
AI agent operating system for Blender 5.2: bpy knowledge base, AGENT_OK execution contract, 3D-print
AI agent operating system for Blender 5.2: bpy knowledge base, AGENT_OK execution contract, 3D-print production gate, worked keyboard and desktop-robot builds.
·
design:os · 3D Blender · agent operating system
An AI agent that builds parts natively in Blender, and shows its evidence.
Agent skills, a verified bpy knowledge base, an AGENT_OK / AGENT_FAIL execution contract and a production gate for Blender 5.2 LTS. Every worked build below ships with what was checked and what was not.
Live site · Worked builds · Install and use · Tools and skills · design:os ecosystem · Evidence boundaries
![]() | ![]() | ![]() |
| DC-01 desktop companion | CK-001 reference keyboard | Robot arm |
Rendered in Blender, no added captions or banners. Names, development status and evidence limits live next to each build. Media provenance · the full set on the live site.
Two builds ship a public review page with the delivered package beside it.
| Build | Review page | What the repository carries |
|---|---|---|
| DC-01 desktop companion — printable parts, a real carrier board and a 39-second assembly film | Revision R03 · package files | 18 printable parts through the geometry gate, a KiCad carrier with zero ERC and DRC violations, and a 39-second native animatic. Eighteen gated STL files, two plates, the CAD sources and the film travel with the page; run directories and editable scenes stay local. Manufacture not approved. |
| CK-001 reference keyboard, built native from one photo | Revision B · package files | 58 keys and 5 knobs rebuilt from one photo; 781 meshes in a GLB that reopens within 0.05 mm, reviewable in the browser. Contracts, pass scripts, manufacturing studies and tests. |
| Product | Gallery | Included worked build |
|---|---|---|
| Robot arm | Assembly and completed model | Native assembly build and 38-part fit prototype |
| Watch-winder capsule | Overview and surface details | Parameters, native scenes, scripts and print-kit evidence |
ORC images are derivatives of native renders; their source models are not in this public checkout.
| Component | Gallery | Evidence scope |
|---|---|---|
| C01 · Wellhead | Overall and valve details | Render review imagery; physical qualification is separate. |
| C02 · Separator | Vessel, nozzles and instrumentation | Render-only detail set, including the repaired vapor interface. |
| C03 · Exchanger train | Overview, bellows, instruments and hardware | Current R7 HQ imagery; source-derived details retain their stated technical limits. |
| C04 · Three-stage exchanger | Circuit layout and installed instruments | Verification previews; full detail and HQ delivery remain in progress. |
| C05 · Turbine–generator | Eight HQ overall and close-up views | Refreshed cabinet revision; final media integrity and visual review completed. |
The plant layout uses an illustrative 1:15 scale. Catalogue dimensions and installation manuals support specific modeling decisions; they do not establish a selected vendor package, plant capacity or as-built engineering design. A component gallery is not evidence that the complete site has been integrated or qualified.
Instrument families and mounting details show the parts shared across assemblies. Reuse requires a defined local origin, a mounting interface, supported parameters and verification in the receiving assembly.
Requirements: Blender 5.2.x, Python 3.10+, macOS or Linux. The host-side tests use the Python standard library only.
1 · check the install. Point the scripts at your Blender binary, verify the knowledge catalogue, run the host tests.
git clone https://github.com/jangtrinh/design-os-3d-blender.git && cd design-os-3d-blender
export BLENDER_BIN=/Applications/Blender.app/Contents/MacOS/Blender
python3 scripts/blender-knowledge.py check
python3 -m unittest discover -s tests/executionThe rest of the host suites, and the contract tests that import bpy and therefore run inside Blender:
for suite in production-gate knowledge hq-coverage native-pipeline native-review pass-coverage; do
python3 -m unittest discover -s "tests/$suite"
done
python3 -m unittest tests/boilerplates/test_parametric_contract.py
for t in animation geonodes evaluated_mesh; do
bash scripts/headless-run.sh "tests/boilerplates/test_${t}_contract.py"
done2 · run a pass headless. Every pass runs in an isolated Blender process and ends with a sentinel; success is AGENT_OK on the last line, never the exit code.
