SPRAPP documentation

Three products, sold separately: a multi-model reasoning panel, a short-string compression API, and offline on-device language models. This page is the hub — each section covers setup, usage, and pricing pointers.

App

SPRAPP Panel

Run one question across many frontier models; converge with dissent visible.

API

smoltext

Lossless compression for payloads under 1 KB, at the Cloudflare edge.

Models

TinyLM

Tiny instruct models running 100% offline in a phone browser. Early access.

SPRAPP Panel

The Panel is a browser app at sprapp.com. Ask once; a panel of models (Claude, GPT, Grok, Gemini, and 300+ via OpenRouter or local Ollama) reasons through the question in three phases — fan-out, peer review, synthesis — and returns one converged answer with disagreements kept visible.

Key points

BYOK on every plan — bring your own API keys; they stay in your browser.
No conversation storage — chats live in your browser (IndexedDB), not on our servers.
Plans — Free forever; Lite from $3–9/mo depending on region. See pricing.
Coordination patterns — 7 ways to combine models: panel debate, routing, cascades, and more.

smoltext — compression API

Short-string compression as an API. General-purpose compressors lose money on payloads under ~1 KB; smoltext layers a semantic JSON codec, dictionary deflate, and a trained codebook, returning whichever is smallest. Lossless by construction — output never exceeds input + 1 byte.

Full reference: smoltext.sprapp.com/docs. Base URL: https://api.smoltext.sprapp.com

Quick start

curl -X POST https://api.smoltext.sprapp.com/v1/compress \
  -H 'content-type: application/json' \
  -d '{"text":"{\"event\":\"login\",\"user\":\"550e8400-e29b-41d4-a716-446655440000\",\"ip\":\"192.168.1.100\",\"ok\":true}"}'
# → 94 bytes in, 36 out — 61.7% savings, lossless roundtrip

Endpoints

Endpoint	What
`POST /v1/compress`	`{"text": "..."}` → compressed base64 + byte stats
`POST /v1/decompress`	`{"compressed": "..."}` → original text
`POST /v1/signup`	`{"email": "..."}` → API key (`sk_live_…`)
`GET /v1/usage`	current usage; requires `Authorization: Bearer`

Plans

Plan	Price	Included/mo
Free	$0	10 calls/day
Starter	$3	100k
Pro	$9	500k
Scale	$29	2M

Overage $1 per 100k calls. Annual billing is 10× monthly (two months free). Anonymous calls share the free quota per IP; signed-up keys get their own.

TinyLM — offline on-device models

Tiny instruct models (the eeny / meeny / miny family, ~50M–135M class, with larger ternary variants) that run fully offline in any phone browser on our own Rust→WASM inference engine — SIMD int8 and 1.58-bit ternary paths, no llama.cpp, no cloud.

Offline by design — the model downloads once (IndexedDB cache) and then works with the network off.
Fine-tune in minutes — LoRA on-device or in the cloud; export SFT pairs from your own data.
Private by physics — inference never leaves the device.

TinyLM is in early access. The browser demo is being prepared at tinylm.sprapp.com. Request access via contact.