Electrical & control panels
Control panel design, automated end-to-end: single-lines, main and control circuits, panel layouts, terminal tables, and the bill of materials, generated from a spec and sized against the loads.
Agents that
design machines.
Outerport designs the electrical and piping systems inside heavy equipment: from spec, to CAD drawing, to verification that drives the CAE and simulation tools you already run.
Trusted by the companies that build the world’s machines
Top 5 semiconductorequipment firmTop 5 EPC firmFortune 1000chemical manufacturerFortune 1000defense contractorWhat Outerport is
Coding agents turned a prompt into working software. Outerport does the same for the systems inside machines.
Every industrial machine runs on designed systems: relay control circuits, power distribution, piping. Outerport’s agents design them as structured models, and everything downstream (the CAD drawing your manufacturer needs, the checks, the simulation) compiles from the model.
What it designs
The two disciplines every machine is built on.
Describe the system. Outerport produces the deliverable, and the structured data behind it.
Piping & instrumentation
P&IDs, line lists, and process systems assembled symbol-by-symbol, then validated for continuity, tagging, and instrumentation coverage.
How it works
From specification to shipped drawing.
01
Design from the spec
Hand the agent a requirements document, a datasheet, or plain language. It designs the system the way a discipline engineer would, down to component sizes and wire labels.
02
Compile to CAD
The design model compiles straight to CAD: drawings you can send to your manufacturer, in the drafting conventions your industry expects.
03
Prove it works
Simulations and design-rule checks run against the model itself, not the picture of it. Continuity, sizing, and interlocks are verified on every revision.
Smart CAD promised this a decade ago. It stalled on the ramp.
Rule-based configurators only produce a drawing after your senior engineers spend months encoding a macro library, and only pay off on designs you repeat. Custom-order work never clears that bar, so the tools sit unused and the drawings stay manual.
Outerport’s agents build the templates and run the automation themselves. The learning curve collapses to describing what you need.
Your existing drawings
Your archive becomes a design library.
Panel schematics, single-lines, and P&IDs, decades of them, parse into the same structured form the agents design in.
What comes out is a design library of templates: the interlock circuits you always use, the breaker frames you trust, ready to use in Outerport’s design tool or exported to the smart CAD you already run. A sheet drawn in 1994 becomes the panel you ship next year, with the ratings resized for the new loads.
Simulation & search
Simulate a design. Then search for a better one.
Validation here goes deeper than rule checks: the design runs. Control circuits execute in Outerport’s own simulator, and you watch the interlock sequence fire the way the panel will. The CAE and simulation tools you already run can be driven by the agent too: it sets up the case, runs it, and reads the results back into the design model.
Then the design space opens: generate a family of candidates and let an optimization agent search it, with passing simulation as the constraint. Thousands of candidates narrow to the few worth an engineer’s time, every survivor already validated.
It runs in reverse, too. Give it test and inspection data, and it searches back for the design and process conditions that produced them. That is the loop R&D runs to get to a working prototype in fewer rounds, and the one the line runs to raise yield.
“We were able to achieve a dramatic leap in parsing accuracy compared to conventional approaches.”
Shohei Hido
Executive Engineer, Daikin Industries
The platform
More than a design tool.
Everything above runs on a data platform: one model of your equipment, built from parsed drawings and new designs alike, with a knowledge graph joining every document.
It is an agent platform, too. The drawing tools the agents use (trace a line, read a legend, query the graph) are open to agents your own team builds. The sections below are that platform.
The parsing pipeline
How a sheet becomes data.
Every page is classified first: a P&ID reads differently from a control circuit, so each type gets its own detectors for symbols, lines, and text. Detected symbols are checked against the sheet’s own legend, and P&ID classes map to CFIHOS.
Connectivity is assembled last, line by line and tag by tag, into the graph the rest of the platform runs on. Every value keeps a citation to the exact region of the sheet, and a review workspace shows it to the engineer who signs off.
Across disciplines
One graph joins the P&ID, the single-line, and the datasheet.
Pump P-2104 is a symbol on the P&ID, a load on the single-line, and a row in its datasheet. In Outerport’s knowledge graph those are one node, joined by tag, so process lines, power feeds, breaker sizes, and rated duty connect across the whole document set. The graph is typed by an ontology: equipment classes and the relations between them, CFIHOS out of the box and extended to your own naming. You query the plant, not the files.
