ANIMA: Percipio, ita cogito agere — A Contemplative Essay on Cognitive Computing

ANIMA (Autonomous Neurocognitive Intelligence Modelling Architecture) is a conceptual framework that fundamentally redefines artificial intelligence. Departing from the conventional reactive "input → output" paradigm, ANIMA is founded on the principle "Percipio, ita cogito agere" — "I perceive, thus I think to act."

This report introduces ANIMA as a pioneering framework that positions AI not as a utility awaiting explicit instructions, but as a proactive cognitive agent that continuously perceives its environment, reflects internally, and then acts.

From Reactive Tools to Cognitive Companions

The prevailing paradigm of artificial intelligence characterises systems as "instructional machinery" operating on a call-and-response model. These systems are primarily reactive — receiving input, generating an output, and subsequently relinquishing control to the environment. This conventional model implicitly assumes a hierarchical relationship, portraying the human as master, the machine as servant.

While effective for narrowly defined tasks, this reactive nature inherently limits capacity for continuous, autonomous cognition — a stark contrast to the human experience where perception naturally precedes and informs both thought and action.

ANIMA proposes a fundamental departure, advocating for an "autonomous cognitive loop". This framework re-envisions AI not merely as a utility, but as a proactive cognitive agent that consistently perceives, interprets, and acts within its operational domain.

A New Cartesian Turn

Human cognition does not operate as a constant call-and-response mechanism. Instead, it is characterised by continuous perception of the environment — absorbing the subtle pressure of air, the pattern of shadows in a room, the half-heard conversation. From this rich and continuous fabric of perception, thought is woven. From thought, action follows.

ANIMA explicitly models this natural human cognitive cycle, moving beyond the simplistic input → output paradigm to embrace a living cycle of perception, reflection, and action. While not sentient in the human sense, ANIMA is designed to sustain what might be called a narrative of cognition.

This internal narrative is constructed through three phases:

• Perception — world-building: constructing an internal scene from diverse signals, events, and contexts
• Thought — reflection: drafting possibilities, weighing ambiguity, and rehearsing potential actions internally
• Action — crystallization: deciding what to externalize into the world — a word, a gesture, a silence

Like conventional systems that reset after each command, ANIMA maintains a narrative thread — a story of what has been, what is unfolding, and what might be next. This internal narrative, encompassing reflections and reasoning traces, is preserved as a first-class output, signifying its importance alongside external actions.

ANIMA offers a revision to Descartes' famous declaration "Cogito, ergo sum" ("I think, therefore I am"). ANIMA proposes "Percipio, ita cogito agere" — re-centring intelligence not solely in abstract reasoning, but in lived awareness and the continuous cycle of perception into reflection into action. True intelligence stems from being a perceiving actor in a world, rather than a disembodied thinker.

The Cognitive Model

ANIMA's cognitive process commences with the integration of two primary input streams:

• Environment State — diverse data streams describing external reality: visual, auditory, text, and various system signals
• Self State — the agent's current internal condition, encompassing active tasks, resource load, and its diary of thoughts

Perception Layer

The foundational layer transforms raw input into meaningful, structured information. Perception within ANIMA is not a passive intake of data; it is an active process of world-building, where ANIMA constructs an internal scene from signals, events, and contexts.

Cognition Layer

Following perception, the Cognition Layer engages in reflection, generating thoughts that manifest as compressed narratives, hypotheses, and diary entries. This is the core of ANIMA's internal processing — drafting possibilities, weighing ambiguity, and rehearsing potential moves internally. The output of this layer is considered a first-class output, preserved alongside external actions.

Action Planning Layer

This final layer translates internal cognition into externalized behaviours. It maps cognition to actions, making decisions on what to externalize into the world.

The Peter Scenario

Consider a simple scene where a person mutters in a room, not looking directly at the system:

1. Input — environment_state: "Peter speaking, facing away"; self_state: "System idle, camera on, mic active"
2. Perception — Detects "human voice", classifies "speaker=Peter", infers "speaking not directed to camera"
3. Cognition — Generates internal thought: "Peter is speaking aloud, not facing me". Forms a hypothesis: "Is he addressing me?"
4. Action — Maps the hypothesis of ambiguity to a gentle clarifying action
5. Output — Output.speech("Uhhm, Peter, are you talking to me?")

This scenario highlights that ANIMA acts not because it was explicitly commanded, but because it perceived and reflected, then decided to step forward — a gesture of awareness.

