Embeddable ESG decision intelligence

Ask better Scope 3 questions. See the evidence and the tools.

CarbonSage brings hybrid retrieval, cited supplier evidence, trusted emissions tools, and interactive decisions into the software where work already happens.

Manage artifacts and test the workspace-scoped decision agent.

What CarbonSage is proving

A useful agent needs more than a chat box. CarbonSage combines recoverable evidence, evaluated retrieval, deterministic tools, and structured interaction inside a focused Scope 3 workflow.

Grounded by design

Answers preserve artifact, document, page, and chunk provenance instead of turning generated prose into a supplier fact.

Evaluated retrieval

Lexical, pgvector semantic, and hybrid search are compared against representative questions to tune routing and ranking.

Tools own the numbers

Typed deterministic services own calculations, scenarios, citations, and chart data while the agent retrieves and explains.

Product-shaped, not SaaS-sized

A small control plane, one embed path, and one connector prove the product thesis without billing, RBAC, or integration sprawl.

Current baseline

Honest foundations before feature breadth

Traceable calculations

Versioned factors, explicit assumptions, normalized inputs, and warnings keep every displayed result inspectable.

Cited supplier evidence

Bounded document ingestion and PostgreSQL retrieval preserve supplier, file, page, and chunk locations.

Isolated workspaces

Signed sessions, server-side quotas, retention, and workspace-scoped repositories protect every user-owned record.

Next release track

From trusted data to embedded intelligence

Artifact control plane

Manage shipment datasets, evidence documents, and generated decision artifacts with provenance.

Evaluated hybrid RAG

Compare full-text, pgvector semantic, and hybrid retrieval to tune ranking against representative questions.

Typed agent tools

Orchestrate evidence search, emissions calculations, scenarios, data quality, and reports through validated commands.

Structured conversations

Return cited text, metrics, tables, charts, warnings, artifact links, and safe suggested actions.

JavaScript embed

Mount the authenticated agent in an existing application through an isolated iframe and small loader.

Selected-file import

Import one explicitly selected Google Drive file through the same bounded artifact pipeline.