diff --git a/docs-site/assets/apple-touch-icon.png b/docs-site/assets/apple-touch-icon.png new file mode 100644 index 0000000..7d15225 Binary files /dev/null and b/docs-site/assets/apple-touch-icon.png differ diff --git a/docs-site/assets/brain.png b/docs-site/assets/brain.png new file mode 100644 index 0000000..1851c98 Binary files /dev/null and b/docs-site/assets/brain.png differ diff --git a/docs-site/assets/favicon.png b/docs-site/assets/favicon.png new file mode 100644 index 0000000..2fd9ab8 Binary files /dev/null and b/docs-site/assets/favicon.png differ diff --git a/docs-site/assets/fig/Amygdala.png b/docs-site/assets/fig/Amygdala.png new file mode 100644 index 0000000..a3df7c1 Binary files /dev/null and b/docs-site/assets/fig/Amygdala.png differ diff --git a/docs-site/assets/fig/Entorhinal.png b/docs-site/assets/fig/Entorhinal.png new file mode 100644 index 0000000..45b6179 Binary files /dev/null and b/docs-site/assets/fig/Entorhinal.png differ diff --git a/docs-site/assets/fig/Hippocampo-neocortical-loop.png b/docs-site/assets/fig/Hippocampo-neocortical-loop.png new file mode 100644 index 0000000..ecc9ae9 Binary files /dev/null and b/docs-site/assets/fig/Hippocampo-neocortical-loop.png differ diff --git a/docs-site/assets/fig/Hippocampus.png b/docs-site/assets/fig/Hippocampus.png new file mode 100644 index 0000000..63a397a Binary files /dev/null and b/docs-site/assets/fig/Hippocampus.png differ diff --git a/docs-site/assets/fig/Neocortex.png b/docs-site/assets/fig/Neocortex.png new file mode 100644 index 0000000..05717c6 Binary files /dev/null and b/docs-site/assets/fig/Neocortex.png differ diff --git a/docs-site/assets/fig/Sensory-cortices.png b/docs-site/assets/fig/Sensory-cortices.png new file mode 100644 index 0000000..351d762 Binary files /dev/null and b/docs-site/assets/fig/Sensory-cortices.png differ diff --git a/docs-site/assets/fig/VTA.png b/docs-site/assets/fig/VTA.png new file mode 100644 index 0000000..f04d24d Binary files /dev/null and b/docs-site/assets/fig/VTA.png differ diff --git a/docs-site/assets/fig/vmPFC.png b/docs-site/assets/fig/vmPFC.png new file mode 100644 index 0000000..c8804f9 Binary files /dev/null and b/docs-site/assets/fig/vmPFC.png differ diff --git a/docs-site/index.html b/docs-site/index.html index b4cd6c4..dc05c20 100644 --- a/docs-site/index.html +++ b/docs-site/index.html @@ -4,257 +4,361 @@ CogniFold — A brain-inspired always-on agent memory - + - + - - - + + - + - - - - + - - - + + + + +
+
- - - -
- -
-
-

Brain-inspired agent memory

-

CogniFold

-

- An always-on agent memory that folds events into self-emerging - cognitive structure, with proactive intent crystallization. -

-
- -
+
-
- - View on GitHub + +
+
BRAIN-INSPIRED AGENT MEMORY
+

CogniFold

+

An always-on agent memory that folds events into self-emerging cognitive structure — and proactively crystallizes them into intent.

+ - -

Apache-2.0 · Python · MCP-compatible

- -
-
12memory systems mapped
-
~60%coverage (weight-normalized)
-
9public benchmarks reported
+
+
19memory systems mapped
+
~53%memory-system coverage
+
4.6×event → concept compression
+
0.614proactivity score
- -
- -
-
-

Coverage map

-

Memory systems

-

- CogniFold maps human memory onto its topology across two layers — the behavioral - functions memory performs, and the neural circuits that implement - them. Hover, tab, or tap a brain region to inspect it; the two columns below give the full - correspondence. These are mappings to the memory taxonomy we can name — well-described - at the behavioral layer, less so at the circuit, and far less at the underlying mechanism. -

+ +
+
DESIGN PHILOSOPHY
+

Memory consolidates — it doesn’t search.

