Claude as a direct AI provider
Use an Anthropic API key directly, next to OpenRouter and OpenAI: Claude models without a router in between, under your own contract with Anthropic.
Roadmap
What we are building now, what comes after, and what shipped recently. Plans change as we learn from people using PrismLCA; the order below is our current best guess.
What we are building at the moment.
Q4 2026No major feature in this phase.
Use an Anthropic API key directly, next to OpenRouter and OpenAI: Claude models without a router in between, under your own contract with Anthropic.
The assistant draws charts and diagrams as part of its answer, with Mermaid.js: a contribution breakdown, a comparison, the structure of your product system. Not only in the AI report or on the results page, but right where you ask.
Planned and designed, starting after the current work.
Q1 2027Today there are two: the short report, built without any AI, and the full study report, whose checks bind every number and every statement to the model. In between comes a mode made for the much stronger models now arriving. The agent gets its tools and more trust, looks things up, works out figures and writes more freely, and you review the result.
The assistant already answers every question about your model in the chat, and the report writer drafts the whole study. In between, the assistant will hand you files: an Excel table of an inventory or of results, or a short Word document such as a memo, a data request or a summary for a colleague.
Feed the data for hundreds of products into PrismLCA automatically, and model and calculate them in one aligned, parallel run, for companies with large product ranges. First building blocks already exist internally.
One process, several products, like a sawmill with sawn timber and wood chips: you choose how the burden is shared, by mass, energy, economic value or your own factors. Allocation already works today with parameters: put the shares into parameters and scale the exchanges with them.
PrismLCA for macOS is on its way. If you would like to try it early, write to [email protected].
Ideas we are exploring. Order and scope may change.
2027Extend PrismLCA with plugins: new importers, data sources, tools for the agent or report sections, without changing the core.
Further background databases, with the freely available ones first.
Further data sources and MCP connections the agent can query while it models and researches. First on the list: waste-management statistics from the World Bank and the OECD, so that end-of-life scenarios outside Europe, where Eurostat ends, also rest on published figures. Also on the list: the open data that US federal agencies publish on data.gov.
Air distances follow the routes actually flown: from Bremen to New York becomes Bremen → Frankfurt → New York, with the distance of each leg.
A complete course from the first steps to a full study, for practitioners and students alike, built on the freely available BAFU database.
How well do the major language models handle real LCA work inside PrismLCA? A benchmark of modelling, research and report writing.
Recently released.
Since August 2026A built-in MCP server gives Claude Code, Codex and Claude Desktop the same tools as the in-app agent, with every change in the audit log.
Import the Swiss Federal Administration LCI Database (BAFU:2026) next to ecoinvent, or on its own.
Full study reports drafted chapter by chapter from the model, with every number bound to it.
Long agent runs follow a visible plan and pick up where they stopped.
Group models with their background database in workspaces.
Road and sea routes from place names, and Eurostat recycling rates, for you and for the agent.
Tell us what you need on GitHub; ideas from people using PrismLCA shape this list.