SYS: PERSONAL_MODEL_FIELD X:--- Y:---

FIG. E1 Continuity and correction as a vector field.

SYS: ROUTING_MATRIX X:--- Y:---

FIG. E2 Routing decisions under explicit constraints.

Systems in the Loop

These projects are not showcases.

They are test benches for agent memory, routing, inference, and deployment.

Project

Inferoa

An inference-native agent harness for long-horizon tokenmaxxing.

Protects prefix-cache discipline, bounded context, and deterministic tool schemas.
Optimizes coding and research sessions through goal, plan, autoresearch, and tokenmaxxing modes.
Routes across self-hosted vLLM models and external frontier models under explicit pressure.

Website GitHub

Type Inference-native agent harness

Scope Prefix cache, context optimization, routing, and vLLM serving

Focus Long-horizon coding, autoresearch, and tokenmaxxing across the inference stack

Project

An open routing layer for choosing the right model, tool, and execution path.

Website GitHub

Type Open-source routing system

Scope Inference, model selection, and agent-facing control flows

Focus Cost, latency, privacy, safety, and execution quality

Flagship Project

A personal-model-first agent for durable, correctable memory.

Correctable user model instead of a hidden profile.
Curiosity that asks only when the answer changes future help.
Continuity across CLI, chat, dashboard, messages, tools, and background learning.

Website GitHub

Type Personal-model-first AI agent

Scope Personal Model, curiosity, memory, and continuity

Focus Correctable understanding, local-first operation, and user-paced learning

Research Fit

Start where agent systems break.

Build Surface

Interfaces beat internal demos.

Deployment Loop

Failure is input to the next paper.