Codex
Public Made by Adomby adom
Codex in Adom Hydrogen: ecosystem setup, dock dashboard and live engineering demos.
Codex Astra: independent PCB placement, routing and native 3D comparison
Meeting video · 4:12 · narrated · real Autodesk Fusion. Start here for the all-engineering Codex Astra discussion. The new native 3D comparison begins at 2:46.
Open/download MP4 · All Fusion progression videos · Fusion Bridge perspective · EDA Skillpack
What Astra did
Starting with Adonis's BQ25792 charger, Astra preserved a reference copy, removed the routes, and moved all 109 elements off the board. It calculated a new component placement and four-layer routing inside the original 50.4 × 26.4 mm outline. The plan accounted for the Molecule grid, four corner machine-pin contacts, connector daughterboard mating patterns, pad access, and critical power components. No original trace geometry or external autorouter was reused.
Astra's reasoning and calculation scripts produced the candidate; Adom Hydrogen and Fusion Bridge made those decisions visible in the real editor. The recording shows native placement and routing replay, copper pours, a connectivity correction, and verification. Calculations were performed before replay; the video labels accelerated action and removes waits. This is not a claim that an arbitrary PCB can be optimally designed in four minutes.
At the end, Fusion generates populated native 3D boards for both designs. The comparison shows the USB-C, barrel-jack and XT60 daughterboards, then temporarily hides them to expose the base board. Matching-scale top views and editorial callouts explain actual component moves. The daughterboards are restored; no components are deleted during 3D inspection.
What changed in placement
Adonis's layout has greater symmetry. Astra's alternative moves IC1 5 mm left and 1 mm down, with its nearby power capacitors; brings the inductor from the upper-right edge toward the middle-right; moves the battery transistor inward; changes the input transistor column into a staggered group; and moves programming contacts to the upper edge. The four corner contact locations and board outline remain fixed. Molecule mating patterns were retained, though some interfaces shifted on the grid.
This is a concrete demonstration of independent placement, not a declaration that the result is better in every respect. The candidate uses 157 routing vias versus the reference's 88. Symmetry, routing complexity, assembly access and thermal paths are distinct objectives, and should be evaluated together.
Measured result and limits
| Check | Recorded Astra result |
|---|---|
| Elements / nets | 109 / 50 |
| Copper layers | 1, 2, 15, 16 |
| Native geometry | 879 wire segments, 157 routing vias, 69 pour outlines |
| Wire direction | Horizontal, vertical or 45° within coordinate quantization tolerance |
| Molecule grid | 53 machine-pin pad centers checked; zero off-grid failures |
| Native DRC after refill and final repair | 0 airwires, 0 errors, 0 warnings |
| Rule profile | 3rd party fab 4L with requested 0.1 mm isolation floor; stricter 0.15 mm pad-to-pad and 0.2 mm edge/drill spacing retained |
| Final movie | 1920 × 1080, H.264/AAC, 252.02 seconds; complete decode passed |
The reference under the same profile reported 14 wire-stub errors and seven warnings; it is preserved as a separate baseline. Medium-contact drills were explicitly updated from 0.84 mm to the current component definition's 0.78 mm. Press-fit validation is still needed.
The target was “up to 4S lithium-ion”; input range, charge/load current and cooling conditions remain unspecified. DRC does not establish current capacity, thermal performance, signal integrity, insulation suitability or mechanical fit. The 3D comparison is a visual review, not a completed interference or tolerance analysis. These are standalone board snapshots; synchronized schematic/PCB and ERC signoff, CAM acceptance, and fabrication qualification remain open.
This four-layer run used an independent planner and native Electron.run replay. It did not treat the bridge's conservative two-layer routing preflight as a general four-layer autorouter.
Try the workflow on your own board
Codex is available through Hydrogen. Select the intended model in your AI session; typing “Astra” does not itself select a model. Use a signed-in Fusion installation and current compatible Adom Bridge/Fusion Bridge releases, preserving any deliberately pinned development installations. Have the AI discover your desktop and inspect live capability schemas first.
