Level 5 — Runnable Vertical Slice. The repository currently supports only the Simple Charm profile. This baseline exists to verify that the canonical Manufacturing Plan architecture works end to end; it is not feature complete and is not yet a production manufacturing system. See the Roadmap for the delivery sequence and current scope.
Open Manufacturing Planner is a design-first, open-source framework for turning a manufacturing request into a reviewable, deterministic manufacturing workflow.
The framework treats the Manufacturing Plan as its only intermediate representation. An LLM may propose a plan, but it must never emit SVG, CAD, CAM, geometry, or machine instructions. Deterministic engines validate the plan and derive downstream artifacts.
Requires Python 3.12 or newer.
git clone https://github.com/restructure-git/open-manufacturing-planner.git
cd open-manufacturing-planner
python -m pip install -e .
omp validate manufacturing-request examples/requests/simple_charm.json
omp demo examples/requests/simple_charm.json --output artifacts/demoThe demo produces manufacturing-plan.json, deterministic geometry.json, and deterministic template.svg under artifacts/demo/. Run python -m pip install -e '.[dev]' followed by python -m pytest -q for self-validation.
Current limitations: one rectangular vegetable-tanned-leather bag charm, one circular attachment hole, millimetres only, no LLM, no manufacturing control, and a preview canonicalization profile. See Runnable State.
Manufacturing knowledge is often trapped in individual experience, one-off scripts, and opaque AI prompts. This project makes decisions explicit, typed, reviewable, reproducible, and progressively replaceable by rules learned from evaluated outcomes.
flowchart LR
R[Manufacturing Request] --> P[Manufacturing Plan]
P --> G[Deterministic Geometry]
P --> A[Manufacturing Artifacts]
A --> M[Manufacturing]
M --> E[Evaluation]
E --> K[Knowledge Base]
K --> X[Rule Extraction]
X --> U[Rule Engine]
U --> P
L[LLM Planner] --> P
ManufacturingPlanis the sole contract between planning and every downstream capability.- LLM output is structured as
proposal,reason,confidence, andalternatives. - Geometry and manufacturing artifacts are deterministic for identical plan, configuration, and engine versions.
- Modules exchange versioned contracts; they do not reach into one another's internals.
- Known cases are handled by rules. Only unknown or low-confidence cases fall back to an LLM.
- No physical execution is automatic: validated artifacts require an explicit human release gate.
- Phase 1 use case: leather bag charms.
- Later product families: coasters, wallets, and key holders.
- Long-term direction: a reusable leather manufacturing planning platform.
- Non-goals for the current roadmap: image generation, 3D modelling, CAM control, and laser control.
docs/: normative architecture and domain documentation.schemas/: versioned JSON Schema contracts and executable profiles.configs/: reviewed configuration profiles.examples/: valid and invalid executable fixtures.src/: installable Python package.tests/: schema, CLI, determinism and vertical-slice tests.scripts/: repository automation only; no manufacturing control.
Start with the documentation index, Architecture Audit, Manufacturing Request Specification v1.0, and Manufacturing Plan Specification v1.0. Terms such as plan, artifact, and release are normative as defined in the Glossary.
Project evolution is governed by the Autonomous Development Charter, evidence-based Quality Gates, a living Top 10 Issue Backlog, and the Project Knowledge Base. Commits, merges, and releases always require human approval.
The project develops vertical slices under architecture and quality gates. Each Issue must make the repository more executable or remove a named delivery blocker. See the Roadmap.
Read CONTRIBUTING.md, AGENTS.md, and SECURITY.md before proposing changes. The project is licensed under Apache License 2.0.
The selected name is open-manufacturing-planner because it is descriptive, vendor-neutral, product-neutral, and discoverable. Alternatives considered:
- manufacturing-plan-core — emphasizes the central contract, but sounds like a library rather than a platform.
- forgeplan-framework — memorable, but “forge” suggests metalworking and narrows the domain.
- manuplan-framework — concise, but less self-explanatory and more likely to be confused with existing brands.