Architecture Intermediate 1 minute read Updated 2026-06-28 UTC

Browser-Local Breeding Workbench

A practical architecture for teaching model breeding in the browser with file-backed manifests, local dashboards, deterministic simulations, and future WASM model packages.

Research statusIntegrates client-side visualization, TinyRustLM source notes, and zero-dependency Rust LLM reports Publication statePublished Reviewed byMichael Kappel Source reports4

Answer first

The first ModelBreeder workbench should run in the browser, use local files, and teach the full breeding cycle without requiring a cloud service. The site can start with deterministic simulations and later attach real .slm packages, adapter deltas, and WASM runtime diagnostics.

Components

ComponentRole
File-backed registryStores model cards, Genome records, FitnessVector records, and release packets.
Browser dashboardShows population state, fitness trends, novelty, lineage, and next actions.
WASM runtime bridgeLoads tiny model packages and adapter deltas when the lab graduates from simulation.
Manifest verifierConfirms artifact digests, tokenizer identity, tensor layout, and adapter compatibility.
Release packet builderConverts evidence into a copyable promotion record.

Why browser-local matters

The source reports emphasize edge compute, privacy, low latency, and local sovereignty. A browser-local workbench teaches those principles by design: no database, no cloud requirement, and no third-party scripts. The user can inspect the artifacts directly and understand how a population changes over time.

pseudocode
PROCEDURE open_workbench(local_folder)
    registry <- LOAD_MANIFESTS(local_folder)
    dashboard <- RENDER_POPULATION(registry.population)
    IF wasm_runtime_available THEN
        runtime <- LOAD_RUNTIME()
        VERIFY_MODEL_PACKAGES(registry.models)
    END IF
    ENABLE_LOCAL_EXPERIMENTS(dashboard, registry)
END PROCEDURE

Source reports used for this guide

These reports are preserved verbatim in the site archive. The guide above is an editorial synthesis and may narrow, qualify, or reorganize claims from the source material.