DSdistributed.systems
Infrastructure for autonomous systems

The operating layer for autonomous work.

Distributed Systems builds the runtime infrastructure that routes models, governs tools, supervises agents, and preserves state—from one machine to a fleet.

Open Router ↗View systems ↓
An inspectable stack for systems
that need to keep working.
05Integrated systems
C + BEAMNative runtimes
1 → fleetOperating topology
01 / Systems

Five systems.
One operating model.

01DSCO HarnessThe native C harness for cross-provider agents, governed execution, durable state, and local-first control.Harness ↗ 02DSCO RouterOne model gateway for provider-independent routing, credentials, metering, budgets, and policy.Control ↗ 03Tool Management APIBounded capabilities behind one discoverable, inspectable registry and execution surface.Tool plane ↗ 04agents.erlA BEAM-native runtime for supervised, concurrent agents with message passing and fault isolation.Runtime ↗ 05GraphSubA graph database in Rust for coordination, topology, state, and credit across machine fleets.Database ↗
02 / Architecture

A system, not a bundle.

Explicit boundaries connect planning to inference, tools, supervised execution, and durable state. Each component works independently; together they form the control loop.

Direct work
DSCO Harness

Plans, dispatches, evaluates, and retains local control over execution.

  • requestaccepted
  • policyloaded
Governed access
Router + Tools

Select models and capabilities within credentials, budgets, and authority.

  • routeresolved
  • capabilityscoped
Operate + remember
agents.erl + GraphSub

Supervise concurrent work while retaining topology, state, and provenance.

  • processsupervised
  • recordcommitted
Operational principles

Provider credentials, tool authority, budgets, routes, state transitions, and outcomes remain visible operating concerns—not hidden implementation details.

Control stays explicit

Every route and capability operates inside declared authority and policy.

Evidence travels with work

Execution produces records that can be inspected, replayed, and evaluated.

Topology is a runtime choice

The same boundaries hold from a local process to a supervised machine fleet.

03 / Research

The research behind the operating model.

Apart Research
2025
Detecting Piecewise Cyber Espionage in Model APIsFragmented attacks evade request-level defenses; full-context analysis restores the missing security boundary.
Read ↗
Apart Research
2026
LidaSim: Testing AI Policies with Persona-Based SimulationsA simulation environment for evaluating policy interventions against heterogeneous, strategically adapting agents.
Read ↗
Preprint
Research program
Implementing Self-Improving Reasoners using DSPyProgrammatic optimization for reasoning systems that improve against measurable objectives.
Preprint
Research article
July 6, 2024
Object-Oriented Reinforcement Learning in Mutable Ontologies with Self-Reflective Meta-DSLA framework for open-ended learning in evolving object–dyad graphs, where message-passing objects inspect and modify their state, goals, beliefs, methods, interactions, and learning strategies.
Read ↗
04 / Company

Built for the systems after the demo.

Distributed Systems is an independent infrastructure company. We build and operate the DSCO stack, publish the research behind it, and work directly with teams moving autonomous systems into production.

Use the stackStart with hosted routing and tool surfaces, or deploy the native runtimes and database.
Build with usArchitecture, integration, reliability, and specialized agent systems for demanding environments.

Bring us a real workload.

Tell us what it needs to do, where it must run, and what cannot fail.

Discuss an implementation ↗