Frequently asked questions

Yes. PrismLCA is free and open-source software under the AGPL-3.0. There is no account, no trial and no paid tier. Costs arise only from services you choose yourself, such as an AI provider or an ecoinvent licence.

You do. The AGPL covers the software, not what you make with it. Your models, results and reports are files on your computer, and PrismLCA uploads none of them. Only when you use the assistant are the parts it works on sent to your own AI provider.

No. There is no telemetry, no crash reporting and no analytics in the app. Every connection the app can make is listed in the privacy notice, and each one needs a switch, a key or a click from you.

PrismLCA is a private open-source project by Jonas Harth. There is no company behind it. You can reach the project through the contact form or at [email protected].

PrismLCA works with your own API key for OpenRouter or OpenAI. A fresh installation uses GPT-6.1 Sol for the chat; you can pick another model per role (chat, report writing, research, review, material assignment) and set how much effort it spends.

There is no PrismLCA server in between: the app talks to your provider directly, under your contract with it.

Only what the current task needs: your message, the conversation, the parts of the model and results it is working on, and the results of the tools it calls. Nothing is sent without a key and without a message from you.

PrismLCA asks OpenRouter for providers that do not collect data and for zero-retention endpoints where they exist. Claude, GPT and Gemini models are exempt from the zero-retention request by default and run under their vendor's own retention terms; you can make each vendor strict in the settings. The full list is in the privacy notice.

Yes. Every change the assistant makes lands in the audit log with its name and can be undone there, one step or many. Actions that cannot be undone, such as deleting a model, always ask you first in the PrismLCA window.

No. Modelling, calculation, results and reports all work by hand. Without an API key PrismLCA makes no connection to any AI provider at all.

PrismLCA itself is free. Your AI provider bills you for the tokens you use, at its own prices. The app shows the context size of each chat, so long sessions do not come as a surprise.

A full study report in chapters: introduction, goal and scope, system model, life cycle inventory, impact assessment, contribution, interpretation, limitations and conclusions, with figures and tables built from your model and results. The assistant drafts the prose; numbers, units, tables and charts come straight from the calculation, not from the language model.

For the introduction and the written chapters the writer can also search the literature through your AI provider — regulation, the product's market, published LCAs of comparable products — and lists what it found with its sources in the references.

If you need something quicker, the short report summarises system model, hotspots and scenario differences on a few pages in one click.

The report follows the structure ISO 14044 asks for, but software cannot make a study compliant. Methodological choices, the critical review and the responsibility for the conclusions stay with you.

To support that, a second model reads every chapter against your figures and study before you see it, and you approve the report yourself. A chapter written or changed after your approval is marked New again.

Yes. Every chapter can be edited by hand, rewritten by the assistant on request, or locked as it is. Figures in the text stay bound to the results they quote and update themselves after a recalculation; chapters whose wording may no longer hold are marked Needs update, and an update changes only the sentences that are no longer true.

Reports are exported as PDF, with a choice of themes and colour sets. They can be written in English, German, French, Spanish or Portuguese; spell checking for these languages works offline with dictionaries that ship with the app.

You can start for free with the Swiss Federal Administration LCI Database (BAFU:2026), which PrismLCA imports as a background database. If you hold an ecoinvent licence, you can import ecoinvent from your local ecoSpold2 files, with no login. Both can sit side by side.

No. ecoinvent is licensed by the ecoinvent Association, and PrismLCA neither ships nor downloads it. You bring your own licence and files; PrismLCA reads them on your computer.

The calculations run on Brightway 2.5, the open-source LCA framework. PrismLCA adds the model canvas, parameters with formulas, scenarios, tags for life-cycle phases, the assistant and the reports.

Export it as a .prism file and send that file. It contains your own processes, parameters and scenarios, but no licensed background data: the links to background datasets are kept, the datasets themselves are not, so your colleague needs the same background database.

The methods that come with your background database, for example EF 3.1 or IPCC climate change. You can also build your own methods and method families from characterisation factors.

Results per indicator, a contribution tree down to single processes, results grouped by life-cycle phase, and comparisons between scenarios and models. Charts can be saved as PNG files, tables exported to Excel.

Windows (signed installer) and Linux (AppImage, .deb and .rpm). A macOS version is coming soon.

PrismLCA searches databases with a language model that runs locally on your computer. It is about 2.3 GB, too large for the installer, so the first start asks whether to download it; PrismLCA needs it to run. After that, searching works offline and sends nothing anywhere.

PrismLCA never checks on its own. Under Settings → About the button Check for updates looks for a new release; the download starts only after you confirm it.

Yes, once the search model is downloaded, apart from the features that need a server by nature: the AI assistant (including its web and literature search), the update check, the news, the transport-distance lookup and the Eurostat waste statistics. Modelling, calculation, results and PDF export work without a connection.

The Model Context Protocol is an open standard that lets AI agents use tools. PrismLCA has a built-in MCP server, so agents such as Claude Code, Codex or Claude Desktop can read and edit your models with the same tools as the built-in assistant.

Under Settings → AI → MCP. The server is off until you turn it on. PrismLCA then shows a ready-made configuration snippet for each supported agent to copy, or writes it into the agent's configuration for you.

The server listens only on your own computer (127.0.0.1) and is not reachable from the network. Every change an agent makes is logged under its name and can be undone; irreversible actions ask you in the PrismLCA window first.

What the agent does with the answers, typically sending them to its own AI provider, is governed by that agent and your contract with its provider.