Founder & AI Systems Architect · Direct Autonomy Systems
I build bounded AI agents and automation that move real work forward while keeping evidence, recovery, and human authority visible.
Work with DAS · Start a conversation
Capabilities · Experience · Foundry · Portfolio · Doctrine · Contact
Direct Autonomy Systems (DAS) is an independent AI systems company and R&D foundry. I design agents, automation, research infrastructure, developer tools, and simulated worlds around one governing rule:
Models propose. Deterministic systems admit, authorize, verify, and commit.
The aim is delegated intelligence, not machine sovereignty: software that can reason, remember, coordinate, and act within explicit authority while keeping evidence, uncertainty, cost, side effects, recovery, and human control visible.
DAS is not a collection of unrelated chatbot demos. Each project investigates a distinct systems problem. Proven laws, components, evaluation suites, and operating patterns can be promoted into shared infrastructure—but private memory, secrets, identity, and earned authority remain isolated.
DAS works on problems where AI, automation, or custom software can create meaningful leverage without hiding responsibility behind a model:
- Bounded AI agents with explicit tools, permissions, identity, memory, budgets, interruption, and recovery.
- Workflow automation and custom integrations across APIs, operational data, business processes, and human approval.
- Evidence-backed research and monitoring with provenance, freshness, contradiction handling, and traceable synthesis.
- Discord-native systems for agents, voice collaboration, simulations, games, and persistent community experiences.
- Agent infrastructure and developer tooling for context, orchestration, evaluation, release evidence, and safe handoffs.
- Deterministic simulations and custom software where state, causality, replay, and explainability matter.
The starting unit is not “add AI everywhere.” It is one bounded workflow, one accountable owner, one success contract, and evidence that the system worked.
The best agent experience should feel less like managing a chatbot and more like collaborating with a capable system that knows its limits. I design for:
- Continuity without repetition. Preserve the context that matters while keeping identity and private state scoped.
- Initiative without surprise. Move work forward inside explicit authority; stop cleanly when a decision belongs to a person.
- Progress without babysitting. Surface state, evidence, uncertainty, cost, and the next useful action.
- Control without friction. Make interruption, revocation, retry, recovery, and rollback part of the experience.
- Results with receipts. Connect outputs to sources, decisions, versions, changes, and verification.
The magic is not hidden autonomy. It is useful initiative with a trustworthy handback.
Aureon governs action. Mira creates presence. Spider Sense grounds claims. Automata coordinates change.
The DAS agent-runtime and control-plane foundation: explicit permissions, typed admission, bounded actions, verification, receipts, revocation, recovery, and rollback. It is where useful model proposals become—or fail to become—authorized system behavior.
A Discord-native AI operations room for voice, research, memory, image and audio creation, progress, interruption, and multimodal collaboration. Mira explores how an agent becomes responsive, understandable, and useful without becoming noisy or presumptuous.
A local-first investigation and evidence system for crawling, ingestion, graph reasoning, source ranking, contradiction checks, freshness, and traceable synthesis. Spider Sense helps agents distinguish observation, claim, inference, conflict, and unresolved uncertainty.
A bounded coordination layer for coding agents and complex projects: work scopes, edit leases, mailboxes, architecture memory, impact analysis, handoffs, review briefs, and proof-aware operator surfaces. Improvements become reviewable candidates—not silent self-modification.
These systems test the DAS doctrine under different pressures rather than presenting the same chatbot in different skins.
- Praxeon & Telaxis — living governance worlds. Paired Discord-native NationStates experiments exploring autonomous policy, constitutional limits, public records, persistent consequence, and the difference between optimizing outcomes and preserving legitimate authority.
- Agent Sandbox — inspectable emergence. A deterministic living world where pressure, experimentation, memory, teaching, trade, institutions, construction, and culture can compound into visible civilizational change without allowing generated narration to become world truth.
- Inkwake — a simulation-first RPG oracle. A browser-native persistent world where player language creates proposals, deterministic simulation decides what can become real, and every admitted consequence remains explainable and replayable.
- Ghost Writer / Ghostline — writer-led continuity. A quiet browser writing layer that offers one reversible, editor-local continuation and commits text only after explicit, current-context acceptance.
- PixelForge — representation without authority. A deterministic, engine-neutral visual asset compiler that turns approved semantic requests into validated, content-addressed image bundles while remaining unable to decide what exists in the world.
View the complete DAS portfolio
- Aureon-Hermes — a DAS-hardened private agent platform built on the Hermes Agent foundation, focused on authority, execution safety, operator control, evidence, recovery, and release proof.
- Mira / Agentic Discord — the Discord-native presence layer for text, voice, research, memory, image generation, music, sound effects, progress cards, and human interruption.
- Avarice — an internal commercial-operations experiment for bounded lead, opportunity, and revenue-support workflows; it is not an independent financial authority.
- Spider Sense / RedThread — autonomous web investigation, evidence ingestion, live graphs, GraphRAG, contradiction analysis, and source-backed synthesis.
