CSL: AI Agent-Driven Meta-Parametric Design for CAD/CAE/CAM and Beyond

Why

Modern hardware programs span CAD, CAE, CAM, BOM, sourcing, factory automation, and project management. Designs change rapidly; syncing geometry, constraints, analyses, and manufacturing outputs across tools is fragile and manual. CSL exists to make complex, multi-domain engineering agent-friendly, reproducible, and automated.

What

CSL (CAD-Specific Language) is a compact DSL and JSON IR that describes geometry, constraints, selections, assemblies, and exports in a deterministic, parametric way. It’s designed for AI agents and humans to co-author designs that are portable across backends. The companion “Triple Lindy” agent converts natural language into CSL, plans across backends, and realizes results in target tools.

Key capabilities

• Meta-parametric design: parameters with units, ranges, and expressions; deterministic evaluation order.
• Robust selections: query by geometry/topology/tags/lineage for stable edits over time.
• Assemblies and joints: mate connectors/LCS, joints; stable instance IDs.
• Export and visualization: STEP/STL/… and thumbnails for downstream workflows.
• Backend capabilities: adapters declare feature support; the agent plans fallbacks.

Multi-backend

• Production ready:
  • Fusion 360 via Python API with real-time control (v12 - fully functional)
  • FreeCAD via Python API with workbench integration (v1.3 - 100% feature coverage)
• In development: Onshape (REST API/FeatureScript)
• Planned: SolidWorks (.NET/COM), NX, Blender (Geometry Nodes)
• Adapters publish capabilities so agents can pick the highest-fidelity route with graceful degradation.

Triple Lindy Real-time Control

• File-based protocol for reliable communication with CAD software
• Universal client (file_client.py) works across all backends
• Query system for debugging and state inspection
• Direct API and CSL transpilation modes
• No silent failures - comprehensive error reporting

Beyond CAD: end-to-end program automation

• CAE: link meshing/solvers and optimization runs to design parameters.
• CAM: export manufacturable geometry with tolerance/finish metadata.
• BOM and sourcing: extract parameters and materials for structured BOM lines.
• Project/PLM: emit milestones, traceability, and build artifacts per revision.
• Factory automation: generate fixtures/tooling geometry and export to downstream.

Architecture (high level)

• Natural language → Triple Lindy agent → CSL (text) → JSON IR.
• JSON IR → Backend adapter (Fusion/Onshape/SolidWorks/FreeCAD/Blender, etc).
• Adapter executes features, applies selections, exports files and thumbnails.
• Capabilities and diagnostics guide planning, fallbacks, and error reporting.

Quick start (Fusion 360 with real-time control)

# 1) Clone
git clone https://github.com/ricfulop/CSL.git
cd CSL

# 2) Install Triple Lindy Fusion add-in (macOS example)
cp -r triple_lindy_fusion_stable "$HOME/Library/Application Support/Autodesk/Autodesk Fusion 360/API/AddIns/"

# 3) In Fusion 360: Scripts & Add-Ins (Shift+S) → Add-Ins tab → Run "triple_lindy_fusion_stable"

# 4) Test real-time control
python3 triple_lindy_daemon/file_client.py ping    # Test connection
python3 triple_lindy_daemon/file_client.py query   # Query current state
python3 triple_lindy_daemon/file_client.py direct  # Create geometry via API

# 5) Execute CSL v1.3 examples
python3 triple_lindy_daemon/file_client.py file --file CSL_v1_2/examples/loft_continuity_demo.json

Optional backends

• FreeCAD workbench: copy triple_lindy/ to ~/.FreeCAD/Mod/ and select “Triple Lindy”.
• Onshape: configure API keys and use FeatureScript/API pipeline.
• SolidWorks: install add-in or IPC shim; Windows host.
• Blender: bpy/Geometry Nodes realization; render/export for visualization.

