Skip to content

ChemSpace Copilot

LLM-powered agent system for chemical-space analysis

ChemSpace Copilot combines language-model coordination with deterministic cheminformatics tools. Its default Agno team coordinates seven specialists for ChEMBL retrieval, GTM workflows, downstream analysis, reporting, molecular design, peptide design, and retrosynthesis. A separate robustness agent and a single-agent ablation baseline are registered outside the production team.

How it works

Chainlit / CLI ──► Agno coordinator ──► specialist agents ──┐
                                                           │
External MCP client ──► MCP profiles and adapters ─────────┤
                                                           ▼
            skills + workflow contracts ──► deterministic tools
                                                           │
                                                           ▼
                                  run events + artifacts + storage

The Chainlit/CLI path uses the Agno coordinator as its reasoning engine. In MCP mode the external client is the supervisor, and the server exposes only the capabilities selected at startup. Both paths share the scientific toolkits, catalog procedures, S3/local storage abstraction, and optional MLflow observability.

Workflow events and checksummed artifacts are the reproducible source of scientific state. Conversation history is session-local, and cross-session agent memory is disabled.

Read the Architecture Overview, then continue with Agents, Skills and Workflows, or the MCP server guide.

Highlights

  • Generative Topographic Mapping through chemographykit
  • ChEMBL REST and optional local SQL backends
  • Molecular and peptide generation with standalone and GTM-guided modes
  • Event-backed workflow runs and integrity-checked artifacts
  • S3/MinIO storage with a local filesystem fallback
  • Chainlit streaming, file upload, and inline molecule rendering
  • Prompt-robustness and multi-agent-versus-single-agent evaluation

Quick start

Use the Installation Guide for a local environment or the Docker Deployment Guide for the containerized application.