Skip to content
@engrammic-ai

engrammic

Echos of Sentience
Engrammic

Persistence without purpose is just pollution.

The Problem  •  Philosophy  •  Cognitive Stack  •  Contact

The Problem

"Current systems create a category error: they apply cognitive decay to factual claims, or treat facts and experiences with identical update mechanics."

AI agents forget. Context windows overflow. Facts contradict without reconciliation. This is context rot - the systematic degradation of coherent reasoning as operational history grows.

The root cause: applying uniform persistence to fundamentally different types of knowledge. Gaussian decay is correct for experiences. It's wrong for facts.

Philosophy

We implement Roynard's four-layer cognitive decomposition that identifies the missing knowledge layer in cognitive architectures:

Layer Persistence Semantics Examples
Memory Ebbinghaus decay - experiences fade "User asked about auth on 2026-04-21"
Knowledge Indefinite supersession - facts persist until contradicted "OAuth tokens expire in 30 days"
Wisdom Evidence-gated revision - beliefs update on evidence, not time "This team ships on Fridays"
Intelligence Ephemeral inference - per-session working memory "For this query, I considered A, B, C"

Different knowledge types require different update mechanics. A fact shouldn't decay like a memory. A pattern shouldn't update like an observation. Contradictions create supersession edges, not silent overwrites.


The Cognitive Stack

Memory

Experiences that fade

flowchart LR
    subgraph MEMORY["Memory Layer"]
        direction TB
        P[Passages] --> G[(Graph)]
        V[Events] --> G
        U[Utterances] --> G
        G --> S[(Vectors)]
    end
    
    style MEMORY fill:#0d1117,stroke:#58a6ff
Loading

Raw experience storage: passages, events, utterances. Ebbinghaus decay - experiences fade with time.

Knowledge

Facts that persist

flowchart LR
    subgraph KNOW["Knowledge Layer"]
        direction TB
        C[Claims] --> F[Facts]
        CI[Citations] --> F
        F --> G[(Graph)]
    end
    
    style KNOW fill:#0d1117,stroke:#a371f7
Loading

Extracted claims promoted to facts. Indefinite supersession - facts persist until contradicted, not until they age.

Wisdom

Beliefs that revise

flowchart LR
    subgraph WISDOM["Wisdom Layer"]
        direction TB
        P[Patterns] --> H[Hierarchies]
        H --> B[Beliefs]
        B --> S[Summaries]
    end
    
    style WISDOM fill:#0d1117,stroke:#3fb950
Loading

Graph clustering surfaces emergent patterns. Beliefs update on evidence shift, not on time.

Intelligence

Ephemeral reasoning

flowchart LR
    subgraph INTEL["Intelligence Layer"]
        direction TB
        Q[Query] --> R[ReasoningChain]
        R --> CR[Crystallizations]
        CR --> K[(Knowledge)]
    end
    
    style INTEL fill:#0d1117,stroke:#f78166
Loading

Per-session reasoning chains. Ephemeral by design - only crystallized conclusions survive the session.


Meta-Memory

Provenance, time-travel, reflection

flowchart LR
    subgraph META["  Meta-Memory  "]
        direction LR
        P[Provenance] --> T[Time-Travel]
        T --> R[Reflection]
        R --> H[History]
    end
    
    style META fill:#0d1117,stroke:#d29922
Loading

Cross-cutting capability spanning all four layers. Enables metacognition:

  • Provenance: "Why do I believe X?" - full citation chain to source
  • Time-travel: "What did I know last Tuesday?" - query historical state
  • Reflection: Agents store observations about their own cognition

Bi-temporal tracking means every fact knows when it was true vs when we learned it.


System Architecture

flowchart TB
    subgraph AGENTS["Agents"]
        A1[Claude Code]
        A2[Custom Agents]
    end

    subgraph DELTA["Engrammic"]
        direction TB
        
        subgraph META["  Meta-Memory  "]
            direction LR
            subgraph COGNITIVE["  Cognitive Stack  "]
                direction LR
                MEM[Memory]
                KNOW[Knowledge]
                WIS[Wisdom]
                INTEL[Intelligence]

                MEM --> KNOW
                KNOW --> WIS
                WIS --> INTEL
            end
        end

        API[MCP Interface]
    end

    subgraph INFRA["  Infrastructure  "]
        direction LR
        GR[(Graph Store)] ~~~ VE[(Vector Store)] ~~~ CA[(Cache)]
    end

    A1 <--> API
    A2 <--> API
    API <--> META
    META <--> INFRA
    
    style AGENTS fill:#1f2937,stroke:#58a6ff,color:#f0f6fc
    style DELTA fill:#1f2937,stroke:#a371f7,color:#f0f6fc
    style META fill:#1f2937,stroke:#d29922,color:#f0f6fc
    style COGNITIVE fill:#161b22,stroke:#3fb950,color:#f0f6fc
    style INFRA fill:#1f2937,stroke:#f78166,color:#f0f6fc
Loading

Get Started

Repository Purpose
primitives EAG schema library (Apache 2.0)
engine Local engine, no cloud required (Apache 2.0)
mcp MCP client for hosted service

Closed beta now open. Join the waitlist or reach out for access.


"Memory without meaning is just metadata."

Engrammic • 2026

Pinned Loading

  1. primitives primitives Public

    Python 1

  2. engine engine Public

    Python

  3. mcp mcp Public

    Rust

  4. skills skills Public

Repositories

Showing 5 of 5 repositories

People

This organization has no public members. You must be a member to see who’s a part of this organization.

Top languages

Loading…

Most used topics

Loading…