v0.4.0 - Simulation Bridge

Released: May 7, 2026

From Words to Worlds — describe a system in natural language, BERT builds and runs it.

Highlights

This release transforms BERT from a visual modeling tool into an end-to-end systems workbench. You can now describe a system in plain language and watch BERT generate a structured model, transpile it into a typed graph, and run an agent-based simulation. Three months, 89 commits, and the biggest feature expansion since launch.

Features

Natural Language Model Creation

The new landing screen greets you with a creation chat. Describe a system — "a coffee shop with supply chain, customer flow, and waste management" — and BERT generates a full model: subsystems, flows, interfaces, boundary, and archetype classifications. Powered by three-tier RAG routing through the bert-rag knowledge engine.

TypeDB Transpiler

bert-typedb is a new CLI tool that converts BERT JSON models into typed graphs in TypeDB. One command transpiles a model into ~140 TypeQL statements, preserving the full 8-tuple structure. This enables cross-model queries, structural comparison, and serves as the bridge to simulation.

Mesa Simulation Bridge

Press Simulate and BERT spawns a Python Mesa simulation derived from your model's structure. Agent behavior is driven by archetype (Economy, Governance, Agent), agency capacity, and flow topology. Results stream back through TypeDB and render as time-series charts in the UI.

Simulation Dashboard

A new bottom-dock dashboard shows the system state vector v(t) during simulation. Each subsystem's state variables appear as horizontal bars with direction indicators (growing, shrinking, stable). First of a planned five-panel observatory (State, History, Flows, Transformation, Hierarchy).

Process Primitive T Functions

All 9 Mobus process primitives (Buffering, Combining, Splitting, Propelling, Impeding, Sensing, Modulating, Inverting, Copying) now have executable T functions. Each is implemented as a pure state transform, designed for portability from Python to Rust. A suite of 9 minimal test models exercises each primitive in isolation.

Agent System

Systems can now carry an AgentModel with behavioral classification (Reactive, Anticipatory, Intentional), agency capacity (0.0–1.0), and process primitive assignments. This bridges the structural model to executable simulation parameters.

For Contributors

BERT is now structured for contribution:

• Lib/bin split (#43) — the bert crate is a library that tools can depend on

• bert-typedb — extend the transpiler to emit new relation types

• bert-generate — standalone CLI for headless model generation

• Test primitive suite — 9 models in assets/models/local/test-primitives/ for validating simulation changes

• System Language Specification v0.1 — formal reference for what valid models look like

Issues tagged good-first-issue and bounties available via merit.systems.

Technical

Layer	Stack
Rendering	Bevy 0.17 (ECS)
UI	Leptos 0.8 (reactive Rust)
Desktop	Tauri 2
Simulation	Python Mesa + TypeDB 3.x
Knowledge	bert-rag (vector + KG hybrid retrieval)

What's Next

• Primitive dispatch via TypeDB — transpiler emits has_primitive relations so simulation behavior is driven by model structure, not just archetypes

• Hierarchical time stepping — child systems step at their own Δt relative to parent

• History as computational state — H conditions future transformations (Buffering learns from overflow patterns)

• Dashboard panels — History sparklines, Flow visualization, Transformation inspector