By /

OpenClaw Goes Chemistry: When Multi-Agent Systems Meet Real Computation

Last autumn, we made a deliberate decision: leave multi-agent orchestration to the big players. Build the tools instead. Let someone else build the conductor.

That bet just paid off.

OpenClaw, the experimental multi-agent framework from the Claw ecosystem, now runs Paramus chemistry tools end-to-end.

No glue code. No middleware. Just a skill file, an API key, and hundreds of validated scientific functions at the agent’s fingertips.

What Happened

We published a Paramus skill to ClawHub (the OpenClaw skill registry). An OpenClaw agent reads the skill, authenticates against the Paramus cloud API, and autonomously decides which chemistry tool to call based on the user’s question.

Ask for molecular weight? It calls `calculate_molecular_weight_from_formula`. Ask for LogP? It picks `calculate_logp`. Ask for drug-likeness? It chains multiple tools.

No routing logic on our side. The agent figures it out.

What Surprised Us

The autonomy. We expected to hand-hold the agent through tool selection. Instead, it:

  • Detected the local server was offline (no on-prem Paramus running)
  • Fell back to the cloud API automatically
  • Chose the right tool from hundreds of options
  • Returned a structured, correct result

All from a single natural-language prompt. No intermediate steps visible to the user.

The Architecture

Two paths, same tools:

On-prem (Paramus INTENT installed locally):

  • Agent detects local server on `localhost:8765`
  • Direct tool calls, no cloud dependency
  • Full 1,200+ tool catalog including HPC tools (GROMACS, LAMMPS, ORCA)

Cloud (via `cloud1.paramus.ai`):

  • 337 tools accessible via authenticated API
  • OAuth 2.0 with PKCE, works with Claude, ChatGPT, VS Code, OpenClaw
  • Free tier available

The skill auto-detects which path is available and routes accordingly.

Agentic Landscape: Why This Matters

Multi-agent frameworks are proliferating. OpenAI has Swarm. Anthropic has tool-use patterns. Google has A2A. The open-source world has OpenClaw, CrewAI, AutoGen, and dozens more.

What none of them ship with: validated scientific computation.

That gap is where Paramus fits. We do not build agents. We build the tools agents need.

Molecular properties, thermodynamics, polymer analysis, electrochemistry, DOE, data science – computed, not hallucinated.

When an LLM says “the molecular weight of aspirin is approximately 180 g/mol”, that is a guess from training data. When Paramus says 180.159 g/mol, that is calculated from the atomic composition C9H8O4 using IUPAC atomic weights.

The difference matters in science.

Try It

Cloud (any MCP client):
  • Create an account at [cloud1.paramus.ai](https://cloud1.paramus.ai/)
  • Copy your API key
  • Connect via Claude, ChatGPT, VS Code, or OpenClaw
OpenClaw specifically:
  • Install the `paramus-chemistry` skill from [ClawHub](https://clawhub.ai)
  • Set the `PARAMUS_API_TOKEN` environment variable
  • Ask a chemistry question
On-prem (full catalog):
  • Download from [portal.paramus.ai](https://portal.paramus.ai/login/)
  • Free for academic use

*) On-Prem requires a background check. Paramus is very powerfull.

Scroll to Top