Software tools

Software worth knowing for AI-assisted IE work. Two sibling pages cover running models locally and hardware and where to find datasets and pretrained models.

Tools change fast, so every entry carries the date it was last checked. If you spot a stale line, fixing it is a one-line pull request.

Entry format: name + link, one line on what it does, good for / not for, free/paid, and whether data leaves your machine (this matters for confidential industry data and for anything under an NDA or a data-management plan).

Assistants and chat interfaces

• Claude: general assistant; strong at long-document analysis, writing, and code. Good for: working through a paper or dataset description, drafting, code explanation. Not for: facts without verification; check citations before using them (see Best practices). Free tier + paid. Data leaves your machine. (checked 2026-07)
• ChatGPT: general assistant with the largest ecosystem (data analysis, custom GPTs). Good for: exploratory data analysis on non-sensitive CSVs, brainstorming. Not for: anything confidential on the free tier. Free tier + paid. Data leaves your machine. (checked 2026-07)
• Le Chat (Mistral): assistant from an EU provider, with EU hosting options. Good for: work that requires EU data residency. Not for: matching the frontier models on hard reasoning. Free tier + paid. Data leaves your machine. (checked 2026-07)
• Everything else: the assistant landscape shifts monthly. For a current view, LMArena ranks models by blind community votes, and Artificial Analysis tracks quality, price, and speed across providers. (checked 2026-07)
• Local chat front-ends (Open WebUI, LM Studio’s chat, Ollama’s app): chat UI over models running on your own hardware; see Local AI & hardware. Data stays on your machine. (checked 2026-07)

Coding assistants

• Claude Code: agentic coding in the terminal/IDE: reads the repo, edits files, runs tests. Good for: multi-file changes, test-driven work, cleaning up research code. Not for: unreviewed use; read the diffs (see the coding checklist). Paid. Code leaves your machine. (checked 2026-07)
• GitHub Copilot: inline completions and chat in the editor; free for verified academics via GitHub Education. Good for: boilerplate, tests, docstrings while you type. Not for: large refactors. Free (edu) / paid. Code leaves your machine. (checked 2026-07)
• Aider: open-source terminal pair-programmer that can run against local models. Good for: AI-assisted coding on code you can’t send to a cloud service (point it at an Ollama/vLLM endpoint). Not for: the polish of the commercial tools. Free (bring your own model/API). Can stay fully local. (checked 2026-07)

Research and literature

• Zotero: open-source reference manager. Add Better BibTeX for stable citation keys. AI integrations exist as plugins. Good for: holding the references you have verified. Add an AI-suggested citation only after you check it. Free. Local-first (sync optional). (checked 2026-07)
• Semantic Scholar: free scholarly search with a citation graph and a good API. Good for: checking whether a paper an AI cited actually exists. Free. (checked 2026-07)
• Connected Papers: visual citation-graph exploration from a seed paper. Good for: getting into an unfamiliar literature quickly. Free tier + paid. (checked 2026-07)
• Elicit: LLM-assisted literature search and structured data extraction from papers. Good for: first-pass screening and extraction tables. Not for: unverified extractions; spot-check them against the PDFs. Free tier + paid. Data leaves your machine. (checked 2026-07)
• NotebookLM: grounded Q&A over documents you upload; answers cite the sources. Good for: interrogating a pile of reports or EPDs. Not for: confidential documents. Free. Data leaves your machine. (checked 2026-07)

Data and modeling libraries

Python packages relevant to AI + IE work:

• Brightway: open LCA framework; the computational backbone for scripted, reproducible LCA. (checked 2026-07)
• Activity Browser: GUI on top of Brightway, developed at CML, Leiden University. Good for: Brightway workflows without writing code. (checked 2026-07)
• premise: prospective LCA databases coupling ecoinvent with IAM scenarios. (checked 2026-07)
• ODYM: dynamic MFA framework. (checked 2026-07)
• pymrio: environmentally-extended multi-regional input–output analysis. (checked 2026-07)
• scikit-learn: classical ML. Start here: fitting a linear baseline before any deep model is one of the ML-reproducibility recommendations. (checked 2026-07)
• PyTorch: the default deep-learning framework in research; most CV/NLP models on the map ship as PyTorch weights. (checked 2026-07)