The drop-in API that restructures any conversation history into topologically-organized loci before your LLM sees them. Get +20 pp accuracy on small models, ~50% fewer context tokens, and on diffusion targets, mathematically guaranteed hallucination-free output on retrieved facts.
Built on the open-source LTMi-XT spec with public benchmarks. Three lines of code. Works with OpenAI, Anthropic, Mercury, Ollama, llama.cpp — any model. Try the live API below — no signup required.
Live demo · No signup
Free tier: 8 calls per 15 minutes per IP. Mercury 2 inference. Your data is processed transiently — nothing is stored.
Input
1409 chars · ~352 tokensOutput · LTMi-XT context
Click Restructure with LENS to see the conversation crystallized into topological loci.
Mercury 2 typically responds in 1–2 seconds.
Why LENS · Already proven
LENS is the productized version of LTMi-XT, our open Apache 2.0 spec. The benchmarks below are public, reproducible, and committed to the GitHub repo — not whitepaper claims.
+20pp accuracy
Topological ordering vs flat retrieval at K=30 on TinyLlama-1.1B. Measured, reproducible, public log.
See the data
0% → 60% rescue
We took a broken 1.3B diffusion model from pure gibberish to coherent QA via forced-anchor decoding.
See the data
100% structured-output
Mercury 2 hits 100% JSONL parse rate at 1.2s avg latency on the LTMi-XT crystallization benchmark.
See the data
Three lines · Any LLM
Wrap your existing OpenAI, Anthropic, Mercury, or local-model call. LENS restructures context using LTMi-XT's lattice-encoded breadcrumb hierarchy and chronological decay, then hands the LLM a denser, topologically-organized prompt. Same model. Better answer. Lower bill.
forced_anchors spec for diffusion targets — provable hallucination-free output# Drop in front of any LLM call (Python)
import requests, openai
def lens(messages, query):
r = requests.post("https://sophiaxt.com/api/v1/lens",
json={"messages": messages, "query": query})
return r.json()["context"]
# Use the structured context with any model
context = lens(conversation_history, next_question)
response = openai.chat.completions.create(
model="gpt-4o",
messages=[
{"role": "system", "content": context},
{"role": "user", "content": next_question},
],
)Pricing · Public, simple, two-axis
Crystallizations are the LLM-backed work — that's the real cost. Retrievals are pure lattice math and effectively free. We tier on both honestly so you never get billed for value we didn't deliver.
Free
$0/mo
Try LENS today, 8 calls/15 min
Hobby
$19/mo
Indie projects, prototypes
Pro
$99/mo
Production apps · diffusion
Scale
$499/mo
High-volume RAG products
Enterprise
from $2K/mo
On-prem, white-label, SLAs
Overage: $0.005/extra crystallization on Hobby & Pro · $0.004 on Scale · $0.003 on Enterprise.
BYOK / BYOPuter: bring your own OpenAI/Anthropic key or use Puter.js for end-user-paid inference. You're not charged for crystallizations on calls where you supply the model — we only meter the lattice + retrieval work.
What's shipping
Live now
Public API · free tier
POST /api/v1/lens · 8 calls / 15 min per IP · returns LTMi-XT context block ready to inject.
Next 7 days
Python + Node SDKs
Three-line drop-in. sophia-lens on PyPI and npm. API key auth for Hobby+ tiers.
Weeks 2–4
Diffusion forced-anchor spec (Pro+)
Provable hallucination-free output on Mercury 2 + future Cassandra T2. Returns position-anchored token specs.
Month 2+
Hybrid mode · public corpus enrichment
Opt-in to read from a curated public LTMi-XT corpus (Python docs, AWS, Postgres, etc.).
Month 3+
Local SDK · Sophia Memory
On-device crystallization & retrieval. Chrome extension wraps ChatGPT.com with persistent memory.
Early access waitlist
Free tier is live above — try the API now. Hobby, Pro, Scale, and Enterprise tiers (with API keys, higher quotas, diffusion forced-anchor specs) ship next week. First 500 signups get 3 months of Pro free.
Or email contactus@sophiaxt.com directly.