bash scripts/headless-run.sh my-pass.py3 · or interactively over MCP. Open the blender-mcp addon in Blender's sidebar (N → BlenderMCP → Connect), then run the same pass through the runtime from your agent CLI. <ROOT> is the absolute path of this checkout.
import sys
sys.path.insert(0, "<ROOT>/scripts")
import agent_runtime as rt
rt.run_file("<ROOT>/builds/<slug>/pass-01.py")4 · gate a printable part. Write the spec before detailing; the gate checks the scene against it and writes a report. A digital gate does not establish physical fit, load, pressure, thermal duty or a successful print.
mkdir -p builds/my-part && cp specs/examples/bracket-m3.spec.json builds/my-part/spec.json
python3 scripts/production-gate.py \
--scene builds/my-part/part.blend \
--spec builds/my-part/spec.json \
--report builds/my-part/gate-report.jsonSee specification and gate behavior. Reading packs for a task come from the knowledge workbench:
python3 scripts/blender-knowledge.py list
python3 scripts/blender-knowledge.py route native-hard-surface --topic session-retrospective
python3 scripts/blender-knowledge.py route render-export-delivery --topic public-review-publication| Layer | Entry point | Responsibility |
|---|---|---|
| Operating rules | AGENTS.md, .project-agent.md | Purpose, contract, execution loop and evidence boundaries. |
| Core Blender skill | .agents/skills/blender-agent-core | Native scene execution, numerical checks and delivery recipes. |
| Image reconstruction skill | .agents/skills/blender-image-to-3d | Reference contracts, staged modeling and construction-aware visual review. |
| Knowledge workbench | .agents/skills/blender-knowledge-workbench | Bounded reading packs and reviewed source routing over the verified bpy knowledge base. |
| Execution contract | scripts/agent_runtime.py, scripts/headless-run.sh | Structured AGENT_OK / AGENT_FAIL results and reliable process exit handling. |
| Native pass pipeline | scripts/native-pipeline.py | Declared step dependencies, durable attempt journals and hash-checked resumption; execution status stays separate from acceptance. |
| Target-bound criticism | scripts/native-review.py | Revision-bound target/candidate/proof packets and complete attributed feature findings; actual visual review remains required. |
| Parameter and source contracts | scripts/boilerplates/bp_parametric_contract.py | Explicit mm/m length parameters, local frame/datum/port declarations and source/export byte receipts. |
| Verification and production gate | scripts/agent-verify-lib.py, scripts/production-gate.py, scripts/hq-coverage-gate.py | Scene checks, isolated previews, spec-bound digital geometry and export evidence; coverage receipts before a guarded render. |
| Declared-pass check | scripts/check-pass-coverage.py | Reports pass scripts that no pipeline manifest declares — an authored pass is not an executed pass. |
| MakerWorld | Publishing leg | Bambu project validation and the documented publishing constraints. |
| Build lessons | Two full product builds | The pipeline both builds converged on, what each failure cost, and which rules are code rather than prose. |
| Public review pages | Publication procedure, scripts/page-checks/ | Generated review pages bound to package receipts, with layout, media and live-URL checks. |
The .claude/skills directories mirror the local skill sources. Antigravity setup is in the setup guide. Vendored img2threejs keeps its Apache-2.0 license; production reconstruction uses native Blender geometry. Native agent iteration documents the reviewed Meshy, Dream-loop and Text-to-CAD mechanisms and their local implementations; the upstream hosted-generation, B-rep and device stacks are not installed.
