Skip to content

Co-Engineer

← Home · Co-Engineer

The Protos Co-Engineer is a multi-agent AI system available across all features. It structures data, configures simulations, and surfaces connections — always with traceable sources. Behind the scenes it calls specialist agents to do focused work, but it stays the one you're talking to.

Important: Co-Engineer is built to surface information it can trace rather than speculate. When a response draws on the Knowledge Library, it carries source traces you can click to see the exact passage behind it.

On This Page

What Co-Engineer Can Do

Capability Example
Schema authoring Create or update a schema based on your description — e.g. "Create a schema for tablet formulation experiments"
Data document creation Extract structured data from an uploaded file and create a data document following your schema
Canvas building Builds and edits a simulation canvas — adds blocks, writes the code, and wires everything together
Running models Run a registered model, or a component on a canvas, and report the result
Model registration Register a model from a script, a repo, or an API endpoint, inferring its input and output schemas
Knowledge surfacing Find relevant entries from the Knowledge Library as you work
Web research Search the web and import a page straight into the Knowledge Library
SharePoint import Pull files and folders from SharePoint into the Knowledge Library
CAD modelling Author and edit parametric 3D geometry in your own Onshape documents, and land the fabrication files back in the project
Requirements parsing Parse an uploaded spec document into structured targets and constraints
Connecting external tools Register an MCP server you name, then take you to the page to finish setting it up
Taking you to an asset Navigate you straight to the schema, canvas, or document it just worked on
Version history "What changed between v2 and v3 of this schema?" — browse and diff version history, check whether a canvas's dependencies are outdated, and restore a schema, canvas, or data document to a past version

How to Use It

The Co-Engineer panel opens from the Co-Engineer icon in the top-right corner of the header bar, on any project screen.

Type in plain language:

  • "Create a schema for tablet formulation experiments with fields for particle size, binder type, and dissolution rate."
  • "Extract the key parameters from this test report and create a data document."
  • "Are there any prior experiments on this material in the Knowledge Library?"
  • "Build a canvas that takes my formulation data and calculates the adjusted capacity."

When a response draws on the Knowledge Library it includes source traces — click one to see where it came from.

You can attach files to any message — PDFs, documents, images, and more. Up to 10 files per message, 32 MB each.

Co-Engineer is a Pro feature. On a free plan the chat panel shows an Upgrade to Pro banner and messages can't be sent.

Build and Brainstorm modes

The mode pill sits in the lower-left of the chat composer, next to the MCP button. It sets the mode for the next message, so you can switch at any point in a conversation.

Some agents don't use modes — guided onboarding and Help — and while one of them has the conversation the slot holds a plain badge. Hover it to see which agent you're talking to. Modes are back as soon as the Co-Engineer has the conversation again.

Mode What it can do
Build (default) Everything. Creates and edits schemas, data documents, canvases, requirements, and runs models.
Brainstorm Reads and researches. It searches the Knowledge Library and the web, and can import sources into the library, but it cannot change a schema, data document, canvas, or requirement.

Use Brainstorm when you want to think something through without the Co-Engineer building as it goes: scoping an approach, reviewing what the library already holds, or asking what a set of results implies.

Importing into the Knowledge Library is deliberately allowed in Brainstorm — filing a source enriches what you can both draw on, without changing anything you have designed.

Brainstorm's limit covers your Protos project, not the tools you've connected to it. Tools reached over MCP are outside it, and some of them can write: a connected issue tracker can open an issue, and Onshape can be driven to change a document. If you want a guarantee that nothing outside Protos is touched either, turn those servers off for the conversation from the MCP button.

Multi-agent system

The Co-Engineer is built on a multi-agent architecture, but you only ever talk to one agent. Behind the scenes it calls specialists to do focused work and folds their results into its own reply.