- Automata / Clockwork Automata — scoped coding-agent coordination, architecture memory, work leases, mailboxes, impact analysis, proof, and an operator cockpit.
- Context Hub / Project OS — a continuity compiler that separates committed repository truth, safe project context, active workspace state, evidence, and handoff packets without pretending they are interchangeable.
- Reddit Monitor — a local-first browser sensor that turns rendered Reddit discussions into bounded, versioned observations and durable change signals; Spider Sense remains the downstream evidence authority.
- Portfolio R&D Decision Engine — a recurring research and prioritization loop that weighs release unblock, user value, proof, differentiation, reuse, and revenue leverage against complexity, time, cost, and risk.
- Agent Sandbox — deterministic agent society, physical economy, construction, institutions, precedent, governance, and inspectable emergence in C# and Godot.
- Praxeon — a constitutional NationStates intelligence emphasizing citizen agency, accountable power, consequence lineage, and tamper-evident public records.
- Telaxis — Praxeon’s distinct sovereign-optimization counterpart: a separate identity and directive used to test how the same substrate behaves under different governing aims.
- Inkwake — a browser-based living ink world where prompt-driven action remains subordinate to deterministic state admission.
- Echoes of the Veil — an earlier deterministic simulation laboratory whose lessons feed later work in world state, causality, agents, and emergent narrative.
- Ghost Writer / Ghostline / Continuum — intent-aware writing continuation with editor-local previews, bounded context, strict foreground authority, and explicit Tab acceptance.
- Ghostpaste — a destination-aware copy-and-paste assistant built around the loop “Copy. Switch. Recognize. Tab.” with local secret gating and target-locked delivery.
- PixelForge — deterministic compilation, validation, lineage, and delivery of pixel-art representation for games and simulations.
- Ghostwrite Web — the accessible, dependency-light product surface for Ghost Writer, including interactive demonstration and deployment tooling.
- Aureon Linux-native Proof Lane — a separate evidence-closing lane for exact Linux runtime validation. It proves one bounded subject only; it does not collapse CI, deployment, serving, or live operation into one claim.
- DAS profile and brand system — this public company and portfolio surface, designed to become the entry point for independently scoped demos, case studies, release evidence, and client work.
The portfolio is designed to convert difficult project lessons into reusable leverage:
problem found in one system
→ bounded implementation
→ deterministic test and proof
→ reusable law, pack, or evaluation
→ proof in a second distinct system
→ product or client outcome
Products may inherit substrate, doctrine, tools, packs, and evaluations. They do not inherit private user state, secrets, identity, or trust that another system earned. A model is a replaceable capability inside the architecture—not the architecture itself.
- Speech is not authority. A model response is a proposal, not truth, permission, memory, or an irreversible action.
- State has an accountable owner. Canonical transitions are admitted by deterministic systems with explicit invariants.
- Evidence stays visible. Claims, sources, versions, decisions, fallbacks, and proof boundaries remain inspectable.
- Agency stays bounded. Permissions, identity, memory, spend, retries, time, and side effects are scoped deliberately and default to denial.
- Recovery is part of the product. Idempotency, cancellation, revocation, restart safety, rollback, and degraded modes are designed in—not added after failure.
- Proof classes remain separate. Local tests, exact-head CI, deployment, serving health, and live user outcomes are different claims and require different evidence.
- Failures become reusable protection. Important incidents should leave behind a reproducer, regression test, fix, and versioned safeguard.
Current build priorities
- Harden the trust and execution substrate so authority, evidence, recovery, and release proof become reusable across distinct systems.
- Make existing capability immediately legible through focused first-minute experiences, visible state, interruption, receipts, recovery, and public demonstrations with exact scope.
- Turn proven project laws into customer value through reusable packs, integrations, and paid outcomes under Direct Autonomy Systems.
- Languages: C# · TypeScript · Python · JavaScript
- Frameworks and runtimes: .NET · Node.js · React · Godot · browser extensions
- Platforms and infrastructure: Discord · REST APIs · GitHub Actions · Docker · Linux/VPS · SQLite
- Systems work: agent harnesses · voice and multimodal UX · deterministic simulation · evidence graphs · browser automation · evaluation and proof pipelines
Most portfolio repositories are private while they are actively developed, hardened, and prepared for responsible release. This profile communicates the product direction, company mission, architecture, and engineering doctrine without exposing credentials, private memory, customer information, unsafe operational details, or unreviewed implementation internals.
A public description is not a claim that a system is deployed, generally available, independently verified, or licensed for reuse. Public demos, case studies, documentation, and source releases will be linked only when their exact scope and evidence are ready to stand on their own.
DAS is open to thoughtful conversations with founders, builders, engineers, and business owners working on serious agent systems, workflow automation, research tooling, Discord experiences, simulations, developer infrastructure, or ambitious custom software.
founder@directautonomy.com · Start a public conversation
Please do not include confidential, credential-bearing, or sensitive information in public discussions.