{"slug": "sand3d-python-3d-modeling-built-for-llms-not-blender-scripts", "title": "sand3d: Python 3D Modeling Built for LLMs, Not Blender Scripts", "summary": "Sandflow-ai's sand3d Python 3D modeling library passed 2,150 GitHub stars by November 19, 2026, roughly two weeks after public development began on November 3, 2026. Sand3d represents 3D objects as short, named, deterministic Python programs rather than thousand-line Blender scripts, and Sandflow's README claims 2–6× fewer tokens per part versus Blender Python, CadQuery and FeatureScript, plus a ~3,600-token API surface that fits beside system instructions. The claims come from Sandflow's own documentation rather than independent labs, and the library targets agents that must emit geometry within context limits, including joint semantics that map to simulators such as Isaac and MuJoCo.", "body_md": "**sand3d is a Python library for 3D objects written as short programs LLMs can read and edit**, not as thousand-line Blender scripts. [Sandflow-ai/sand3d](https://github.com/Sandflow-ai/sand3d?ref=explainx) on GitHub had **more than 2,150 stars** when explainx.ai checked on **November 19, 2026**, after roughly two weeks of public development since **November 3, 2026**.\n\nThe project sits at the intersection of **[physical AI datasets](https://explainx.ai/blog/nvidia-cosmos-data-factory-two-prompts-physical-ai-datasets-november-2026)**, **[robotics code generation](https://explainx.ai/blog/forceharness-frozen-frontier-models-force-robot-harness-2026)**, and **[agent tool design](https://explainx.ai/blog/rust-anything-incremental-rewrite-recorded-parity-mcp-2026)**: if your agent must emit geometry, you need an API that survives context limits.\n\n## What problem sand3d solves\n\nFrontier models can describe shapes in natural language, but **executable 3D** usually collapses into **bpy** or **CadQuery** snippets that burn context on coordinates and booleans. Sandflow's pitch is **named, deterministic programs**: every part, edge, face, and joint has a label; the same source rebuilds the same mesh bit-for-bit.\n\nThe [README](https://github.com/Sandflow-ai/sand3d?ref=explainx) lists four measurable claims:\n\n| Claim | What it means for builders | \n|---|---|\n| **2–6× fewer tokens per part** | Median comparison versus Blender Python, CadQuery, FeatureScript (vendor chart) | \n| **~3,600-token API surface** | `docs/api.md` fits beside system instructions | \n| **Named geometry** | No hand-computed vertex lists in prompts | \n| **Deterministic builds** | Same program yields identical meshes for eval replay | \n\nThose numbers come from Sandflow's documentation, not independent labs. They are still useful **design targets** when you pick a 3D tool for agents.\n\n## How a sand3d program is structured\n\nsand3d exposes one flat namespace: `import sand3d as s3d`. Families include **profiles** (rect, slot, text), **solids** (extrude, loft, sweep), **edits** (hole, fillet, shell), **joints**, and **assembly**.\n\n### Example: bracket in millimetres\n\nThe README's starter bracket uses extrude, patterned holes, and fillet:\n\n``` python\nimport sand3d as s3d\n\ndef bracket(w=80, d=40, t=6, hole_d=6.6, pitch=50):\n    plate = s3d.extrude(\"plate\", s3d.rect(w, d, r=4), t)\n    s3d.hole(plate, hole_d, on=s3d.grid(2, 1, pitch, 0))\n    s3d.fillet(plate, 1, edges=\"top\")\n    return plate\n\ns3d.export(bracket(), \"bracket.glb\")\n```\n\nMillimetres are default. One `export` call picks format from the extension.\n\n### Example: washing machine with joints\n\nSandflow's flagship demo models a front-loading washer: cabinet, drum, door, drawer, dial, plus **four joints** (hinge, continuous spin, prismatic slide, continuous dial). The README shows roughly **38 lines** for the full assembly—small enough for a single model completion with room for critique loops.