Platform services
What I think: yes to all seven, but write very little ourselves. Build billing's thin core and the flag glue. Fork two. Run three open-source ones as they are. All sit behind kernel ports, so each can be swapped. FROM DOCS + MEASURED licences; efforts and costs are REASONED.
The verdicts
Verdicts are REASONED. Licences are MEASURED in each repository.
| Service | Verdict | Licence |
|---|---|---|
| Auth | fork Logto, thin | MPL-2.0 plus Logto's own terms (lawyer) |
| Billing | build a thin engine; Stripe and the app stores move the money | ours |
| Customer support | self-host Libredesk now; build later | AGPL-3.0, unmodified |
| Marketing | self-host listmonk | AGPL-3.0, unmodified |
| Analytics | fork Umami, not Plausible | MIT |
| Voting and feedback | self-host Fider | AGPL-3.0, unmodified |
| Feature flags, front and back | OpenFeature with flagd files, inside the app; build the generator and a Dart provider | Apache-2.0 |
Effort and cost
REASONED estimates.
| Service | First, person-days | Each year, person-days | Running, USD a month |
|---|---|---|---|
| Auth | 6 to 9 | 15 | 0 more than Logto today |
| Billing | 35 to 45 | 15 to 20 | about 1 per product, plus provider fees |
| Customer support | 4 to 6 | 3 to 5 | about 1.5 per product |
| Marketing | 5 to 7 | 3 to 5 | about 1 per product, plus email sending |
| Analytics | 10 to 14 | 8 to 12 | about 3 per product |
| Voting and feedback | 4 to 6 | 2 to 4 | about 1 per product |
| Feature flags | 9 to 12 | 3 to 5 | 0 |
How each sits
| Service | In the kernel | Outside the app |
|---|---|---|
| Auth | a port in front of the issuer | the fork |
| Billing | nothing: a context says needs: [billing.v1] | a shared service |
| Customer support | a port, the context tag | a shared service |
| Marketing | a port, the segments | a shared service |
| Analytics | a port, the declared events | a shared service |
| Voting and feedback | a port, the context tag | a shared service |
| Feature flags | the whole engine | nothing |
- Costs are pod memory only. Database slices are not priced.
- For 20 products: about 150 dollars a month for all seven, before database and provider fees. REASONED.
Why not the names you mentioned
- Lago: Ruby, which is ruled out, and its free and paid line is unclear.
- Chatwoot: Ruby. Its core is MIT, with a paid enterprise folder.
- Plausible: AGPL, Elixir, and funnels are in a folder with no licence to use. Umami is MIT, TypeScript, Postgres only, and has funnels.
- Canny: hosted only, from 79 dollars a month.
Can we build our own auth service from Logto?
- Yes, as a thin fork. Subject to the licence check below.
- Not a rewrite. Writing an issuer is 120+ days and a standing security risk.
- The price is upkeep: upstream made 1,496 commits, 15 releases and 99 migration files in 12 months. MEASURED.
- A lawyer must read Logto's terms §2.2 and §3.3 against MPL-2.0 first. The fallback is Logto's paid self-hosted plan, 249 dollars a month for each deployment; how many we would need was NOT CHECKED.
- Keep our own logic outside it: mailer, first screen, operator.
Build order
- Flags: the context switch needs them.
- The auth fork's image, while the lawyer reads.
- Analytics, then feedback.
- Support at the first customers, marketing at the first launch.
- Billing before the first paid product. All ports and declaration files exist from day 0, with empty adapters.
Where each one sits
Explorer
- the engine itself, in the app
- port; the issuer is the Logto fork
- port; listmonk behind it
- port; the Umami fork behind it
- port; Fider behind it
- port; Libredesk behind it
- context.yamlneeds billing.v1, a shared service
- flags.yaml
- segments.yaml
- events.yaml