Cognia documentation

Cognia is a declarative language and runtime for building plastic, recurrent neural systems. A .cognia file describes the substrate—neurons, populations, connectivity, timing, chemistry, learning, attention, outputs and optional controllers. The runtime validates that description, expands it into a concrete graph and executes the graph on a GPU.

Cognia is not a conventional layer-by-layer deep-learning framework. It is designed for stateful systems whose activity continues over time and whose structure can include recurrence, delayed connections, modulation, competing coalitions and explicit outputs.

Start here

• Getting started — validate and run your first model.
• Language guide — understand neurons, populations, modules, networks and brains.
• Language reference — look up declarations, connection patterns and expressions.
• Runtime — see what crosses from source code into the GPU implementation.
• Learning and memory — choose plasticity, supervised readout training or stateful sequence processing.
• Controllers and polarization — steer a distributed network state with or without a learned controller.
• Http API — embed a live Cognia mind in another application.
• Examples — follow tested models from a single neuron to controller-driven polarization.
• Limitations — check implementation status before designing a system.

The shortest useful mental model

.cognia source
  -> lexer and parser
  -> semantic validation
  -> concrete topology
  -> GPU neural graph
  -> stateful steps, outputs and introspection

A model is compiled into a persistent neural graph. Calling step changes its state. Calling evaluate resets state and produces an isolated response. This distinction matters whenever recurrence, delays, adaptation, stochastic firing or polarization is involved.

Current platform

The reference runtime targets Windows with an NVIDIA CUDA-capable GPU. The standalone parser and semantic validator are pure C++ and can validate models without running the full sensory application. Cognia is currently pre-1.0: language and binary compatibility may change as research features become stable.

Where to go next

If this is your first visit, continue with Getting started. If you are evaluating Cognia for research, read Runtime, Controllers and polarization, and Limitations before interpreting experimental results.