Minimalist R⁴ Terminal Client & Local Vendor API

Preserved interface reference: this page documents the existing terminal client, launcher, and HTTP surfaces from the TLA/R4G1 serving era. It does not show that the route-native geometric-intelligence programme can answer a prompt. There is no working source-free geometric chat path yet; #961 -> #952 -> #953 must establish lexical state, recursive attention, and generation before this interface can carry that claim. See the current programme.

The interface exposes r4 client plus vendor-shaped endpoints such as POST /v1/chat/completions, GET /v1/models, and GET /uor/v1/status. Provider, engine, and multiplication-free claims apply only to the exact legacy execution tier and artifact selected for a request; the existence of the endpoint is not evidence about which mechanism produced its response.

1. Single-Command Launcher (`./r4-app.sh`)

The simplest way to run R⁴ interactively is with the zero-dependency ./r4-app.sh launcher:

./r4-app.sh

What `./r4-app.sh` does automatically:

Model Selection Menu: Presents an interactive choice between smollm2-135m-instruct, smollm2-360m-instruct, and smollm2-1-7b-instruct.
4-Stage Pipeline Compilation & Live Progress: Automatically runs any missing compilation stages (download, corpus compilation, R4G1 graph scoring) while displaying claude-code-styled live progress bars ([█████████████░░░░░░░░] 65% (18s)).
Background Server: Launches r4 serve in the background on port 8000 (or $PORT).
Interactive Client: Automatically opens the interactive client once ready.
Signal Cleanup: Traps EXIT, INT (Ctrl-C), TERM, HUP (window close), and QUIT (Ctrl-\) signals to cleanly terminate the background server process.

2. Interactive Terminal Client (`r4 client`)

You can also run the client against an existing local server instance:

cargo run --release -- client --remote http://127.0.0.1:8000/v1

Upon launching, r4 client displays an ANSI R⁴ ASCII banner and shortcuts guide:

  ██████╗ ██╗  ██╗     ██████╗██╗     ██╗
  ██╔══██╗██║  ██║    ██╔════╝██║     ██║
  ██████╔╝███████║    ██║     ██║     ██║
  ██╔══██╗╚════██║    ██║     ██║     ██║
  ██║  ██║     ██║    ╚██████╗███████╗██║
  ╚═╝  ╚═╝     ╚═╝     ╚═════╝╚══════╝╚═╝

R⁴ Holographic Graph & Transformerless Engine v0.1.0
Zero-Multiply Local Intelligence Runtime • Pinned Multiplication-Free Execution

Connected to local vendor endpoint: http://127.0.0.1:8000/v1/chat/completions
Active teacher model             : smollm2-135m-instruct

Commands & Shortcuts:
  • Type /help to view available slash commands (/status, /models, /clear, /quit)
  • Type / for interactive slash command suggestions & autocomplete
  • Type exit or press Ctrl-D to quit session

Interactive Slash Command Completion

Typing / triggers interactive command suggestions:

Typing / $\rightarrow$ Displays available commands list.
Typing /m or /mo $\rightarrow$ Auto-completes and runs /models.
Typing /s or /st $\rightarrow$ Auto-completes and runs /status.
Typing /h $\rightarrow$ Auto-completes and runs /help.
Typing /q $\rightarrow$ Auto-completes and runs /quit.

3. Local Vendor-Style HTTP API Endpoints

The local R⁴ server implements OpenAI/vendor-compatible endpoints for local tools, IDE integrations, and custom frontends.

`GET /uor/v1/status`

Returns 4-stage pipeline readiness JSON for the active teacher model. (The bare /v1/status path is a deprecated alias — it still works but is served with a Deprecation header pointing at /uor/v1/status, keeping /v1 a pure OpenAI surface.)

`GET /v1/models`

Returns the local model availability manifest.

`POST /v1/chat/completions`

Accepts standard vendor Chat Completions requests.

UOR-R4