# Workflows > Draw a pipeline on the builder canvas, wire its steps together, save it as a version, and activate it. A workflow is a graph. Steps are the boxes, and an arrow from one box to another means "wait for that one to finish, then run". Hodoflow executes the graph, not a script, so anything without an arrow between it and another step runs in parallel. **Workflows** in the sidebar lists everything you have saved. **New Workflow**, or opening an existing one, takes you to the builder. The builder is desktop only — on a phone you get the read-only graph view instead. ## The canvas A brand-new workflow opens on the trigger wizard, because a workflow with nothing to start it can never run. Pick **Webhook**, **Scheduled**, **Kafka**, or **No trigger (subflow)**; the next screen configures whichever you chose. [Triggers and webhooks](/help/workflows/triggers-and-webhooks) covers all of them. Steps come from a searchable palette rather than a drag-and-drop tray. Press `Cmd`/`Ctrl` + `K`, or click the `+` beneath a selected step, or click the `+` under the **Start here — add the first step** card on an empty canvas. Type in **Search steps…** and the palette narrows; a **Recent** section remembers what you reached for last. The `+` under a selected step also offers **On error…** and **Compensate…**. Those add the same step types, wired with a different kind of arrow: an on-error edge runs its target only when the parent fails, and a compensate edge runs its target to undo work when the run unwinds. You can change any existing arrow between **Normal**, **On error**, and **Compensate** by right-clicking it. Connections are drawn by dragging from one step's handle to another's. While you drag, every step the connection cannot legally reach outlines itself in red — that is either a cycle, or a step already reachable from where you started. Hodoflow keeps the graph free of arrows that another path already implies, so adding a shortcut can silently remove one; a toast at the bottom of the screen tells you when that happened and why. A step with no arrow coming into it shows **No parent step** on the card and counts toward the **invalid nodes** total in the toolbar. It will not stop you saving, but it will stop you activating. Click a step to open the inspector on the right. **Name** is the label you see on the canvas, **API Name** is the identifier formulas use to reach this step's output, and the **Configuration** and **Policy** tabs hold everything else. Select several steps and the panel switches to **Duplicate**, **Copy**, and **Delete** for the whole selection. Positions do not matter. Only the connections are saved, and Hodoflow runs its own layout every time you open a workflow — **Auto Layout** in the toolbar re-runs it on demand and never counts as a change. ## Step types Thirteen step types exist. Seven have an entry in the step palette and can be added to a workflow today; the other six run in the engine but have no way into the builder yet. The identifier in the first column is what the **Type** column shows when you inspect a run. | Step type | Palette name | What it does | |---|---|---| | `http_request` | **HTTP Request** | Calls any HTTP endpoint and passes the response to the steps below. | | `api_request` | **API Request** | Calls one operation from an imported API Spec, with its parameters typed for you. | | `harmonizer` | **Map to Data Model** | Maps incoming fields onto a data model and saves the result to the warehouse as a record. | | `filter_list` | **Filter List** | Keeps only the items of an upstream list that match a condition, and skips the branch below when nothing survives. | | `data_transform` | **Data Transform** | Reshapes, renames, and computes fields with formulas before the next step reads them. | | `conditional` | **Conditional** | Evaluates its branch conditions top to bottom and sends the run down the first one that matches, or down **Default**. | | `subflow` | **Subflow** | Runs another workflow as a run of its own and continues with whatever it returns. | | `delay` | — | Parks the run until a fixed instant and frees the slot until then. The deadline is recorded when the run suspends, so a resume never restarts the clock. | | `kafka_produce` | — | Publishes one message to a Kafka topic through a Kafka connector. | | `snowflake_query` | — | Runs a `SELECT` against Snowflake through a Snowflake connector. | | `snowflake_write` | — | Writes rows into a Snowflake table, by insert, merge, or truncate-and-insert. | | `s3_read` | — | Downloads one object, or lists objects under a prefix, from S3-compatible storage. | | `s3_write` | — | Writes one object to S3-compatible storage. | ### Running a step once per item **HTTP Request**, **API