← Retour au plan
Live Artifacts Mastery: From Mission to Marketplace · Leçon 3 sur 8

3. Forging from Plain Language on the Forge Page

Create a Live Artifact from scratch by describing the workflow in plain language and letting @oupi identify the variable inputs.

The Forge page is where you create a Live Artifact from scratch, without needing a prior mission. Navigate to the forge page and describe the workflow you want in plain language — for example, "Every week, summarize the top five news articles about renewable energy and format them as a briefing for my team." @oupi reads your description, identifies what changes between runs (the variable inputs such as a topic, a date range, a file, or a list of keywords), and builds a structured input form for you. This is the second creation path: instead of converting a successful mission, you start from intent and let @oupi extract the parameters.

When @oupi analyzes your plain-language description, it proposes a set of declared inputs — the values that will change from one run to the next. Each input has a type (text, file, date, list of choices), a required-or-optional flag, and validation rules. Files are pulled from your file space. For example, a "weekly competitor report" artifact might declare inputs for the competitor name (required text), the reporting period (required date range), and an optional reference document (file). Getting these declarations right at forge time is critical: they determine what the input form looks like for every future user who runs or installs your artifact.

Astuce

Write your forge description the same way you'd write a good mission: state the desired outcome, who it's for, and the output format. Don't list steps — @oupi decides those. Then explicitly name what varies between runs ("the client name changes", "the CSV file changes each time"). The clearer you are about what's fixed vs. variable, the fewer clarification rounds @oupi needs.

During forging, @oupi may ask clarification questions if your description is ambiguous — just as it does in any mission. For instance, if you say "generate a report from data," it might ask which data source, what format, or whether charts are needed. Answer concisely; your answers are persisted and shape the final workflow. If you'd rather @oupi proceed with reasonable defaults, say so. Once the inputs are confirmed, the artifact is created and ready for its first run.

After forging, your new Live Artifact appears in your artifact list. You can immediately fill the declared inputs and run it on demand, attach a schedule (e.g., every Monday at 8 AM), or wire a trigger (a new email matching a rule, a webhook from another tool). Each run tracks progress live, stores deliverables in a dedicated folder in your files, and logs the full history — inputs, cost, and verification verdict. You can also share the artifact with your team: each member gets their own instance with their own inputs and history, billed to them.

À vous de jouer

Go to the Forge page. Describe a real workflow you repeat regularly in two to three sentences — for example, "Take a PDF invoice, extract the vendor name, amount, and due date, and add a row to a summary spreadsheet." Let @oupi propose the variable inputs, review them (adjust types, mark required vs. optional), confirm, and run the artifact once with sample data.

Suivre ce cours dans OUPI → Cet exercice se fait dans la plateforme OUPI.
Astuce

After your first successful run, check the run's deliverables and verdict in the history. If the output isn't quite right, reply from the artifact to refine the workflow — @oupi keeps the context. Once satisfied, consider sharing it with your team or publishing it to your organization's marketplace.

À retenir

You can create a Live Artifact entirely from plain language on the Forge page: describe the workflow, let @oupi identify the variable inputs and build the input form, then confirm. Write your description like a good mission — outcome-focused, with explicit mention of what changes between runs. After forging, run it on demand or automate it with schedules and triggers. Every run is tracked with its inputs, cost, deliverables, and verdict. Share with your team or publish to the marketplace when it's proven.