How Construct Builds Internal Tools in Your Workspace
Ankush · Author
Writes about Construct, AI agents, and developer tools.
Published · Updated · @ankushKun_
On this page12 sections · 7 min read
- Why build it in Construct?
- Start with the repeated process
- The app lives beside the work
- What a workspace app can do
- Permissions are part of the design
- Validation, then verification
- Draft changes do not replace the last good build
- Internal tool or workflow?
- Before you start: where workspace apps fit
- Already tracking this in a spreadsheet? Move it in 15 minutes
- Which should you choose?
- Related resources
Construct builds a private app for your team's repeated process, such as a request tracker, an intake form, a review queue, or a dashboard, inside the same workspace as the agent and files doing the work, from $9 a month. The usual alternatives are a spreadsheet that people copy values between, or a separate hosting project someone has to set up, secure, and maintain.
For founders and small teams, Construct is the faster way to get an internal tool. You describe the job in plain language; the agent writes the app, checks that it builds, and opens it in your desktop. The source stays in your files, permissions are declared up front, and a broken edit never takes the working version offline.
Why build it in Construct?
- Describe the process, get the app. Construct writes the React and TypeScript, validates it, and opens it in the web desktop. No separate hosting project.
- It lives beside the work. Source sits in Files under
Apps/<app-id>/, and the app's data inAppData/<app-id>/, next to the reports and files it tracks. - Least privilege by default. Each app declares the capabilities, connected apps, and network origins it needs, and other calls are denied at runtime.
- The last good build keeps running. A failed edit never replaces the working app while the agent repairs it.
- The agent maintains it. Construct reads the app's own diagnostics, fixes the source, and validates again when something breaks.
- Private to your workspace. Teammates with access can open it from the desktop, Launchpad, or Spotlight.
Start with the repeated process
The best request describes the work rather than prescribing a software architecture. For example:
- "Build a research request tracker with an owner, priority, status, and link to the finished report."
- "Turn this recurring launch checklist into an app the team can update."
- "Create a dashboard for the JSON reports stored in this workspace."
- "Build an intake form for customer-review requests and keep the queue in one place."
Construct can start from a list, form, dashboard, or blank interface. The agent creates the app package, writes the React and TypeScript source, declares the capabilities it needs, and validates the result. A successful build opens directly in the Construct web desktop instead of sending you to a separate hosting project.
The request still matters. "Build an operations dashboard" leaves important choices unresolved. "Show open research requests, let an operator assign an owner, and mark the linked report as reviewed" gives the agent a bounded workflow and a clear way to tell whether the app is useful.
Silent product demo: Construct writing, validating, and publishing a working internal app, then opening it in the product interface.
The app lives beside the work
A workspace app is not a hidden artifact. Its source lives under an Apps/<app-id>/ folder in Files, where it can be inspected and edited. After a successful build, the app appears as a private workspace install that can be reopened from the desktop, Launchpad, or Spotlight.
The app's durable state is kept separately under AppData/<app-id>/. That separation is important: changing the interface does not overwrite the records the app already manages. Rebuilding updates the code while existing app data remains in the workspace and counts against the normal workspace storage quota.
This model also makes the surrounding agent useful after launch. The same workspace contains the source, relevant files, conversations, and diagnostics. If the interface needs another field or a runtime error appears, the next task starts with the actual app rather than a screenshot and a vague bug report.
What a workspace app can do
The runtime is purpose-built for internal tools, which keeps each app close to its workspace job.
| Capability | Workspace-app support |
|---|---|
| React interface | Yes |
| TypeScript and TSX | Yes |
| Read or write workspace files | With declared permission |
| App-scoped JSON state | With declared Files permission |
| Call an allowed connected app | With an explicit app allowlist |
| Make direct network requests | To declared HTTPS origins |
| Packages | The Construct SDK; arbitrary npm packages are not installed |
| Terminal and agent orchestration | Run by the agent, not from the app UI |
| Visibility | Private to your workspace; the public app registry has its own review process |
The app UI receives a small Construct SDK for workspace files, app-scoped storage, notifications, window controls, and explicitly permitted calls. It does not inherit every tool available to the agent that authored it. The agent uses broader execution surfaces while building, but the finished interface runs with its own narrower contract.
Permissions are part of the design
An internal tool should not gain broad access simply because it was generated inside a trusted workspace. A workspace app declares the native capabilities, connected apps, and network origins it intends to use. Calls outside those allowlists are denied at runtime.
The platform also rechecks current workspace membership and permissions when the app makes a gateway call. File operations remain subject to the user's workspace access, and an app cannot use its file access to rewrite another app's package. Direct network access is limited to exact HTTPS origins rather than wildcard domains.
For a request tracker that only reads and writes workspace files, that can mean declaring Files access and nothing else. A tool that posts an approved record to a connected service needs that specific app target as well. The default is the smallest permission set that completes the job.