Request**, **Data Transform**, and **Subflow** each get an **Iteration** section in the inspector. Leave **Mode** on **Run once with the whole payload** and the step behaves normally. Switch it to **Run once per item** and you get: - **Source list** — the path to the list to walk, like `fetch.data` for an upstream step's whole output or `fetch.data.stations` for a list inside it. Each element is available to formulas as `item`. Upstream lists Hodoflow can already see are offered as clickable chips. - **When an item fails** — **Fail the run**, or **Skip and count**, which attempts every element and lets the step succeed on the rest. - **Advanced** — how many items run at once, a delay between them, and a **Max items** ceiling that refuses the step before any item runs. ## Saving and versions Nothing is written until you press **Save**. Until then the toolbar shows **Unsaved changes**, and you can rearrange, delete, and reconnect freely. Every version of a workflow is its own record, sharing a group with the others. Editing a version that has been published turns the button into **Save as New Version** and the indicator into **Draft changes**: your edits fork a new version rather than rewriting the one that is live. A version that was never published keeps taking plain saves. The version number sits beside the workflow name in the header as `v1`, `v2`, and so on. **Edit details** changes the name and description without touching the graph. Back on the **Workflows** list, each workflow collapses into one row with a version count. Expand it and every version is there with its state — **Active**, **Inactive**, or **Draft** — plus **Published** date, and **Activate**, **Deactivate**, **Delete**, and **Open** as appropriate. Deleting a version deletes its runs too, and cannot be undone. ## Activation A saved workflow is not a running workflow. Press **Activate**. The button stays disabled while anything obvious is wrong, and its tooltip says which thing: *Save changes before activating*, *Fix invalid nodes before activating*, *Add at least one trigger before activating*, *Add at least one step before activating*, *Connect every step to the flow. Some have no incoming path*, or *Re-sample stale sources before activating*. When you press it, Hodoflow checks the version properly and refuses with a message naming the first problem it found: - A workflow must have at least one trigger, unless it is a subflow. Subflows are invoked by other workflows, so they are exempt. - Every step must be fully configured. - Every **Map to Data Model** step needs field mappings, and its data model needs a primary key field that is mapped. - The steps must connect to the trigger, with no disconnected steps. - The graph must have no cycles. - No source schema may be stale. If a source's configuration changed since it was last sampled, a banner offers **Re-sample** per source. Activating a version deactivates whichever version of that workflow was live before, so exactly one version of a workflow ever runs. **Deactivate** takes it offline: webhooks for that workflow start being rejected, its schedules are cancelled, and its Kafka consumers stop. Runs already in progress continue to completion. Deactivating is always allowed and never forks a version. ## Watching runs **Workflows → Monitor** is the run history. Its table shows **Time**, **Workflow**, **Status**, **Duration**, and **Source**, filtered by workflow version and by time range. Turn on **Live Tail** and new runs stream in as they happen. A run's status is one of `pending`, `running`, `completed`, `failed`, or `suspended`. Suspended is normal — it is what a run looks like while it waits out a delay or waits for a subflow. Click a run to open its steps in order, each with its own status, duration, and error. A failed run also gets a **Run again from the failure** panel with a **Retry Run** button. Every step that finished is kept, and only the steps that never finished run again. Occasionally a run stops between issuing an external call and recording what came back, and the panel above the retry asks you to say what happened under **Unaccounted-for external calls**. Check with the provider, then answer **Never sent** — which lets the step run again — or **Went through**, which keeps the failure rather than repeating the call. Retry stays disabled until you do, because Hodoflow cannot know whether repeating the call is safe. The **Analytics** tab on the same page totals **Total Runs**, **Completed**, **Failed**, and **Avg Duration** for whatever the filters are set to, with breakdowns by workflow and by status. From inside the builder, the **Monitor** button in the toolbar opens the same page pre-filtered to the workflow you are editing.