\n\nsand3d washing machine assembly with named joints for LLM procedural 3D modeling\n*Image: [Sandflow](https://github.com/Sandflow-ai/sand3d?ref=explainx)*\n\nJoint semantics map cleanly to simulators:\n\n- **Revolute** door hinge with numeric range\n- **Continuous** drum rotation\n- **Prismatic** drawer travel on an axis\n\nThat is the difference between **pretty meshes** and **things Isaac or MuJoCo can load**, a gap teams hit when moving from **[Cosmos synthetic data](https://explainx.ai/blog/nvidia-cosmos-data-factory-two-prompts-physical-ai-datasets-november-2026)** to **deployable URDF**.\n\n## Token efficiency: what Sandflow measures\n\nToken count chart comparing sand3d to Blender Python for LLM 3D code\n*Image: [Sandflow](https://github.com/Sandflow-ai/sand3d?ref=explainx)*\n\nThe README chart plots **median tokens per part** across comparable objects. Sandflow argues low-rank-style **coordinate math** in other APIs forces models to spend tokens on arithmetic instead of structure.\n\nFor agent builders, the actionable test is simpler: **count tokens for your part catalog** under sand3d versus bpy on ten representative brackets, enclosures, or gripper fingers. If you run **[replica-skill-style](https://explainx.ai/blog/replica-skill-eleven-claude-skills-clone-any-app-recon-entrepreneur-guide-2026)** app cloning, geometry is often the longest hidden cost.\n\n## Export matrix: where files go\n\n`s3d.export(model, path)` routes by extension:\n\n| Format | Carries | Typical downstream | \n|---|---|---|\n| **GLB** | Meshes, materials, joint extras at current pose | Web viewers, game engines | \n| **URDF** | Links, joint limits at rest; OBJ meshes alongside | ROS 2, PyBullet, MuJoCo, Isaac | \n| **USDA / USDC / USDZ** | OpenUSD scene graph | Omniverse, Isaac Sim | \n| **IFC4** | BIM assemblies (current pose; joint metadata omitted) | Revit via IFC link, Tekla | \n| **OBJ / STL** | Meshes only | Slicers, generic DCC tools | \n\nSandflow notes **Revit `.rvt` is not writable** by third-party libraries; IFC is the interchange path. **IFC GlobalIds** can shift when geometry edits—do not use them as stable part IDs across exports.\n\n## Giving sand3d to your coding agent\n\nSandflow ships **`llms.txt`** pointing at **` docs/api.md`** for the full vocabulary. For multi-part objects, **` docs/guidelines.md`** lists eight assembly rules: plan first, then parts, then joints.\n\nThat mirrors how **[MCP servers](https://explainx.ai/blog/what-is-mcp-model-context-protocol-guide)** document tool schemas: keep the contract in-repo so **Claude Code**, **Cursor**, or **Codex** can loop on `export` failures without guessing bpy internals.\n\nRecommended workflow:\n\n1. Pin **Python 3.13 + bpy 5.2** in your devcontainer or VM.\n2. Paste **api.md** into system context once per session.\n3. Ask the agent for **named parts** and**joint declarations** before mesh detail.\n4. Run **`s3d.export`** to GLB for visual QA; URDF when sim semantics matter.\n5. Store programs in git—the source of truth—not binary meshes alone.\n\n## Hugging Face datasets\n\nSandflow published five datasets under **CC BY 4.0**, each rebuilding external corpora as sand3d programs:\n\n- **sand3d-objaverse** — everyday objects from Objaverse\n- **sand3d-cad** — mechanical parts from public CAD\n- **sand3d-articraft** — articulated objects with declared joints\n- **sand3d-simready** — NVIDIA SimReady assets\n- **sand3d-automate** — AutoMate mechanical assemblies\n\nAn [explorer](https://sandflow.ai/sand3d/?ref=explainx) on Sandflow's site browses examples. For fine-tuning, pair programs with **[serverless SFT stacks](https://explainx.ai/blog/baseten-loops-serverless-sft-rl-training-glm-5-3-qwen3-6-per-token-billing-ga-what-docs-confirm-2026)** once you define grading rubrics on exported meshes.