+

Human memory doesn’t look things up — it settles. Borrowing the brain’s complementary learning systems, CogniFold writes events down fast, replays what recurs into durable concepts, and turns that understanding into goals and next steps. Nothing is fetched on demand: structure emerges, and the agent gradually learns what matters to you and what’s worth acting on.

+
+
COMPLEMENTARY LEARNING SYSTEMS
+
HIPPOCAMPUSWrites new events down fast, the moment they happen.
+
NEOCORTEXSettles what keeps recurring into durable concepts.
+
PREFRONTALTurns that understanding into goals and next steps.
+
-
-
-
- -
- - + +
+
+
RESULTS
+
CogniFold’s own figures
+
+
- -
-
-
-

Behavioral · phenomenological

-

What memory does — the level we describe best. Each function maps to its CogniFold mechanism.

-
-
-
-
-
-

Systems · circuits

-

Where memory lives — the CLS substrate. Each region maps to its CogniFold structure.

-
-
-
-
- -
+
+ - -
-
-

Architecture

-

The folding loop

-

- A Complementary Learning Systems design for agents: episodes are captured, folded into - concepts, and crystallized forward into intent. Proactive intent is the differentiator. -

-
+ +
+
-
- + +
+
+
Systems · circuits
+
Where memory lives
+
+
+
+
+
Communications · subspaces
+
How memory moves
+
+
+
-

- Every event enters as a verbatim EVENT node in the hippocampal layer. - As events recur, the agent folds them into generalized CONCEPT nodes in the - neocortical layer — grounded knowledge with explainable provenance back to the episodes that - formed it. From accumulated concepts and time anchors, the prefrontal layer crystallizes - INTENT nodes: commitments to act at the right future moment. This - proactive intent crystallization is what sets CogniFold apart from passive - memory stores — remembering to act becomes a property of the topology, not a separate policy. - Retrieval then closes the loop, letting intent reshape what the agent attends to next. -

+ +
+
THE FOLDING LOOP
+

One map, three node types.

-
-
- -

Hippocampal · EVENT

-

Verbatim episodes captured the moment they happen — what occurred, when, and to whom.

+ +
+ CogniFold brain: event, concept and intent over a complementary-learning-systems graph +
-
- -

Neocortical · CONCEPT

-

Repeated events fold into grounded, generalized knowledge with explainable reasoning.

+ +
+
+
EVENTα
+
Hippocampal
+
Verbatim episodes captured the moment they happen — what occurred, when, and to whom.
+
+
→
+
+
CONCEPTβ
+
Neocortical
+
Repeated events fold into grounded, generalized knowledge with explainable provenance.
+
+
→
+
+
INTENTγ
+
Prefrontal
+
Concepts and time anchors crystallize into proactive intent — remembering to act, by topology.
+
-
- -

Prefrontal · INTENT

-

Concepts and time anchors crystallize into proactive intent — remembering to act, by topology.

+
Time is the fourth axis, threading through every layer · 4.6× compression · 0.614 proactivity
+ +
+
Memory-system coverage~53%
+
+
+
+
+ COVERED 7 + PARTIAL 6 + PLANNED 6 +
-

Time is the fourth axis, threading through every layer.

- -
- -
-
-

Results

-

Benchmarks

-

CogniFold's measured results on public benchmarks. Figures are CogniFold's own; no competitor numbers are shown.

+ +
+
Behavioral · phenomenological
+
What memory does
+
+
-
- -
- -
+
+ - -
-
-

Quickstart

-

Get started

-

Install, run the server, and connect any MCP-capable agent.

-
+ + -
-
-
shell
- 
# install
-pip install cognifold
+
+
-# run the memory server -cognifold serve --port 8800 + + // ---- fit the fixed 1680px canvas to any viewport ---- + (function(){ + const W = 1680, stage = document.getElementById('stage'), frame = document.getElementById('frame'); + function fit(){ + const s = Math.min(1, window.innerWidth / W); + frame.style.transform = 'scale(' + s + ')'; + frame.style.marginLeft = Math.max(0, (window.innerWidth - W * s) / 2) + 'px'; + stage.style.height = (frame.offsetHeight * s) + 'px'; + } + window.addEventListener('resize', fit); + window.addEventListener('load', fit); + fit(); + })(); +