Deliver results somewhere

Goal

One destination connected and one workflow result written to it. Done means data you produced shows up where someone else works — a warehouse table, a file drop, a downstream system — with an approval guarding the write.

Image: Integrations Discovery tab filtered to destinations.

Steps

1. Open Integrations and switch to the destination class. Sources bring data in; destinations are where results go when they need to live outside Semogram.
2. Pick from the Discovery catalog the system that should receive results. Prefer the place your team already reads — the warehouse they query, the storage they share — over a new location nobody checks.
3. Let setup assist: which account may write, which target (schema, path, project), and what shape the writes take (append, overwrite, upsert). Writes deserve narrower scope than reads — grant the minimum the job needs.
4. Read the configuration preview, then validate and commit exactly like a source: reachability, permissions, and a test write the setup can roll back. A destination that validates is one you can trust with real outputs later.
5. Back in Pipeline Studio, attach the destination to the workflow whose results matter: a destination node referencing the installed integration, with its write intent explicit in the graph.
6. Put an approval on the writing node before the first real run. External writes are the highest-consequence action in the loop, so the human gate goes as close to the write as possible — node level, not just workflow level.

Image: graph with a destination node carrying an approval gate, shown in permissions view.

7. Dry-run, inspect the preview, commit. Then verify at the other end: open the target system and confirm the rows, files, or records arrived in the expected shape.

When writes misbehave

• Validation fails on write permission: the authorizing account can read but not write. Re-authorize with write scope for that target only.
• Wrong shape at the target: the mapping into the destination is off, not the connection. Fix the upstream mapping node and dry-run again.
• Duplicates on reruns: the write mode appends where it should upsert. Change the mode deliberately and rerun.

What good looks like

• One destination, healthy, receiving scheduled or on-demand writes.
• Every external write passes through an approval you placed.
• The receiving team can confirm the data without asking you.

Next

• Inspect what ran: Read your first run.
• Approval and policy depth: Operators.