\n\n## Install constraints builders hit early\n\nThe README is explicit about **bpy wheels**:\n\n- **Python 3.12 is unsupported** (no bpy wheel).\n- **Intel Macs** from bpy**5.2 onward** should stay on**Python 3.11** .\n- CI runs against **bpy 5.2** on Python 3.13.\n\nIf your agent sandbox is **Node-only** or **Alpine musl**, you will need a **Linux glibc VM** with Blender's Python module—similar friction to **[AnyPS5 native builds](https://explainx.ai/blog/anyps5-ps5-games-native-pc-no-emulator-how-relinking-works-2026)** or GPU harness containers.\n\n## How sand3d fits explainx.ai's agent stack\n\nsand3d does not replace **[LeRobot datasets on Lance](https://explainx.ai/blog/lerobot-v0-7-0-interactive-rollouts-disk-less-streaming-remote-policy-server-cosmos-3-yam-unitree-g1-2026)** or **[NVIDIA Cosmos factories](https://explainx.ai/blog/nvidia-cosmos-data-factory-two-prompts-physical-ai-datasets-november-2026)**. It is a **authoring layer** when the model must **write** geometry instead of retrieving meshes.\n\nTeams building **Snapchat-style AR** or **sim-ready props** might chain:\n\n1. sand3d program from the agent\n2. GLB preview for humans\n3. URDF or USD for sim validation\n4. Promotion into a Lance table for training rows\n\n## Verb surface: what the API actually exposes\n\nSandflow deliberately avoids **category magic** like `chair()` or `table()`. Everything is built from **profiles**, **solids**, **edits**, **frames**, **joints**, and **assembly** verbs on the flat `s3d` namespace.\n\nThe README counts roughly:\n\n| Family | Examples | Count listed | \n|---|---|---|\n| Profiles | rect, slot, text, offset | 16 | \n| Solids and paths | extrude, loft, sweep, torus | 15 | \n| Edits | hole, pocket, fillet, shell | 15 | \n| Articulation | joint, pose, seat | core trio | \n| Assembly | group, export, fingerprint | 8 | \n\n**Materials** and **gears** sit beside those families. The design goal is that an agent **composes** primitives rather than calling opaque generators—similar philosophically to **[MIT Isola's agent loops](https://explainx.ai/blog/llm-agents-self-improve-without-prescribed-loop-mit-isola-2026)** where structure beats hidden helpers.\n\n## Testing and determinism in CI\n\nBecause bpy ties sand3d to Blender's Python module, Sandflow runs CI on **bpy 5.2** with **Python 3.13**. Deterministic rebuilds matter when you:\n\n1. Store programs in git and diff them in code review.\n2. Replay agent trajectories in eval harnesses.\n3. Compare mesh fingerprints with `s3d.fingerprint` style utilities documented in the README.\n\nIf your organization already runs **[Spaceship IDE on Cerebras](https://explainx.ai/blog/spaceship-ide-cerebras-realtime-llm-steering-github-2026)** for tight feedback loops, sand3d pairs well: short programs mean faster critique cycles before you export heavy GLB assets.\n\n## Who Sandflow is\n\nThe repository credits **Thibault Neveu** and **Andrée Loy** in **CITATION.cff**, with the public site at [sandflow.ai](https://sandflow.ai/sand3d/?ref=explainx). The project is young; treat release tags and **CHANGELOG.md** as your stability signal rather than a decade of semver history.\n\n## Limits and honesty\n\n- **Category verbs are intentionally absent** —no`chair()` helper; categories are programs.\n- **Vendor token benchmarks** need your own rerun on production parts.\n- **IFC and USD** dependencies pull in heavyweight tooling on CI runners.\n- The repo is **young** (late 2026); breaking changes may land despite**CHANGELOG.md** .\n\n## FAQ alignment with search intent\n\nPeople who land on this page often ask whether sand3d replaces Blender. It does not: it **generates** bpy work through a constrained vocabulary. Others ask about **robotics**: prioritize **URDF** exports and joint tests in PyBullet before trusting decorative GLB.