instruments for imagination

The power of imagination lives in everyone. The arrival of modern AI lets us imagine at scale. Artists and engineers are at the frontier of collective exploration. The boundary of art and engineering is blurring.

"The idea becomes a machine that makes the art," Sol LeWitt wrote in 1967. We see that unfold in autonomous systems: agents follow a protocol, state accumulates, behavior emerges, artifacts materialize. Imagination becomes a system that runs rather than an image that sits. The idea of living systems has deep roots in artistic practice, but only the recent advances in AI made it possible to ship pieces that are genuinely alive.

We give artists and engineers a shared set of instruments to play with: reusable, composable primitives for inputs, state, agents, orchestration, and all the other scaffolding a living system needs. Borrowing the language artists already know makes AI-enabled imagination immediately expressible, without a new dialect to learn. With those primitives in common, we compose rather than rewrite.

We serve a category larger than art. Any autonomous assembly that carries state, relate to context, and makes decisions based on teleological heuristics is a living system, and the software that carries one carries the others. Art is the avant-garde we enter through, not the scope we limit ourselves to.

1. Living systems

A system is living when three properties are present.

State

The system changes over time. The system is evolving, and never truly returns to its original state. The existence of the system leads to changes in the environment it is embedded in.

Relationality

The system is in relationship with something outside its control. That relation can be external (an audience, a community, other agents, environmental data, neighboring systems) or internal (a self-contained ecosystem whose parts couple to each other richly enough to produce genuine evolution). Its protocols encode metabolism, adaptation, and the capacity to reproduce itself, sometimes in mutated forms.

Autonomy

The evolution of the system is driven by some form of agentic, heuristic, or trained behavior rather than deterministic playback. A living system demonstrates decisional and operational autonomy. The system forms an identity over time, self-sustains and makes decisions based on internal and external motivators.

Mortality is optional. Some living systems have arcs (they grow, resolve, end). Others are ecosystems meant to run indefinitely. Both can exhibit aliveness.

Living protocol art

When a living system is also a work of art, it is living protocol art. Protocols range from the rules behind a single artifact to the physics behind a whole world. Static protocols are implemented in discrete artworks: rule systems, instructions, or generative parameters for individual pieces (Sol LeWitt's wall drawings, Oulipo constraints, Tyler Hobbs's Fidenza). They are not living by our definition. Evolutionary protocols function as genomes: they encode metabolism, adaptation, and decision-logic for autonomous beings (symbients: Plantoid, Botto, terra0, Truth Terminal). World protocols generate and sustain whole environments with their own physics, economics, and ecologies: living protocolic worlds in which many living systems coexist.

The three properties above (state, relationality, autonomy) make a piece living, and the protocol is the medium through which they are expressed. Living protocol art is our avant-garde entry: museums offer a sandbox that allows us to pressure-test living systems. Our focus is on the evolutionary layer. We reach toward the world layer through composability. A sufficient accumulation of living systems relating to each other begins to constitute a world.

2. What we build

Guiding principles

Familiar, flow-inducing

Artists already use instruments to build art pieces and even whole worlds. Our instruments aim to feel familiar. We borrow the metaphors (tracks, clips, arrangements, groups, transport controls, mute and solo) from the creative software artists spent decades becoming fluent in, rather than forcing them to learn the abstractions of AI infrastructure. We aim to induce flow state: the instrument disappears and the artist's attention is entirely on the work.

Symbiotic

We are not building against the machine. We are not building around the machine. The artist and the system collaborate, each doing what the other cannot.

Open

Every piece of software we ship is open, extendible, and customizable: by the artists, for the artists. Artists read the code, fork it, patch it, contribute back. The instruments belong to the community that plays them. Closed software for a practice that prizes autonomy would be a contradiction, and it would make preservation unkeepable in principle.

Playful

We build instruments for imagination. We give form to worlds that existed only in the imaginary space: scenarios, environments, possible futures. Our instruments invite experimentation. Our instruments invite play. The division of labor makes play possible. Others build the models. We build the environment that brings the models, the components, the artist, and the imagination together in one intuitive place. We make the imaginary tangible.

Joyful

A protocol is a set of rules that govern a game. Games are meant to be joyful. Joy is what permits the exploration that produces emergent beauty. Our instruments invite joy.

Components

Model invokers, memory stores, agent schedulers, input adapters, live visualization, state persistence, output surfaces, verifiability layers are plug-and-play modules behind stable interfaces. We provide the sandbox that allows artists to combine them in meaningful ways. All components can be implemented locally, the artist can run their own infrastructure.

Applications

We actively develop two applications:

Rabbithole

A multi-agent orchestration environment built on the DAW arrangement view. Agents sit on tracks, prompts and tasks become clips placed on a timeline, groups hold ensembles, and the transport plays the arrangement forward in turns instead of seconds. The human performer is part of the digital space, not outside of it. Artists compose a living system the same way a musician composes a piece.

Glitchbox

A live visualization interface. Real-time image and video generation at performance speed. An instrument you play on stage, in the club, in the installation. Playable standalone, or driven from Rabbithole.

What we do not build

• A marketplace. Pieces may live on any platform. We do not aim to own distribution.
• Foundation models. We are instrument-builders, not a model lab. We integrate with what exists.
• A specific chain, engine, or rendering target. Portability is the point. We empower our users to build their own artistic environment.
• Curation. We build the instrument. The artist plays the piece. Editorial judgment is out of scope.