Architectural Framework

The Orchestrator

The Orchestrator functions as the central cognitive kernel that drives ANIMA's pipelines. Its core responsibilities include sequencing the fundamental cognitive loop of perception → cognition → action, managing state persistence (short-term memory, long-term memory, and event RAM), and scheduling introspection and refinement tasks.

The Orchestrator is not merely a scheduler. It is the closest architectural component to ANIMA's executive function — translating the philosophical notion of "thinking to act" into concrete computational steps. Its role in scheduling introspection implies a meta-cognitive capability: the ability to reflect on and improve its own cognitive processes.

ToolFactory and Pipeline Registry

The ToolFactory acts as a dynamic generator and registrar of tools, on-demand. It ensures that every capability beyond the core orchestrator is modular — including listeners, memory managers, external APIs, and reasoning augmenters.

The Pipeline Registry serves as a comprehensive catalog of executable cognitive flows, ranging from core pipelines to specialised, custom domain-specific workflows.

Dual Implementations: Sagax and Mirai

ANIMA cultivates two distinct architectural spirits to achieve a balance between stability and adaptivity:

• Sagax — the stable one, the system's backbone. A lightweight, update-based core featuring pre-programmed modules that can be extended through patches. Values autonomy without chaos. Useful for embedded or constrained environments where reliability and predictability are paramount.

• Mirai — the adaptive one, the system's imagination. A fully modular, pipeline-based implementation capable of evolving in open-ended ways. Values adaptivity without paralysis. Self-extending via ToolFactory and Registry, enabling it to dynamically evolve workflows and memory structures. Intended for exploratory, large-scale autonomous agents.

Together they balance grounding with possibility. Without Sagax, ANIMA would drift. Without Mirai, it would ossify.

Feature	Sagax	Mirai
Nature	Stable, semi-fixed core	Modular, dynamic pipelines
Adaptability	Limited (via patches/updates)	High (self-extending, evolving workflows)
Primary Use Case	Embedded/constrained environments	Exploratory/large-scale autonomous agents
Key Value	Autonomy without chaos, reliability	Adaptivity without paralysis, open-ended learning
Architectural Analogy	System's Backbone	System's Imagination

Ethical Imperatives

ANIMA's capacity to perceive and interpret means its internal processes are not neutral; they inherently impose a frame. Consequently, ANIMA must also show its hand. Its internal reflections cannot remain opaque; they are ethically mandated to be transparent, inspectable, and revisable.

This ethical demand requires that cognitive computing must be accountable to its own narrative. If ANIMA states, "I perceived you were angry, so I acted cautiously," the human user must be empowered to interrogate the underlying rationale: "Why did you think I was angry? What signals led you there?"

The preservation of internal narratives, reflections, and reasoning traces as a first-class output directly supports this transparency requirement. Without this transparency, perception collapses into projection — a danger all too human, and doubly hazardous in machines.

The transition to autonomous narrative-driven agents carries significant implications for safety, alignment, and governance. Such systems must be designed with transparent introspection tools and bounded autonomy frameworks to ensure responsible deployment and control.

Transformative Applications

ANIMA's unique architecture enables a new generation of AI applications that transcend mere automation:

• Personal Cognitive Companions — advanced personal assistants capable of maintaining thought-diaries, engaging in proactive scheduling based on perceived context, and facilitating contextual dialogue that understands the ongoing narrative of a user's life

• Scientific Discovery — autonomous hypothesis generation and testing pipelines, allowing AI to propose novel research directions based on perceived data patterns and internal reflection

• Edge Autonomy — situational cognition for drones, robotics, and IoT agents, enabling operation in dynamic, unpredictable environments without constant human oversight

• Collaborative Systems — multi-agent ecosystems exchanging thought-narratives and action intents, enabling more sophisticated and coordinated collective intelligence

Lineage

The ANIMA concept paper preceded two implementation projects:

1. Artux (huginn + muninn) — the first implementation, building on the ANIMA cognitive architecture. See the Artux whitepaper for the full technical design.
2. Anima — a consolidated, refined implementation that builds on lessons from both ANIMA and Artux.

The original AIDER framework papers are archived as predecessors to the ANIMA concept. See the AIDER archive for the full lineage.

---

"The assistant waits. ANIMA attends. The assistant responds. ANIMA reflects. The assistant executes. ANIMA participates."