\n\n## Security and supply-chain notes\n\nsand3d executes **native bpy** code. Treat agent-generated programs like any other **arbitrary Python** in your sandbox:\n\n- Run inside disposable VMs with no outbound mail or secret mounts.\n- Pin **bpy** versions and hash**lockfiles** in CI.\n- Scan exports before loading GLB in production viewers—mesh bombs are rare but cheap to guard against.\n\nThat posture matches how teams already isolate **[Codex sandboxes](https://explainx.ai/blog/openai-inside-codex-sandbox-permission-profiles-november-2026)** and **[Microsoft agent frameworks](https://explainx.ai/blog/microsoft-agent-framework-pypi-1-26-credential-stealer-november-2026)** after typosquatting incidents.\n\n## When not to pick sand3d\n\nSkip sand3d when you only need **single-mesh props** with no joints: a glTF from an artist may ship faster. Skip it when your runtime cannot install **bpy** (many serverless functions). Skip it when you need **NURBS CAD** exactness for machining—FeatureScript or native STEP workflows still win on tolerance stacks.\n\nsand3d shines when **an LLM must author and revise** articulated assets repeatedly—robot cells, appliance demos, warehouse props—without ballooning token use. Start with the README bracket example, export one GLB, and only then scale to multi-joint assemblies your sim team can load. Sandflow's explorer and Hugging Face collection pages are the fastest way to sanity-check joint naming before you automate generation.\n\n## A practical evaluation plan\n\nBefore you adopt sand3d in a team workflow, run a small, honest bake-off. Pick ten parts your agents already generate today: a bracket, an enclosure, a gripper finger, a hinge, a shelf unit. Write each one twice, once in your current bpy or CadQuery style and once in sand3d, using the same model and the same prompt template.\n\nThen measure four things:\n\n1. **Prompt and completion tokens** per part, including retries after export errors.\n2. **First-pass success rate** : how often the program exports without a traceback.\n3. **Edit cost** : how many tokens a follow-up change such as \"make the holes 8 mm\" needs.\n4. **Sim load rate** : whether the URDF opens in your simulator with the joint limits you intended.\n\nNamed parts help most in the edit step. A model can say \"fillet the top edges of the plate\" without recomputing coordinates, so revisions stay small and diffs stay readable in code review. If your numbers do not show a gain on your own catalog, the vendor chart does not matter for you. If they do, you have a concrete case for moving your asset pipeline to programs stored in git.\n\n## Related reading\n\n*Star counts and dataset names follow the Sandflow GitHub README as of November 19, 2026.*", "url": "https://wpnews.pro/news/sand3d-python-3d-modeling-built-for-llms-not-blender-scripts", "canonical_source": "https://explainx.ai/blog/sand3d-sandflow-3d-modeling-python-llms-november-2026?utm_source=rss&utm_medium=feed&utm_campaign=blog", "published_at": "2026-11-19 00:00:00+00:00", "updated_at": "2026-11-19 02:14:01.033388+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "developer-tools", "large-language-models", "robotics"], "entities": ["Sandflow-ai", "sand3d", "GitHub", "Blender", "CadQuery", "FeatureScript", "Isaac", "MuJoCo"], "also_reported_by": [], "takeaway": "sand3d's named, deterministic Python programs claim 2–6× fewer tokens per part than Blender Python, CadQuery and FeatureScript, with a ~3,600-token API surface that fits beside system instructions.", "recommendation": null, "alternates": {"html": "https://wpnews.pro/news/sand3d-python-3d-modeling-built-for-llms-not-blender-scripts", "markdown": "https://wpnews.pro/news/sand3d-python-3d-modeling-built-for-llms-not-blender-scripts.md", "text": "https://wpnews.pro/news/sand3d-python-3d-modeling-built-for-llms-not-blender-scripts.txt", "jsonld": "https://wpnews.pro/news/sand3d-python-3d-modeling-built-for-llms-not-blender-scripts.jsonld"}}