Specialists

  • Knowledge — searches the Knowledge Library and the web, and imports sources into the library
  • Data — creates and edits schemas, data documents, and requirements, and sets up the Data Studio
  • Simulation — builds and edits canvases, and runs models
  • CAD — models parametric 3D geometry in your Onshape documents (see below)

You never talk to Knowledge, Data, or Simulation directly — they run in the background and the Co-Engineer folds their results into its reply. CAD and Help work differently: each takes over the conversation, so while one of them has it you are talking to that agent, until it hands back. Help answers questions about Protos itself (type /help).

CAD (Onshape)

The CAD specialist authors and edits parametric 3D geometry in your own Onshape documents. It writes native FeatureScript, so the geometry stays parametric and Onshape remains the single source of truth for it. It can measure live geometry, check the result visually as it builds, and save fabrication files back into the project.

CAD work is iterative — a build can run across dozens of steps — so CAD takes over the conversation and drives the work itself, then hands back to the Co-Engineer.

It needs Onshape connected as an MCP server. If it isn't, the agent explains what it needs and points you at MCP Connections.

Sub-agent progress card

While a specialist is working, a card appears in the chat with its name and a spinner. Expand Task to see the brief the Co-Engineer sent it, and watch the steps as it works. The card collapses to a green check when the specialist finishes, or a red cross if it failed — click it open again at any time to see what it did.

Knowing which agent you're with

Which agent has the conversation shows in its colour: the pill's tint, the accent ring on the composer, and the border on each message all follow the active agent. Where the slot holds a plain badge — guided onboarding, or Help — hovering it names the agent outright.

Slash commands

Typing / in the chat composer opens a command picker. Available commands:

  • /help — get context-aware help from the docs
  • /feedback (or /idea) — submit feedback about Co-Engineer

Versioning skill

Ask about version history, whether a canvas's dependencies are current, or to restore something to an earlier state, and the Co-Engineer loads its versioning tools automatically — there's nothing to turn on. See Versioning for what it can and can't do (model and knowledge-document versions can only be browsed, not restored, this way).

MCP servers

An MCP button in the lower-left of the chat composer lets you toggle individual MCP servers on or off for the current conversation. You can also check Default in new conversations per server so your preferred setup carries over automatically. The popover also has Enable all and Disable all. If no servers are configured yet, it links you to MCP Connections to add one.

You can also just ask the Co-Engineer to connect a server — give it the URL and it will register it and take you to the MCP servers page to run tool discovery.

Sharing a chat session

Co-Engineer sessions can be shared with community members. Open a chat session, click the Share icon, and assign Editor or Viewer access to individuals, teams, or the whole community — chats can't be made public. Sessions shared with you appear under Shared with me in the project's Co-Engineer sessions panel, and open as a read-only transcript. See Collaboration & Sharing for details.

Co-Engineer and Your Data

Co-Engineer can take direct actions in your project — creating schemas, data documents, canvases, and requirements — based on what you ask it. These actions happen through the chat; the Co-Engineer will tell you what it created or changed.

You can review all Co-Engineer sessions for a project by opening the Co-Engineer panel and browsing past sessions.

How Co-Engineer Gets Smarter

Co-Engineer's quality depends on the richness of your workspace:

  • Models: The more models registered and documented, the better Co-Engineer's simulation configuration suggestions.
  • Knowledge Library: The richer the library, the more relevant connections Co-Engineer can surface as you design.
  • Schema quality: Well-defined schemas with units and descriptions give Co-Engineer more signal when extracting data from files and creating documents.

Investing in these foundations makes Co-Engineer progressively more useful over time.

See Also

  • Knowledge Library — Co-Engineer draws on this to surface relevant prior work
  • Models — Co-Engineer uses this for simulation configuration recommendations
  • Schemas — Co-Engineer can create and update schemas based on your description
  • Simulation Studio — Co-Engineer can recommend model and input configurations
  • MCP Connections — connect external tools the Co-Engineer can use in chat
  • Versioning — what the Co-Engineer's version tools can browse, diff, and restore

← Back to Home