green rose systems

“growth, forever”
A Software Studio · Established 2026

substrate

A Python runtime that coordinates model calls and computations through one append-only log.

01  ·  description

substrate is a Python 3.12+ runtime for running many computations together. Model calls, transforms, subprocesses, parsers, simulators, anything that takes typed input and emits typed events. It runs them concurrently and coordinates them through one append-only log. Every event, and every runtime decision about what to start next, lands on that log. The log is framed, CRC-protected JSONL. Replay it, diff it, or inspect any point.

Nine named pieces carry the design. A Producer takes typed input and streams typed Events. It is a callable: an LLM, an ML model, a deterministic transform, a subprocess, a parser. Events land on the Bus, one totally-ordered append-only log per run. Producers coordinate only through it. A View maintains a running summary over the log. A Predicate is a yes/no over the Views, evaluated when an event lands. A Trigger starts a new Producer when its Predicate holds. After the initial Producers, Triggers are the only way new ones appear. A Route carries data from past events into a future Producer’s input. A TerminationPolicy decides when the run ends or pauses for outside input. The bus writes to a run record on disk.

These pieces compose into topologies whose shape grows as the run unfolds. Recursion falls out naturally because a Trigger can start its own kind. Eight topologies ship with committed records. code_review runs several cheap models on the same task, has a stronger model adjudicate, and cancels the losing runs when the verdict lands. coding_flow streams code from a writer Producer while a checker Producer fires on each complete declaration as it arrives; writer and checker run at the same time, and the check is real (ruff check, mypy strict, pytest on the writer’s output). debate puts two model Producers on opposite sides of a claim; the run ends on a convergence condition. adversarial_pair runs a writer and an attacker streaming from t=0. planner_solver emits subtasks that each start solvers that can themselves emit more subtasks. simulation runs many Producers each tick against a shared world-state Producer. tool_loop chains model, tool, model as a sequence of Producers, each call independently replayable. retry_pipeline feeds the failure reason into the next attempt, escalates after N tries, and pauses for a human when it cannot recover.

On PyPI as substrate-kernel. PolyForm Noncommercial 1.0.0.

02  ·  repository

github.com/laffeyp/substrate

Contact

We’re open to a small number of new engagements. Write with a short note about the work and the timeline.

info@greenrosesystems.com