by Calera Labs · Infinite Context · FinanceSec
Context without limits. Memory without loss.
Domain AI that recalls what it knows — and refuses the rest.
Drop-in OpenAI-compatible API with persistent lattice memory.
Crystallize documents once into Lattice Nodes. Recall Grounded Facts through a scoped viewport.
ICX product → · Sign in → · Head-to-head → · caleralabs.com →
Paired runs used Gemini 3.5 Flash-Lite as the BYOK upstream. Token cuts are versus uncached full-context prefill — not a cached-agent invoice. SWE-bench is predicted pass, not official test-resolved.
|
Mean token cut LongBench v2 official 503 |
Faster on one 852k row 1,375 ms vs 6,227 ms |
LongBench v2 exact 276 / 503 · 0 ICX errors |
RULER MRCR v2 exact 466 / 484 |
The warehouse is the lattice. The desk is a 500–1,500 token viewport. Prompt caching discounts re-reading a haystack; it does not take the haystack off the model.
| Repository | What it is |
|---|---|
| icx-skill-harness | Go client: skill libraries live in ICX, each turn injects a 1–2 tool micro-viewport, BYOK models |
| icx-mcp | Hosted ICX Model Context Protocol server for Cursor, Claude Desktop, and agent hosts |
| finsec-mcp | FinanceSec MCP — SEC EDGAR facts with filing provenance, or honest refusal |
Starter tools in the harness are sandbox mocks. Live memory is the hosted ICX API.
from openai import OpenAI
client = OpenAI(
api_key="icx_live_your_api_key",
base_url="https://icx.api.caleralabs.com/v1",
default_headers={"X-Space-ID": "your_space"},
)Get a key at dashboard.caleralabs.com. Docs: ICX · MCP · Paper
- Infinite Context (ICX) — persistent lattice memory, scoped viewports, OpenAI SDK drop-in
- FinanceSec — deterministic SEC filing Q&A grounded to filings, pages, and XBRL
- BioSec / LegalSec / Compliance — in development on Calera subdomains
One Calera account can subscribe to ICX and FinanceSec separately. They do not share a token pool.
© 2026 Calera Computing, Inc. · Infinite Context is a Calera Labs product · support@caleralabs.com