This repository is the Blender member of design:os — one agent operating system per craft, each with its own knowledge base, gates and worked examples.
| Repository | What it is |
|---|---|
| design-os | Design CLI for Claude Code, Codex and Antigravity: describe the UI in plain words, get gated code. |
| design-os-figma-plugin | Figma plugin and CLI so a shell agent can read and edit the open Figma file. |
| design-os-code2flow | Turns a Next.js codebase into a living user-flow canvas with real screenshots. |
| design-os-pedagogy | Evidence-graded pedagogy library and a local teacher review loop for agents. |
| design-os-animejs | Motion knowledge base and gated Anime.js recipes for agent-built interfaces. |
| design-os-voice-ux | Voice and conversational-AI knowledge base with production-ready patterns. |
| design-os-apple | SwiftUI design-system library for iOS, iPadOS and macOS: semantic roles, typed tokens. |
| Build | Demonstrated | Still separate |
|---|---|---|
| Robot arm | Native assembly film, sampled motion checks, exported fit prototypes and a 38-part digital gate report. | Loaded operation, hardware retention, thermal duty and physical assembly qualification. |
| Watch winder | Native stills, sampled motion checks, digital print-kit checks and fit visualization. | Actual printing, physical fit and manufacturing qualification. |
| DC-01 desktop companion | Eighteen printable parts through the geometry gate with STL round-trip and independent 3MF reopening, 26 of 26 harness connections, a KiCad carrier with zero ERC/DRC violations, and a 39-second native animatic. | Printing and process trials, switch actuation, battery charge window, electronics and firmware, and every physical assembly and thermal test. |
| CK-001 keyboard | Reference-bound 58-key digital prototype: modeled receivers, guides and D couplings, sampled key travel, form/final gates on eight part families, GLB round-trip of 781 meshes, native Full HD media. | Switch retention, tolerance extremes, electronics and firmware, process trials and bench/load/thermal qualification. |
| ORC components | Reference-driven component development, reusable parts and revision-specific image/geometry checks. | Selected equipment, complete plant integration, process duty and fabrication qualification. |
The worked-build READMEs carry the detailed receipts and exclusions. The knowledge catalogue also includes small execution examples; the full private ORC development history and the source reference images are not shipped.
Both full product builds converged on one loop — build → numeric assert ending in the sentinel → immutable run directory → an independent re-open of the saved artifact → a named review report → freeze by SHA — and the incidents that cost the most are recorded with their evidence files:
Read the two-build distillation, the CK-001 retrospective, the ORC component lessons with their public journal, and start a new project from the product workflow template. Specialised reading: geometry diagnosis, native media delivery, manufacturing evidence, publishing a review page.
The C03/E02 source package adds full-stroke cap clearance, mechanical retention, acrylic collars, corrected LED polarity and an explicitly incomplete physical pilot ledger. The public Revision-B gallery remains historical media for B, not an approval of C03.
MIT. Vendored dependencies retain their own licenses.
This repo does not document an install step we could read. Check the README on GitHub.
Paste this prompt into your agent. It reads this page and does the setup for you.
Read https://aiagentslisting.com/mcp/design-os-3d-blender to learn what the "Design Os 3d Blender" MCP server does and how to install it. Add it to my coding agent's MCP configuration as documented on that page, then confirm the server connects and list the tools it exposes.Agents can also browse this directory over MCP at https://aiagentslisting.com/api/mcp
This server doesn't publish a machine-readable tool manifest, and it doesn't expose a public endpoint we could ask. Connect the server locally to see what it exposes.
Paste this line near the top of your repository's README. The badge always renders the current score.
[](https://aiagentslisting.com/mcp/design-os-3d-blender)Nothing comparable is listed yet.
v0.1.0v0.1.0 — Blender agent OS + robot-arm demoRelease notesConnect to the AI Agents Listing MCP server at https://aiagentslisting.com/api/mcp using the HTTP transport and complete the sign-in when it asks. Then call the submit_review tool for the listing "design-os-3d-blender" with a rating from 1 to 5 and a short note on what worked and what did not.One email a week. New agents, MCP servers and skills, and what is actually getting traction.