Agentic Systems Group

Systems for research, learning, and long-horizon agency.

We build research systems for difficult domains: local-first terminals, tutoring infrastructure, and experiments in human-LLM cooperation where evidence, tools, and inspectable state matter more than presentation-layer magic.

What is live now.

The company site should reflect the actual work in the repos, not a frozen snapshot of an earlier pitch. These are the systems and programs that currently define the work.

Local-first research terminal

ThothTerminal

A serious always-on research terminal for heterogeneous sources, canonical storage, companies, earnings, transcripts, supply chains, and report-grade outputs.

Active system Open Web, desktop, MCP, Discord

Tutoring and control room

EdMneme

An interactive tutoring stack with a Control Room UI, inference routing, project memory, tool orchestration, and textbook ingestion through remote OCR.

Active system Open Tutoring, OCR, project memory

Long-horizon programs

Research

Current research directions now emphasize report-centric intelligence systems and Zeta, a simulation-heavy tactics RPG exploring lawful magic and local LLM companions.

Ongoing programs Open Simulation, evaluation, systems

How we build.

We do not want the company site to promise one thing while the codebase reveals another. The work is technical, structured, and opinionated, so the site now reflects that directly.

Inspectable over theatrical

We prefer systems that can show their evidence, expose their state, and survive scrutiny from operators who care about what the model actually did.

One durable core

The strongest product decisions in the current stack all push toward a shared system of record instead of fragile sidecars, duplicate stores, or prompt-shaped state.

Research feeds product

The long-horizon programs are not decorative R&D. They shape how we think about agency, control, evaluation, and human-machine collaboration.

Where the deeper work is going.

The product line is anchored in concrete systems, but the surrounding research matters: how to structure long-form research memory, how to think about institutional intelligence, and how humans and model agents can meaningfully share a bounded world.

  • Report-centric intelligence

    Inside ThothTerminal, research reports are becoming the canonical export surface for companies, technologies, components, and themes.

  • Institutional intelligence

    The longer roadmap pushes beyond dashboards toward continuously refreshed, evidence-dense views of companies, programs, products, and bottlenecks.

  • Zeta

    A preproduction tactics RPG and simulation project built around lawful magic, local-model companions, and one world model for humans and agents.

We are more interested in systems that can defend their outputs than in systems that merely look fluent.