Tölvera

Tölvera is a Python library for composing self-organizing simulations and using them as creative or musical instruments. The name is an Icelandic kenning the team coined: from tala (number), volva (prophetess), and vera (being), composed as "number being." It is developed by Jack Armitage at the Intelligent Instruments Lab, hosted at the Icelandic University of the Arts in Reykjavik.

The lab is the project home of the ERC-funded Intelligent Instruments (INTENT) project, led by Professor Thor Magnusson under European Research Council Consolidator Grant agreement No. 101001848 (Horizon 2020). The grant funds a five-year, EUR 2M effort with a team that spans music, computer science, and philosophy: Magnusson, Armitage, Halla Steinunn Stefansdottir, Victor Shepardson, Nicola Privato, Miguel Angel Rozzoli, Halldor Ulfarsson, and others. The stated research question is what twenty-first-century AI means for creative music technology and what new instruments fall out of treating an ML model as a performer rather than a tool.

Technically Tölvera is built on Taichi, a Python-embedded domain-specific language that compiles parallel kernels for CPU and GPU. Built-in behaviors include boids-style flocking, slime-mold (Physarum) growth, particle-life-style swarming, and reaction-diffusion-class patterns; the library exposes these as composable building blocks that can be combined into hybrid systems. Two adjacent libraries from the same lab handle the interfacing: iipyper provides Open Sound Control (OSC) for two-way communication with audio environments such as Max, SuperCollider, and TouchDesigner, and anguilla provides interactive machine learning so a performer can train mappings from sensor input to simulation parameters in real time.

The Mozilla Builders Accelerator selected Tölvera in the September 23, 2024 cohort with funding of up to $100K, and the project was presented at NIME 2024 (New Interfaces for Musical Expression). In the stack it sits at the evaluation and runtime layers as a vehicle for probing what counts as intelligent behavior outside the LLM frame: closer to artificial-life research with a performance interface than to deep learning.