AI Infrastructure for the Next Decade

We build the stack
that makes AI learn

Your AI doesn't forget. It evolves. From continual learning engines to skill marketplaces to managed hosting — we build the infrastructure that makes small models brilliant over time.

Our CL Engine Research & Valuation

Our Thesis

"Scaling context windows to millions of tokens isn't solving continual learning — it's buying your way around it. If AGI only works at datacenter scale, it's not general intelligence, it's expensive intelligence. True generality means the architecture itself supports learning over time."

— The problem we're solving

MARKET CONVERGENCE

The world is building toward this

Every major lab and analyst firm is independently arriving at the same conclusion: AI needs memory, deliberation, and edge-native architecture. Here's what they're saying.

Google Research — NeurIPS 2025

"Nested Learning" & the HOPE Architecture

Google's Nested Learning paradigm reframes AI models as nested, multi-level optimization systems — fast-updating modules for immediate context, slow-updating modules for long-term knowledge. Their proof-of-concept HOPE architecture demonstrates continual learning without catastrophic forgetting. This is exactly the architecture philosophy Graphonomous implements at the infrastructure level: multi-timescale memory consolidation, edge-native, no retraining required.

Behrouz & Mirrokni, NeurIPS 2025 — "Nested Learning: The Illusion of Deep Learning Architectures"

ACL 2025 + arXiv 2025

Structured Consensus > Opaque Confidence

A wave of 2025 research (ACL Findings, multiple arXiv papers) demonstrates that multi-agent deliberation protocols — where AI agents argue, critique, and iterate toward consensus — consistently outperform simple voting and single-agent approaches. Deliberatic operationalizes this: Dung's argumentation framework extended into weighted bipolar systems with Byzantine fault tolerance and Merkle-chained evidence. This isn't theoretical — it's the deliberation infrastructure the research says agents need.

Kaesberg et al., ACL 2025 — "Voting or Consensus? Decision-Making in Multi-Agent Debate"

Industry — 2025/2026

MCP + A2A: The HTTP Moment for Agents

Anthropic's MCP (agent-to-tool) and Google's A2A (agent-to-agent) are establishing the equivalent of HTTP for AI agents. MCP standardizes tool access; A2A standardizes agent collaboration. Gartner predicts 40% of enterprise apps will embed AI agents by end of 2026. The entire [&] stack is MCP-first and A2A-ready — not as an afterthought, but as the foundational design decision. Every product in the portfolio speaks these protocols natively.

Gartner 2025; Linux Foundation Agentic AI Foundation; A2A Protocol (Google + 50 partners)

VentureBeat (Jan 2026): "Continual learning shifts rigor toward memory provenance and retention… The winners will not only pick strong models, they will build the control plane that keeps those models correct, current, and cost-efficient." — That control plane is what we're building.

Hero Products

Graphonomous

graphonomous.com
In Development

A continual learning engine that makes small language models (1B–8B) get smarter over time in their deployment context. Self-evolving knowledge graphs that learn, consolidate, and prune at inference time — no retraining, no cloud dependency, no catastrophic forgetting.

Learn

Graph-based memory that grows with every interaction. Episodic, semantic, procedural, and temporal knowledge — structured relationships, not flat vectors.

Adapt

Multi-timescale consolidation inspired by the brain's sleep cycles. Fast memory promotes to slow memory. Weak connections decay. Strong patterns crystallize.

Decide

MCP-first API means any AI client can connect directly. Graph-aware context injection makes every inference smarter than the last.

Visit Graphonomous →
Elixir/OTP SQLite + sqlite-vec MCP Server Edge-Native

Deliberatic

deliberatic.com
In Development

An open-source deliberation protocol for multi-agent AI. Structured argumentation, Byzantine fault-tolerant consensus, constitutional guardrails, and Merkle-chained evidence chains — giving agent clusters auditable, principled decisions instead of opaque confidence scores.

Argue

Extends Dung's argumentation framework into a weighted bipolar system. Agents submit formal positions with typed evidence. Attack and support relations are explicit, not implicit.

Decide

Two-phase adaptive consensus — Raft fast-path when the winner is clear, PBFT conflict-path when positions are close. Byzantine fault tolerant.

Prove

Merkle-chained evidence logs, constitutional governance with hard boundaries and soft preferences, and vindicated dissent tracking. Every decision ships a proof.

