Product · Cortex Chat

LCA data retrieval, in a conversation.

Ask in Chinese or English. Upload a BOM. Cortex searches twelve databases in parallel and returns ranked candidates — DQI-scored, every assumption attached.

Open Cortex Chat See a BOM walkthrough

Where Chat runs

Two paths in. One reasoning engine behind both.

Web

cortex.hiq.earth/chat

Sign in, start typing. Conversation history syncs to the cloud. Supports Chinese and English input without switching modes.

API

docs.x.hiq.earth

Four protocols: REST, MCP, AG-UI, A2A. REST for batch lookups. MCP for Cursor / Claude Desktop / any MCP-compatible client. AG-UI for front-end streaming. A2A for agent-to-agent calls.

What people use it for

§ I–IV

Four scenarios LCA practitioners run inside one conversation.

Each returns a named result, not a hedged summary.

§ I

Bulk BOM matching

Upload a Bill of Materials — Excel, CSV, or pasted text. Cortex asks the clarifying question that matters: alloy grade, production route, region. Silence on low-signal rows. It then searches HiQLCD, Ecoinvent, EF, CarbonMinds, and more in parallel and returns DQI-scored candidates per row. Three gates pause for your decision: coverage below 80%, proxy substitution, or cross-database GWP spread over 2×.

100-row BOM · ~10 min · full provenance export

See the full walkthrough →

§ II

Single-material search

Type a material name — "hot-rolled steel, China" or "HDPE granules, extrusion grade" — and Cortex returns a ranked list across all connected databases. Each hit shows GWP value, unit, geographic region, system model, and a link to the source record on HiQLCD. Results in Chinese or English; the databases answer either way.

Natural-language query · multi-database · DQI-ranked top-k

§ III

CBAM scope checking

Walk a product row through CBAM Article 5.2 requirements in conversation: embedded emissions boundary, direct vs. indirect emissions split, and which proxy assumptions the verifier will question. Cortex cites the dataset key and GWP value at each step — not a hand-wave, a named record.

Named dataset · embedded emissions boundary · reasoning chain per row

§ IV

Methodology Q&A

"What is the difference between Ecoinvent cut-off and APOS for my stainless steel row?" Cortex answers from the dataset record and system-model documentation — not from a training-data summary. Ask about ISO 14040/44 system boundaries, allocation approaches, EPD standards, or CBAM default values.

Cites dataset key + system model · not a paraphrase

Top-k, not top-1. DQI, not confidence score. Three gates where a human decides — not a threshold a model crossed silently.

Cortex returns the candidates ranked by fit: geographic region, technology match, system model. Each carries a DQI across five dimensions. Where coverage drops below 80%, where a proxy substitution is required, or where two databases disagree by more than 2× on GWP, Cortex stops and asks.

How Cortex reasons →

Which Cortex

§ V

Chat is the web surface. Cowork is the desktop one.

Chat and Cowork share the same retrieval engine. The project scope decides which to open.

	Cortex Chat	Cortex Cowork
Where it runs	Browser (cloud)	Desktop app (local-first)
File access	Upload per session	Reads and writes your project folder
Conversation history	Cloud-synced, searchable	Local only
Local tool execution	Not available	Python, shell, custom MCP tools
Skills system	Core search skills	Full skills marketplace + custom skills
Best for	Single-session lookups: a BOM, a material, a methodology question	Projects spanning days: decision logs, multi-session memory, openLCA modeling

Chat answers questions. Cowork runs workflows.

Meet Cortex Cowork →

Ask Cortex about a material.

One question is enough to start. No schema, no setup — type the material name, and Cortex returns candidates with sources attached.

Open Cortex Chat API access · info@hiqlcd.com