3. Who we build for

Protocol artists

The artists already shipping multi-agent, autonomous, or living work (Plantoid, terra0, Botto, and the symbient community around them) built their own scaffolding for years. We studied that scaffolding and are proposing shared instruments they can inspect and fork.

Generative artists in the Fidenza lineage who have mastered static-generative form and are ready to carry state across time find a credible pathway here. Rabbithole's DAW framing and Glitchbox's live-instrument framing speak in vocabulary they already know.

Installation and performance artists bringing AI into physical, temporal, and collaborative practice find Glitchbox especially native: real-time instruments to play on stage and tune during rehearsal.

Cultural institutions

Living protocol art lives inside a field of curators, conservators, critics, collectors, and the engaged cultural public who carry the cultural memory of the form. We build instruments worthy of works that aim to be read as serious art a generation from now.

4. Why now

Society is in a metacrisis

Interlocking climate, institutional, epistemic, and technological disruptions are reshaping every frame we operate through, art included. The response we need is not more dystopia. We need positive visions of the future: futures worth imagining, arguing with, inhabiting before they arrive. Art is one of the vessels through which those futures become experienceable. Living protocol art is a high-bandwidth version of the vessel, because it lets a future be experienced as a system rather than as an image. Our applications empower artists to build such vessels at the speed this moment requires.

Open source technology is available

General-purpose LLMs crossed the threshold where multi-agent, tool-using, memory-keeping behavior became tractable on top of the open-source models. The open-source model ecosystem (Llama, Mistral, Qwen, Kimi, and others) made local-first a real practice. Artists can now run our instruments on their own infrastructure. Real-time image and video generation got fast enough for projection and stage use, which is what Glitchbox needs under the hood. New forms of art are emerging.

The preservation problem has become visible

Pieces built on third-party model APIs do not keep playing. Models get deprecated: a checkpoint an artist composed against is quietly retired, and the piece's outputs drift or stop entirely. A vendor hosting the pipeline sunsets its API, shuts down, or pivots, and the work goes dark. With the latest deprecation of GPT-4o and Claude Opus 3, we already witnessed this happening. The open-source stance described above is not an aesthetic preference. It is the only preservation-viable way to build. Local-first has to be the default.

Institutional legitimacy is forming

Serpentine-scale venues, serious collectors, and museum curators are programming AI and protocol work. The category has the institutional oxygen it lacked a few years ago. Work built with our instruments is read by the art world as work, not as a novelty.

5. Theory of change

Art first, living systems next

[protocol]works builds software for living protocol art. We build software for living systems of any kind: autonomous assemblies that carry state, relate to context, and make decisions over time. Art is the avant-garde: the highest bar we know (aesthetic legibility, cultural stake, institutional scrutiny) and the place where the argument for living systems is made most vividly. If the software can carry a living piece across decades, and do so in a way that reads to curators as serious, it can carry other living systems the world is starting to need.

Why art goes first

Preservation is the highest bar in art. Most software is judged on its roadmap. An artwork meant to live for fifty years is judged on whether it keeps playing. Art practices carry cultural scrutiny that exposes the claims of AI-era infrastructure to rigorous critique before the infrastructure gets widely adopted. If the software holds up under that scrutiny, the broader applications inherit the trust. Art is both the most demanding test case and the public-facing argument for why the category matters.

The philosophical shift we assume

Classical software treats the environment and the individual as separable: a program, its inputs, its outputs. Living systems see inputs, program, and outputs as one. An agent's behavior is shaped by the environment it is embedded in. The agent's actions have a direct effect on its environment. The line between the two is porous by design and the relationship between humans and technology is becoming more symbiotic. This is the foundation we build on: technology that coexists with us and allows us to become more human. It determines how memory is designed, how agents compose, how state persists, how a system deploys. We are aiming to foster living systems that foster us back.

Museums as infrastructure

Museums sit in this theory of change as more than collectors or preservation venues. Museums are institutions of future building: the places where what will be remembered is decided, where cultural imagination gets legitimized, where a piece is stewarded not for a season but for a civilization. Lighthouses of knowledge, distributed across cities and continents, each held by its community of curators, conservators, scholars, and patrons, forming a network of institutions that, in aggregate, constitutes decentralized infrastructure for collective good.

Software that serves museums-as-infrastructure is different from software that serves museums-as-customers. We aim for interoperability to avoid lock-in. Open source stops being a posture choice and becomes a precondition: a runtime whose source cannot be audited by conservators cannot be trusted to carry the works a museum collects. Documentation aligns with conservator workflows. Governance has to contemplate the possibility that [protocol]works might one day no longer exist, while the works it carried still do. We design software so the commitment outlives the commitmaker.

Museums are partners in the decentralized infrastructure. Every living piece is a node. Every museum that carries a node is a lighthouse. We aim to create space for an emergent network.

The world we are betting on

In ten years, the following should feel inevitable in hindsight:

• Living protocol art is a recognized institutional category. Museums collect it, conservators have vocabularies for it, curators program it with the same rigor as video or sculpture.
• Artists can reliably ship work that is still playing decades after it was made.
• The vocabulary of tracks, clips, transports, and arrangements extends into serious non-art practice. Researchers, community designers, and practitioners in adjacent fields use the same artist-built instruments to compose their own living systems.
• We build instruments of imagination that culture needs to imagine its way through the metacrisis.