Agent session
Which loads share a feeder with P-2104?
trace_line · E-101 · upstream of P-2104 → MCCB-2
trace_line · E-101 · downstream of MCCB-2
read_legend · E-101 · breaker frame sizes
MCCB-2 (30AF/30AT) feeds the duty and standby pumps P-2104 and P-2105. No other loads share the feeder; each pump branch has its own thermal relay, THR1 and THR2, set 5-8A.
E-101 · SINGLE-LINEE-103 · CONTROL CIRCUIT
Agents
Ask a question that crosses three documents.
What trips if this pump overloads? The agents that design systems also answer questions about them, working on the graph and on the sheets themselves: tracing a line to its source, checking a symbol against the legend, citing every sheet they used.
The same tools are open to your team: build an agent that reviews vendor drawings against your design standards, or one that counts and prices what is on a sheet for an estimate.
Build on it
Run code next to your drawings.
Agents and your engineers execute Python in a sandbox with the parsed data in scope: query the graph, check a loads table with pandas, plot the result. Every session is captured as a notebook your team can rerun and review.
Personas define what an agent is for, and skills package the procedures it follows: both are yours to edit. The whole platform deploys in your VPC, so drawings never leave your network.
pump_station_audit.opnb
SANDBOX · PYTHON
Which pumps are close to tripping their breakers?
pumps = graph.query("pump", join=["feeder", "datasheet"])
margin = pumps["breaker_a"] / pumps["fla_a"]P-2105 · margin 1.1 · below minimum
PERSONA · PANEL REVIEWERSKILL · BREAKER MARGINRUNS IN YOUR VPC
In and out
One model between your documents and your tools.
YOUR DOCUMENTS
PDF · PNG · TIFFscanned drawings
DWG · DXFCAD sheets
XLSX · CSVdatasheets, BOMs, specifications
THE DESIGN MODEL
components · connections · ratings · tags
connected as one knowledge graph
YOUR TOOLS
DWGyour CAD
DEXPIAVEVA
HYSYSflowsheets
CSVCMMS / ERP
JSONAPI
Industries
From semiconductor tools to mining machinery.
Semiconductor equipment
Control systems for the machines that make chips: gas delivery, RF and vacuum subsystems, tool control panels.
Energy & process
Process units, utilities, and safety instrumentation, from FEED through as-built.
EPC
Estimation, detail design, and as-built reconciliation across project document sets.
Heavy machinery & mining
Control panels and hydraulic systems for equipment engineered in variants.
Deployment
Fitted to your drawing standards.
No two teams draw alike. Deployments start from your own document set: our engineers tune the parsers, symbol libraries, and design checks to your title blocks, tag conventions, and drafting standards (JIS, IEC, or ANSI), and keep tuning them as those standards evolve.
Custom schemas, custom agents, and integrations into the systems you already run are part of the engagement. We operate the service alongside your engineers, on the problems your floor actually has, and we stay until it succeeds in production.
Our vision
The people who design power systems, process plants, and piping networks carry an enormous amount of judgment that has never been written down, only drawn.
We are building the system that learns that judgment: an agent that knows a discipline well enough to draft in it, and to keep the drawing and its data in agreement. Engineers stay in charge of the design. The agent does the drafting, checking, and redrawing that eats their weeks.
Who we are
Built by people who design the physical world.
Our founding team wrote NVIDIA’s open-source 3D machine learning library and the neural 3D representations that led to NeRF and Gaussian splatting. We have put manufacturing execution software on factory floors, autonomy stacks on trucks, and models into defense and disaster-risk systems. Previously at NVIDIA, Meta, Oracle, Embark Trucks, and Tulip Interfaces.
Towaki Takikawa
Co-Founder & CEO
I’ve been building machines since I was a child: first robots, then autonomous vehicles, then AI that understands the 3D world. At NVIDIA I was a research scientist leading work on the neural 3D representations that NeRF and Gaussian splatting came out of, research cited over 5,000 times at CVPR, ICCV, and SIGGRAPH, and I worked on Kaolin, NVIDIA’s open-source 3D machine learning library. Before that I built an autonomous vehicle from scratch, and wrote software that ran on factory floors at Tulip Interfaces. I left to unlock the creative freedom blocked by the drafting, checking, and redrawing that eats an engineer’s weeks.
Backed by
Put an agent on
your next design.
Tell us what your team designs and we’ll show it running on one of your own drawings.
Or write to us at info@outerport.com