Visit Deliberatic →
Elixir/OTP A2A + MCP DAF Semantics Apache 2.0

The Ecosystem

Every product connects. Every domain has a role. Together they form the infrastructure for AI that learns over time.

Memory

Graphonomous

graphonomous.com

A continual learning engine for edge AI. Self-evolving knowledge graphs that make small models brilliant — no retraining, no cloud, no forgetting.

Explore →
Decisions

Deliberatic

deliberatic.com

The decision protocol. Formal argumentation semantics, Byzantine-tolerant consensus, constitutional guardrails, and Merkle-chained evidence chains for auditable agent decisions.

Explore →
Runtime

OpenSentience

opensentience.org

The agent runtime. Manages agent lifecycles, permissions, tool routing, and MCP server connections. The OS for intelligent agents.

Explore →
Skills

FleetPrompt

fleetprompt.com

The skill marketplace. Install CL strategies, agent workflows, and domain-specific knowledge as versioned packages.

Explore →
Orchestration

Delegatic

delegatic.com

Multi-agent delegation and orchestration. Coordinate learning agents across teams with systematic task routing and governance.

Explore →
Specifications

SpecPrompt

specprompt.com

Spec-driven agent development. Define capabilities, contracts, and acceptance criteria before building. The source of truth for agent behavior.

Explore →
Agent Builder

Agentelic

agentelic.com

Premium agent creation platform. Build, test, and deploy individual AI agents with built-in reputation scoring and capability profiles.

Explore →
Automation

AgenTroMatic

agentromatic.com

Automatic multi-agent workflows with deliberation. Agents bid on tasks, argue fitness, and self-organize — no human orchestrator required.

Explore →
Hosting

WebHost Systems

webhost.systems

Managed Graphonomous instances. PostgreSQL-backed, multi-tenant, with monitoring dashboards and usage-based billing.

Explore →
Spatial

GeoFleetic

geofleetic.com

Geo-distributed fleet intelligence collaboration. Vehicles and devices that learn routes, predict demand, and optimize logistics over time.

Explore →
Temporal

TickTickClock

ticktickclock.com

Time-series continual learning. Detect patterns, anomalies, and trends that evolve over time — the temporal intelligence layer.

Explore →

ARCHITECTURE

One system, not ten products

Every product in the portfolio occupies a specific layer. Together they form a complete infrastructure for AI that learns, decides, and acts.

Interface
Define agent behavior with specs and contracts
SpecPrompt
Creation
Build and deploy individual agents
Agentelic BendScript
Orchestration
Coordinate multi-agent teams and workflows
Delegatic AgenTroMatic
Deliberation
Structured argumentation and consensus for decisions
Deliberatic
Memory
Continual learning engine — the knowledge substrate
Graphonomous
Distribution
Package and distribute CL strategies as skills
FleetPrompt
Intelligence
Temporal and spatial intelligence layers
TickTickClock GeoFleetic
Infrastructure
Managed hosting and multi-tenant deployment
WebHost Systems

Built on Solid Foundations

Elixir / OTP

Fault-tolerant concurrency

Phoenix

Real-time LiveView

SQLite + pgvector

Edge to enterprise

Hermes MCP

Industry-standard protocol

Bumblebee / Nx

Local ML inference

Fly.io

Global edge deployment

PHILOSOPHY

What we believe

1. Memory is infrastructure, not a feature

Intelligence without memory is just expensive pattern matching. Real learning requires structured, persistent, evolvable knowledge — not bigger context windows. We build memory as a first-class infrastructure layer, not a bolted-on retrieval system.

2. Decisions should ship with proofs

When agents make decisions, you deserve to know why. Not a confidence score — an auditable chain of evidence, arguments, and constitutional constraints. Deliberatic exists because opacity is a liability, not a feature.

3. Small models, long horizons

The future isn't a single trillion-parameter model running in a datacenter. It's millions of small, specialized models that get smarter in their deployment context over weeks, months, and years. Edge-native. Always learning. Never forgetting.

4. Open protocols, open standards

MCP-first. A2A-ready. Apache 2.0 where possible. We build on open standards because vendor lock-in is the opposite of interoperability, and interoperability is the whole point of an ecosystem.

OpenSentience.org
Our open research initiative exploring the boundaries of machine cognition, structured memory, and what it means for an AI system to "understand." Because the questions matter as much as the products.