Copy this prompt into Codex in Hydrogen:
Read
eda-placement-routing-review,eda-component-placement,eda-board-3d-comparisonandeda-design-summary-videofrom the EDA Skillpack at https://wiki.adom.inc/adom/eda-engineering, together witheda-visible-routingand the installed Fusion Bridge skills. Use the latest BQ placement/routing/3D example at https://wiki.adom.inc/adom/codex/files/docs/BQ-PLACEMENT-ROUTING-3D.md as the workflow reference. Find Adonis's BQ25792 charger on the wiki, preserve the original and make my own working copy. Ask which manufacturer I want, load and read back its rules, and retain stricter electrical constraints. Confirm the board outline, Molecule corner/grid and daughterboard interfaces, and ask about current and cooling requirements before making qualification claims. Prove the working copy is stripped, move its components off-board, calculate a fresh placement and routes, then show the accepted edits visibly in Fusion. Refill copper and run native DRC on the actual result. Compare my result with Adonis's in native Fusion 3D from matching views, explain component moves and tradeoffs, and make a narrated summary video. Preserve failed candidates and report unresolved checks honestly; do not reuse a saved routing plan and call it fresh calculation.
The EDA Skillpack source is readable today. Its installable package is not yet released; have the AI read the linked skill files directly rather than assume pkg install succeeds. These general skills teach the method; they do not bundle a turnkey guaranteed optimizer or guarantee identical placement. Session-specific live-control scripts are retained as task evidence, not shipped as universally safe commands.
Inspect the comparison artifacts
Final Astra board · Reference board · Detailed result and limitations
Video SHA-256: 63058b8acf289e31d9db5012af98496afcd5d08b0530ba7a875582ec8442e8f0.
# Codex Astra: independent PCB placement, routing and native 3D comparison
**Meeting video · 4:12 · narrated · real Autodesk Fusion.** Start here for the all-engineering Codex Astra discussion. The new native 3D comparison begins at **2:46**.
<video width="100%" controls playsinline preload="metadata" poster="https://wiki.adom.inc/api/pages/adom/codex/files/docs/bq-placement-3d-poster.png" src="https://wiki.adom.inc/api/pages/adom/codex/files/videos/fusion/13-bq-astra-placement-routing-3d.mp4"></video>
[Open/download MP4](https://wiki.adom.inc/api/pages/adom/codex/files/videos/fusion/13-bq-astra-placement-routing-3d.mp4) · [All Fusion progression videos](https://wiki.adom.inc/adom/codex/files/docs/ASTRA-FUSION.md) · [Fusion Bridge perspective](https://wiki.adom.inc/adom/fusion-bridge/files/docs/astra-placement-routing-3d.md) · [EDA Skillpack](https://wiki.adom.inc/adom/eda-engineering)
## What Astra did
Starting with [Adonis's BQ25792 charger](https://wiki.adom.inc/adom/bq25792-charger), Astra preserved a reference copy, removed the routes, and moved **all 109 elements off the board**. It calculated a new component placement and four-layer routing inside the original **50.4 × 26.4 mm** outline. The plan accounted for the Molecule grid, four corner machine-pin contacts, connector daughterboard mating patterns, pad access, and critical power components. No original trace geometry or external autorouter was reused.
Astra's reasoning and calculation scripts produced the candidate; **Adom Hydrogen and Fusion Bridge made those decisions visible in the real editor**. The recording shows native placement and routing replay, copper pours, a connectivity correction, and verification. Calculations were performed before replay; the video labels accelerated action and removes waits. This is not a claim that an arbitrary PCB can be optimally designed in four minutes.
At the end, Fusion generates populated native 3D boards for both designs. The comparison shows the USB-C, barrel-jack and XT60 daughterboards, then temporarily hides them to expose the base board. Matching-scale top views and editorial callouts explain actual component moves. The daughterboards are restored; no components are deleted during 3D inspection.
## What changed in placement
Adonis's layout has greater symmetry. Astra's alternative moves IC1 **5 mm left and 1 mm down**, with its nearby power capacitors; brings the inductor from the upper-right edge toward the middle-right; moves the battery transistor inward; changes the input transistor column into a staggered group; and moves programming contacts to the upper edge. The four corner contact locations and board outline remain fixed. Molecule mating patterns were retained, though some interfaces shifted on the grid.
