Platform vision

Summary

Point is a general-purpose, AI-first language for entire applications: logic, UI, HTTP, realtime, data, agents, and automation — all authored in .point and checked by the Point toolchain.

See How Point runs for check, run, and optional build output.

One sentence

Point lets teams and coding agents author application logic, application UI, HTTP services, agent pipelines, and automation in one checked semantic layer.

What Point owns

Capability	Semantic blocks
Data modeling	`record`, `variant`
Business logic	`calculation`, `rule`, `label`, `policy`
Side effects	`action`, `external`, `workflow`
HTTP services	`route`, `middleware`, `stream route`
Web applications	`view`, `page`, `layout`, `navigation`
Realtime & jobs	`schedule`, workflow retry/timeout
Agent orchestration	`pipeline`, `session`, `prompt`, `guard`
Database access	`action` + `external` or `std.sql` (`touches database`)
CLI & dev platform	`command`, `point dev`, integration tests

What stays outside Point (by design)

Layer	Why
Xcode, Swift, iOS apps	Native platform; Point invokes via externals/commands
Apple signing, simulators, store upload	macOS toolchain
Database engines	PostgreSQL, SQLite, etc. — you run the server; Point calls drivers via `external`
Third-party SaaS APIs	OpenAI, FAL, etc. — `external` or std packs, not reimplemented runtimes

Point composes with these layers. It does not replace them.

Architecture

.point (author) → semantic AST → core IR → check → emit → JS | TS | PY
                                                      ↓
                              Bun/Node runtime · React emit · SQL drivers · externals

Public source stays semantic. Internal core syntax (fn, let, type) is a compiler implementation detail and must not appear in author .point files.

Platform phases (14–21)

Phase	Focus
14	Language foundations — stdlib, route middleware, variant types
15	Application platform — `layout`, `navigation`, data loading
16	Realtime & processes — `stream route`, workflow retry/timeout
17	Data interop — `std.sql` + external database drivers (any DB)
18	Agent orchestration — `pipeline`, `session`, `prompt`, `guard`
19	Python full parity
20	Dev platform — `point dev`, full-stack template
21	Hardening — conformance, performance, docs

Connect any database

Point does not ship an ORM. Persist data with:

1. `std.sql` — local SQLite via parameterized queries 2. `external postgres driver` (or any driver) — fixed SQL templates + List<Text> params 3. `load data from action` — views call your database actions

See Database interop.

Principles

New features must pass the principles gate: semantic blocks, agent refs, effect honesty, boring emit, no vendor overfit.