translation
Two threads, one language each — the bot is a bidirectional relay:
- Main thread e.g. English.
- Parallel thread e.g. Spanish.
- Messages from main → translated into the Spanish thread.
- Human messages in the Spanish thread (not from the bot) → translated back into main.
The bot does not “own” conversation content; it relays between the two threads so each lane stays monolingual for its audience.
(Distinct from Language Threads: Parallel is a 1↔1 pair; Language Threads is main + N language sidecars.)
4. Language Threads
Multiple language threads branch off a bilingual or trilingual main group chat. Monolingual members live in sidecars; the bot bridges main ↔ all active language threads. See docs/language-threads.md and #9.
Client website + Stripe
Build a client-facing website where people can:
This is the commercial front door for the assortment strategy — not just docs for operators.
Phala deployment (cost + privacy)
Host on Phala (TEE / TDX CVMs) so plaintext stays in attestable environments, with manageable OpEx:
| CVM |
Contents |
Target size |
| Transcription CVM |
Voice transcription product only |
4 GB RAM |
| Translation CVM |
In-chat + Parallel + Language Threads |
4 GB RAM |
Rationale:
- Transcription is latency/memory heavy → isolate it.
- The three translation products share similar Signal + NEAR translation workloads → one CVM is efficient.
- Two × 4 GB boxes keep operating costs predictable while products stay composable for the customer (billing/pairing is orthogonal to which CVM hosts which service).
Scope (this umbrella)
Related
- #8 — extract voice transcription to its own 4 GB CVM
- #9 — split in-group translation vs Language Threads
docs/language-threads.md — Language Threads behavior
Summary
Bring the product suite together as a coherent offering: four products users can subscribe to all-in or pick a-la-carte, sold on a client website with Stripe, and hosted on Phala as two lean CVMs (transcription alone; the three translation products together).
This issue is the umbrella that ties together the existing product issues (#8, #9) and adds the commercial surface (website + payments) plus a clearer third translation mode (Parallel Translation).
Business strategy
Offer an assortment of interoperable products instead of funneling every group into a single “what we think you need” mode.
The four products
1. Voice transcription
Standalone product (see #8). Heavy Whisper work should not share RAM with translation fan-out.
2. In-chat / group translation
Classic V1 shape: the bot translates (or dual-posts) inside a single Signal group, in the main thread. Simple mental model for smaller groups. Related framing in #9.
3. Parallel Translation
Two threads, one language each — the bot is a bidirectional relay:
The bot does not “own” conversation content; it relays between the two threads so each lane stays monolingual for its audience.
(Distinct from Language Threads: Parallel is a 1↔1 pair; Language Threads is main + N language sidecars.)
4. Language Threads
Multiple language threads branch off a bilingual or trilingual main group chat. Monolingual members live in sidecars; the bot bridges main ↔ all active language threads. See
docs/language-threads.mdand #9.Client website + Stripe
Build a client-facing website where people can:
This is the commercial front door for the assortment strategy — not just docs for operators.
Phala deployment (cost + privacy)
Host on Phala (TEE / TDX CVMs) so plaintext stays in attestable environments, with manageable OpEx:
Rationale:
Scope (this umbrella)
Related
docs/language-threads.md— Language Threads behavior