This is a concrete demonstration of independent placement, not a declaration that the result is better in every respect. The candidate uses **157 routing vias versus the reference's 88**. Symmetry, routing complexity, assembly access and thermal paths are distinct objectives, and should be evaluated together.
## Measured result and limits
| Check | Recorded Astra result |
|---|---|
| Elements / nets | 109 / 50 |
| Copper layers | 1, 2, 15, 16 |
| Native geometry | 879 wire segments, 157 routing vias, 69 pour outlines |
| Wire direction | Horizontal, vertical or 45° within coordinate quantization tolerance |
| Molecule grid | 53 machine-pin pad centers checked; zero off-grid failures |
| Native DRC after refill and final repair | **0 airwires, 0 errors, 0 warnings** |
| Rule profile | 3rd party fab 4L with requested 0.1 mm isolation floor; stricter 0.15 mm pad-to-pad and 0.2 mm edge/drill spacing retained |
| Final movie | 1920 × 1080, H.264/AAC, 252.02 seconds; complete decode passed |
The reference under the same profile reported 14 wire-stub errors and seven warnings; it is preserved as a separate baseline. Medium-contact drills were explicitly updated from 0.84 mm to the current component definition's 0.78 mm. Press-fit validation is still needed.
The target was “up to 4S lithium-ion”; input range, charge/load current and cooling conditions remain unspecified. DRC does **not** establish current capacity, thermal performance, signal integrity, insulation suitability or mechanical fit. The 3D comparison is a visual review, not a completed interference or tolerance analysis. These are standalone board snapshots; synchronized schematic/PCB and ERC signoff, CAM acceptance, and fabrication qualification remain open.
This four-layer run used an independent planner and native Electron.run replay. It did not treat the bridge's conservative two-layer routing preflight as a general four-layer autorouter.
## Try the workflow on your own board
Codex is available through Hydrogen. Select the intended model in your AI session; typing “Astra” does not itself select a model. Use a signed-in Fusion installation and current compatible Adom Bridge/Fusion Bridge releases, preserving any deliberately pinned development installations. Have the AI discover your desktop and inspect live capability schemas first.
Copy this prompt into Codex in Hydrogen:
> Read `eda-placement-routing-review`, `eda-component-placement`, `eda-board-3d-comparison` and `eda-design-summary-video` from the EDA Skillpack at https://wiki.adom.inc/adom/eda-engineering, together with `eda-visible-routing` and the installed Fusion Bridge skills. Use the latest BQ placement/routing/3D example at https://wiki.adom.inc/adom/codex/files/docs/BQ-PLACEMENT-ROUTING-3D.md as the workflow reference. Find Adonis's BQ25792 charger on the wiki, preserve the original and make my own working copy. Ask which manufacturer I want, load and read back its rules, and retain stricter electrical constraints. Confirm the board outline, Molecule corner/grid and daughterboard interfaces, and ask about current and cooling requirements before making qualification claims. Prove the working copy is stripped, move its components off-board, calculate a fresh placement and routes, then show the accepted edits visibly in Fusion. Refill copper and run native DRC on the actual result. Compare my result with Adonis's in native Fusion 3D from matching views, explain component moves and tradeoffs, and make a narrated summary video. Preserve failed candidates and report unresolved checks honestly; do not reuse a saved routing plan and call it fresh calculation.
The [EDA Skillpack source](https://wiki.adom.inc/adom/eda-engineering) is readable today. Its installable package is not yet released; have the AI read the linked skill files directly rather than assume `pkg install` succeeds. These general skills teach the method; they do not bundle a turnkey guaranteed optimizer or guarantee identical placement. Session-specific live-control scripts are retained as task evidence, not shipped as universally safe commands.
## Inspect the comparison artifacts
[Final Astra board](https://wiki.adom.inc/api/pages/adom/codex/files/docs/bq-placement/astra-comparison-native.brd) · [Reference board](https://wiki.adom.inc/api/pages/adom/codex/files/docs/bq-placement/adonis-comparison-native.brd) · [Detailed result and limitations](https://wiki.adom.inc/adom/codex/files/docs/bq-placement/RESULTS.md)
Video SHA-256: `63058b8acf289e31d9db5012af98496afcd5d08b0530ba7a875582ec8442e8f0`.