Language and schema

• Spec (v1.3 Draft): CSL_v1_3/csl_v1_3_spec.md
• JSON Schema (v1.3 Draft): CSL_v1_3/csl_v1_3_schema.json
• Spec (v1.2 Draft): CSL_v1_2/csl_v1_2_spec.md (extends v1.1)
• JSON Schema (v1.2 Draft): CSL_v1_2/csl_v1_2_schema.json
• Migration: MIGRATION_v1_1_to_v1_2.md
• Conformance checklists: CSL_v1_1/conformance/README.md, CSL_v1_2/conformance/README.md
• Examples: CSL_v1_2/examples/ (v1.2) and CSL_v1_1/examples/ (v1.1)
  • Notable v1.1 example: constraints_patterns_demo.csl (construction/driven constraints, per-instance patterns)

What's new in v1.3

• Surface ops with explicit queries (patch/extend/trim/knit).
• Patterns per‑instance controls (instances/table).
• Constraints QoL: construction flags, reference (driven) dims, equal‑length arrays, curvature continuity diagnostics.
• Loft: continuity (G0/G1/G2), per‑section continuity, orientation options, rails/centerline.
• Export: STL tessellation controls (deviation, angular tolerance, aspect ratio, max edge length).
• Capabilities: refined JSON and versioned publishing (make publish-caps).

What's new in v1.2

• Sketch: spline, ellipse, elliptical_arc, text.
• Loft: continuity G0/G1/G2, orientation controls, rails/centerline.
• Sweep: orientation, twist, scale, optional guide rail.
• Fillet/Chamfer: per‑edge variable groups; transitions/setbacks (best‑effort).
• Draft: explicit faces, neutral_plane, optional pull_dir.
• Wrap/Emboss/Project: depth/draft/direction; project along direction.
• Threads: standardized designations/classes/handedness; modeled vs cosmetic.
• Queries: created_by, owner_feature==, pattern_instances, tangent_connected, largest_by, curvature≈/radius≈/area≈, by_material.
• Assemblies & joints: limits {linear, angular}, optional damping/preload.
• Sheet metal: base/edge flange/bend; unfold/refold (best‑effort).
• Helix curve generator; light PMI notes/frames.
• Validation: make validate-v12 checks IR against the v1.2 schema.

Gap analysis and roadmap

• Backend parity and gaps: BACKEND_GAP_ANALYSIS.md.
• Roadmap priorities: Fusion adapter completeness, robust selection predicates, manufacturing exports.

See Production-Grade Closure Plan in BACKEND_GAP_ANALYSIS.md for remaining items (native wrap/emboss, cross-session stable IDs, units parity tests, full joints, materials/PMI, APS hardening).

Conformance harness (Fusion backend)

Run representative cases (loft continuity G1/G2 with rail, variable fillet/chamfer including angle, emboss/native where available, joints with revolute/slider limits) and emit a JSON report.

python scripts/conformance_harness.py

Outputs:

• out/conformance_report.json: summary with per-case ok, mapping_size, and recent diagnostics.
• out/golden_summary.json: short summary for CI gates.
• Dry-run friendly: if Fusion API is unavailable, the adapter returns a stable mapping and diagnostics for CI.

APS (Autodesk Platform Services)

• Auth test:

export APS_CLIENT_ID=... APS_CLIENT_SECRET=...
make aps-auth

• Orchestrate example (local Fusion run + upload when configured):

make aps-orchestrate-example

• Design Automation (DA) stub:

make aps-da-list-engines

Set APS_SCOPES and optionally APS_BUCKET (defaults to csl-artifacts-<env>).

Export units/parity test

Quickly exercise STL/STEP unit and resolution options.

python scripts/export_units_test.py

Contributing

• Issues and proposals welcome. For large changes, start with a short design note.
• Adapters should implement a minimal interface and publish capabilities for planning.

Getting started for contributors

1. Create a virtualenv and install dev tools (pytest optional).
2. Review the adapter stub at triple_lindy/transpilers/fusion360_backend.py.
3. Run the demo script:

python scripts/run_fusion_example.py

4. Implement real Fusion calls in FusionBackend.realize/export/thumbnail.
5. Extend capabilities and add tests that validate exported artifacts.

