[KongchangAI]
ProductChemE-as-Code

ChemE-as-Code

An open-source platform for chemical engineering that encapsulates chemical computation tools as agent-callable skills (Agent Skills). An LLM handles intent understanding and tool scheduling, while a deterministic computation engine produces reliable numerical results. It supports tasks such as process simulation, plant layout, HAZOP analysis, and LCA, with engineering files managed under Git version control and automated consistency checks.

Timeline (last 90 days)

Nov 6

ChemE-as-Code is an open-source project aimed at building an executable, auditable chemical engineering computing environment for agents.

Unverified50%
Nov 6

ChemE-as-Code packages chemical engineering computing tools as skills callable by agents, covering modules such as catalyst calculation, process simulation, plant layout, HAZOP analysis and life cycle assessment.

Unverified50%
Nov 6

The design principle of ChemE-as-Code is that computation runs outside the LLM, every numerical value records its provenance, and the model is only responsible for task understanding and tool orchestration.

Unverified50%
Nov 6

A "skill" in ChemE-as-Code is a modular tool packaging specification, typically containing a function signature, a parameter schema, call examples and a description of the return format.

Unverified50%
Nov 6

The agent skill mechanism of ChemE-as-Code is in the same lineage as Anthropic's Agent Skills and OpenAI's function calling approach.

Unverified50%
Nov 6

ChemE-as-Code designs engineering files in a form that is linkable, machine-readable and version-managed by Git.

Unverified50%
Nov 6

The automated checks in ChemE-as-Code can report inconsistencies and outdated results, down to the file and line number.

Unverified50%
Nov 6

In the ammonia plant conceptual design example, the agent produced simulation files, P&IDs, layout drawings, HAZOP analysis, data sheets and relief design, with the API cost of the entire design session being 224 US dollars.

Expired50%
Nov 6

This ammonia plant conceptual design has not yet been reviewed by independent engineers.

Expired50%
Nov 6

The project team validated the checking mechanism by injecting faults and engineering changes, and the results show that problems within the coverage of the rules can be found while problems outside the rules are missed.

Unverified50%

2 more timeline events

All Facts (10)

Unverified

ChemE-as-Code is an open-source project aimed at building an executable, auditable chemical engineering computing environment for agents.

50%
Unverified

ChemE-as-Code packages chemical engineering computing tools as skills callable by agents, covering modules such as catalyst calculation, process simulation, plant layout, HAZOP analysis and life cycle assessment.

50%
Unverified

The design principle of ChemE-as-Code is that computation runs outside the LLM, every numerical value records its provenance, and the model is only responsible for task understanding and tool orchestration.

50%
Unverified

A "skill" in ChemE-as-Code is a modular tool packaging specification, typically containing a function signature, a parameter schema, call examples and a description of the return format.

50%
Unverified

The agent skill mechanism of ChemE-as-Code is in the same lineage as Anthropic's Agent Skills and OpenAI's function calling approach.

50%
Unverified

ChemE-as-Code designs engineering files in a form that is linkable, machine-readable and version-managed by Git.

50%
Unverified

The automated checks in ChemE-as-Code can report inconsistencies and outdated results, down to the file and line number.

50%
Unverified

The project team validated the checking mechanism by injecting faults and engineering changes, and the results show that problems within the coverage of the rules can be found while problems outside the rules are missed.

50%
Unverified

ChemE-as-Code is positioned to assist human review rather than replace engineer judgment.

50%
Unverified

Whether the technical framework of ChemE-as-Code can pass industry validation and enter engineering practice workflows still requires more independent review and testing in real projects.

50%

Source Articles