Pebble Dynamics

Pebble Dynamics

Autonomous engineering
intelligence.

We build an AI agent framework in Rust — layer by layer, for systems that have to be correct, governed, and durable.

Request a technical brief › Learn more ›

What we build

A framework for building agents that can be reasoned about — and audited.

Most AI systems are written for demos. Ours is written for production: a Rust workspace with strict layer boundaries, deterministic builds, and machine-checkable quality gates that run on every change. The result is a substrate for autonomous engineering work that holds up under scrutiny — by reviewers, by regulators, and by the next engineer to read the code.

87
Rust crates
5
Architectural layers
8
Quality gates per change
42001
ISO/IEC aligning to

Architecture

Five layers. One direction of dependency.

Lower layers cannot depend on higher ones. The rule is enforced in code, not in documentation, and verified on every commit. It is what lets a system this size remain comprehensible.

  1. Immutable
    Shared types and protocol definitions. Zero runtime dependencies. The contract that everything else honours.
  2. Platform
    Foundational primitives: storage, transport, telemetry, cryptography. No business logic.
  3. Domain
    Reasoning, consensus, orchestration, learning. The intelligence of the system. No HTTP, no I/O surfaces.
  4. Interface
    User-facing surfaces: APIs, CLIs, dashboards. No machine learning or consensus logic ever lives here.
  5. Infrastructure
    Deployment, environments, observability. Configuration, never application code.

immutable → platform → domain → interface

ISO/IEC 42001

An AI Management System, by design.

ISO/IEC 42001 is the international standard for AI Management Systems. Our compliance posture is built into the engineering process, not bolted on at audit time.

  1. 01

    Risk identification

    Every domain system is registered with its purpose, data inputs, and potential harms. New components cannot ship without a risk record.

  2. 02

    Data Protection Impact Assessment

    For systems processing personal or regulated data, a DPIA is generated from machine-readable templates and reviewed before deployment.

  3. 03

    Controls & quality gates

    Eight automated gates run on every change: provenance, format, compilation, lint, layer integrity, dependency policy, size limits, and tests.

  4. 04

    Continuous audit

    Decisions, learnings, and provenance are persisted as structured records. The audit trail is a by-product of how the system is built, not a separate document set.

Contact

Request a technical brief.

For grant programmes, accelerators, and collaborators evaluating our work — tell us a little about your interest and we'll walk through the architecture, compliance posture, and roadmap.