PLATFORM

A runtime platform for governed agentic execution.

Subbasis brings agents, durable workflows, model routing, connectors, context, evidence, local execution and federation into one governed runtime layer.

Agents->
Runtime->
Workflows->
Systems
ContextEvidencePolicy

TECHNICAL THESIS

Probabilistic intelligence needs deterministic execution.

Subbasis is built around a deterministic distributed runtime designed to turn AI-driven work into durable, supervised and reviewable execution. Agents may reason probabilistically; the system that coordinates actions, approvals, context, connectors and evidence must remain deterministic.

Deterministic distributed runtime

A runtime core designed to coordinate execution, action state and persistence across nodes without turning every LLM decision into an uncontrolled side effect.

Context Intelligence

Context becomes part of the execution state, moving with the workflow across agents, approvals, systems and local execution.

ACC open connector contract

A connector contract proposal that wraps tools, MCP servers, HTTP / gRPC APIs and internal systems into governed execution boundaries.

Federated durable execution

A path toward workflows that coordinate across nodes, organizations and execution locations while preserving supervision, evidence and control.

RUNTIME CORE

The deterministic core behind agentic execution.

At the center of Subbasis is a Rust-based runtime core responsible for execution, coordination and persistence of actions across the system. It separates probabilistic reasoning from deterministic execution, so agents can act without turning the business process into an uncontrolled chain of side effects.

  • Rust-based runtime core
  • Deterministic execution path
  • Distributed coordination
  • Persistent action state
  • Evidence-aware execution
Intent / agent decision->
Action contract->
Runtime core->
Execution / evidence

WORKFLOW & ORCHESTRATION

Durable execution for real business workflows.

Subbasis is designed around durable execution principles: workflows can wait, resume, retry, escalate and continue over hours, days, weeks or months — while keeping state, context and supervision attached.

  • Wait and resume
  • Human approval
  • Retry and recovery
  • Long-running business processes
  • API-driven workflow execution
Started->
Waiting->
Approval required->
Agent resumed->
Retry handled->
Evidence recorded

CONTEXT INTELLIGENCE

Context is not just memory. It is operational state.

Subbasis keeps the execution context attached to the workflow: request, rules, decisions, approvals, failures, resumes, system state and evidence. This makes sophisticated multi-agent work possible without losing the thread when the workflow spans tools, people, models or local execution.

  • Context preserved
  • Decisions remembered
  • Approval state attached
  • Execution history
  • Own data control
  • Local model continuity
Workflow step
Context object attached
Next step with state

MODELS & INTELLIGENCE

Orchestrate the right model resources for each task.

Subbasis treats models as governed execution resources. A single workflow can use premium remote models, BYOD providers, specialized models or local LLMs depending on the task, policy, cost, privacy and execution context.

  • Local LLM available
  • Governed model orchestration
  • Per-task model binding
  • BYOD provider support
  • Cost-aware execution
  • Private model workflows
  • Fallback and provider control
Task->
Model router->
Local LLM->
Hosted model->
Specialized provider
PolicyCostPrivacyFallback

CONNECTORS & ACC

MCP connects tools. ACC governs execution.

MCP is useful because it standardizes how models reach tools and context. ACC goes further: it wraps MCP tools, HTTP / gRPC APIs, SaaS connectors and internal services into governed execution contracts with permissions, policy and evidence.

  • ACC open connector protocol proposal
  • MCP tools
  • HTTP / gRPC APIs
  • Custom connectors
  • Internal services
  • Evidence-bound tool use
External tools->
ACC->
Runtime->
Policy / Evidence

LOCAL & OMNICHANNEL EXECUTION

Execution can happen where the work is.

ARIN enables persistent local execution close to customer systems, while the messaging layer opens the path for omnichannel workflow entrypoints — from internal apps to chat, WhatsApp, Telegram or support channels.

Omnichannel entrypoints and messaging-driven execution are an active path, not a blanket production claim.

  • ARIN local execution
  • Internal APIs and files
  • Persistent local workflows
  • Matrix-based messaging hub
  • WhatsApp -> agent -> workflow
  • Governed execution contracts
Channel / local system->
Agent->
Workflow->
Evidence / response

FEDERATION

From runtime to execution fabric.

Federation is Subbasis' path toward multi-node and multi-organization execution: workflows that can operate across environments while preserving supervision, evidence, policy and control.

  • Multi-node execution
  • Cross-environment workflows
  • Organization boundaries
  • Remote trust / policy
  • Federated evidence
  • Controlled distributed execution
Node A<->
Node B<->
Node C

Policy / Evidence layer

EVIDENCE, POLICY & CONTROL

Execution remains governed by design.

Every important operation can be supervised, approved, logged and reviewed through policy and evidence controls. Subbasis does not claim certification by default; it provides the technical primitives that regulated environments need to build toward certification and audit readiness.

  • Permissions
  • Approval gates
  • Policy checks
  • Evidence records
  • Reviewable history
  • Audit-ready posture
Action->
Policy check->
Approval->
Evidence record->
Review

BUILD ON THE PLATFORM

Start with one workflow. Scale into governed execution.

Use Subbasis as the execution layer for agents, teams, systems, models, connectors and business operations.