License

MIT (see LICENSE).

Project size (LOC)

• Current snapshot (as of 2025-01-13):

  • Total lines: 1838992,058
  • Python lines: 17737,335
  • Markdown/docs lines: 1325283,726
  • JSON lines: 57417,686
  • CSL language files: 301
  • Fusion backend file triple_lindy/transpilers/fusion360_backend.py: 5388,388

• File counts by type:

  • Python files: 52
  • Markdown files: 44
  • JSON files: 16
  • CSL files: 15
  • TypeScript files: 9
  • Manifest files: 9

• Recompute locally (from repo root):

# Totals (excluding .git, out, __pycache__)
find . -type f -not -path "./.git/*" -not -path "./out/*" -not -path "*/__pycache__/*" -print0 | xargs -0 wc -l | tail -n1

# By type
find . -type f -name "*.py" -not -path "*/__pycache__/*" -print0 | xargs -0 wc -l | tail -n1
find . -type f -name "*.md" -print0 | xargs -0 wc -l | tail -n1
find . -type f -name "*.json" -print0 | xargs -0 wc -l | tail -n1
find . -type f -name "*.csl" -print0 | xargs -0 wc -l | tail -n1

# Key file
wc -l triple_lindy/transpilers/fusion360_backend.py

# File count by extension
find . -type f -not -path "./.git/*" -not -path "./out/*" -not -path "*/__pycache__/*" | sed 's/.*\.//' | sort | uniq -c | sort -nr

Learn more

• Documentation index: DOC_INDEX.md
• CSL Glossary: CSL_GLOSSARY.md
• Triple Lindy implementation guide: triple_lindy_complete with CSL.md

Fusion backend status (v1.3)

• Adds: surface ops explicit queries, patterns per‑instance (instances/table), constraints QoL, loft continuity/orientation/rails, STL controls.
• Continues: v1.2 features (loft, sweep, wrap/emboss/project, threads, face ops, patterns, booleans), robust queries, diagnostics, export/thumbnail.
• Assemblies/joints: creation with limits; best‑effort mate connectors and assembly patterns.
• APS: token/bucket upload helpers and an orchestrate(plan) helper for local/hosted agent models.

Query determinism and tolerances

• Predicates supported: created_by/owner_feature==, pattern_instances, tangent_connected(seed, tol_deg), largest_by(axis), curvature≈/radius≈/area≈ with tol, by_material.
• Tolerances: use absolute tolerances via tol (e.g., tol: 0.1 mm) or tol_deg for angular floods; values are unit-checked.
• Determinism policy: ambiguous queries fail with E12xx diagnostics and guidance to strengthen predicates/tags; lineage/attributes are used for stable reconciliation across sessions.

Automation

• Golden check: make golden-check
• Dump capabilities: make caps (writes out/capabilities_fusion.json with version/commit and determinism metadata)
• Publish/Archive capabilities: make publish-caps (writes versioned archive under out/capabilities/archive/<version>/)
• APS Design Automation helpers:
  • List engines: make aps-da-list-engines (requires APS_CLIENT_ID/APS_CLIENT_SECRET)
  • Bootstrap check: make aps-da-bootstrap
  • Create AppBundle: make aps-da-create-appbundle NAME=<name> ENGINE=<engineId> ZIP=<path.zip>
  • Create Activity: make aps-da-create-activity NAME=<name> ENGINE=<engineId> APPBUNDLE=<bundleId>
  • Submit WorkItem: make aps-da-submit-workitem ACTIVITY=<activityId> ARGS=<args.json>
  • Clean: make aps-da-clean
  • DA pipeline: make aps-da-pipeline (ensure + submit; see env below)

Environment

• Optional env for metadata: CSL_VERSION, GIT_COMMIT/CI_COMMIT_SHA, FUSION_BUILD
• APS DA env: APS_ENGINE, APS_BUNDLE_ZIP, optional APS_BUNDLE_NAME, APS_ACTIVITY_NAME, APS_INPUT_URL, APS_OUTPUT_URL