Validation, then verification
Before launch, Construct checks the package structure and manifest, follows relative imports, enforces package boundaries, compiles the TypeScript and JSX, and reports design findings. A successful validation creates an immutable artifact with a deterministic build ID and opens the app.
Validation proves the package builds under the workspace runtime. Click through the main actions once after a change, especially after changing permissions, storage behavior, or an external call, and Construct helps with the rest: it captures console, network, gateway, and runtime-error diagnostics from workspace apps, inspects recent errors, repairs the source, validates again, and asks you to verify the changed behavior. The loop is closer to maintaining a small internal product than generating a disposable code snippet.
Draft changes do not replace the last good build
A successful build becomes the current runtime artifact. Later source edits mark the app as having draft changes, but opening the app continues to run the last successful build until the new source validates.
If validation fails, the working artifact is not replaced. Cancelling an edit also leaves both the source and prior build in place. The desktop shows whether an app is ready, has draft changes, or has never built successfully. This is a last-good-build fallback that protects the working tool while a repair is in progress, rather than a browsable history of every build.
Internal tool or workflow?
Not every repeated process needs an interface. Use a workspace app when people need to view, enter, filter, or review state. Use a Construct workflow when the main value is running a known sequence of agent steps, connected-app actions, and notifications. Use a scheduled agent job when the process should run in the background without waiting for someone to open a screen.
The two support the same operation. A workspace app provides the queue and review surface, files hold durable records, and a scheduled job prepares new work. Automation still runs through the execution surfaces and permissions designed for it, so the app UI stays a focused interface.
Before you start: where workspace apps fit
Good fits are narrow tools with a clear operator and a durable workspace context:
- Intake forms and review queues
- Checklists for repeated operations
- Lightweight trackers over workspace records
- Dashboards for files or app-scoped JSON data
- Focused interfaces over an explicitly allowed connected service
- Small utilities that make an existing workflow easier to run and inspect
A workspace app is an internal interface for one workspace, not a public product. For a customer-facing site, a product that needs the wider npm ecosystem, or a system with its own infrastructure and security controls, keep a dedicated hosting stack, and let Construct build the internal tools around it.
Already tracking this in a spreadsheet? Move it in 15 minutes
- Start on Lite for $9, or take the 7-day Pro trial.
- Upload the spreadsheet or checklist you use today to the workspace.
- Describe the app: the records it manages, the actions an operator takes, and what "done" looks like.
- Open the first build and use it on real records; ask for the missing field or the clearer status.
- Pair it with a schedule if new items should be prepared automatically, using one of your Calendar tasks.
There is no import wizard: Construct reads the file you upload and builds the app around it.
Which should you choose?
Choose Construct for small internal tools. It starts at $9 a month, builds the app from a plain-language description, keeps it private to your workspace with declared permissions, and protects the working build while it improves the next one.
Give Construct the process your team tracks in a spreadsheet today, such as research requests or review queues: name the records, the actions an operator takes, and the files it may read, then ask for the smallest interface that makes the process easier to run. Open the first build and use it on real records the same day. Start on Lite at $9, or try Pro free for seven days.
Related resources
Frequently asked questions
- Can Construct build internal tools for my team?
- Yes. From $9 a month, Construct builds private workspace apps such as request trackers, intake forms, review queues, and dashboards. You describe the process in plain language, and the agent writes the React and TypeScript, validates it, and opens it in your web desktop.
- What happens if a build breaks?
- Construct keeps the last successful build running, so a failed update does not take the working tool offline while it reads the app's diagnostics, repairs the source, and validates again.
- Do I need to be a developer to use it?
- No. You describe the records the app manages, the actions an operator takes, and the files it may access. Construct writes, validates, and publishes the app, and updates it later as the process changes. The source stays in your Files if anyone wants to read it.
- What can a workspace app access?
- Only what it declares. Each app lists the native capabilities, connected apps, and exact HTTPS origins it needs, and calls outside those allowlists are denied at runtime. File operations stay subject to the user's workspace access.
- When is a workspace app the wrong choice?
- A workspace app is a private internal interface for one workspace. For a public customer-facing site, a product that needs the wider npm ecosystem, or a system with its own infrastructure, keep a dedicated hosting stack and let Construct build the internal tools around it.
Keep reading
What is an AI agent with its own computer?
An AI agent with its own computer keeps working in a cloud workspace after you close your laptop. How it works, the risks, and who offers one, from $9 a month.
- ai-agent
- always-on-agent
- ai-employee
- product
Nobody merges an email
Agents made work cheap to produce and no cheaper to check. Why software absorbed the flood, why the rest of the business did not, and the five things that make agent work checkable.
Your agent has a half-life
Why AI agents keep failing on long multi-step jobs: a 95% reliable agent finishes 48 steps 8.5% of the time. The fix is a resumable run, not a better model.
What is an AI employee?
Learn what an AI employee is, how it differs from a chatbot, and how Construct completes work across email, Slack, browser, and connected apps.
