# Blocks reference (/docs/agents/flow-nodes)

Blocks are the building pieces of a flow. Each one you add becomes a step on the canvas, and its settings open in the side panel. This page lists every block in the order the **Blocks** dock shows them, with what it does and what you can set.

- **Who can use it:** Owners and admins can add blocks and change their settings. Members and viewers can open any step and read its settings, but the fields are read-only and the Blocks dock is hidden. See [Roles and permissions](/docs/settings/roles).
- **Before you start:** Open a flow in the builder. See [Flow builder basics](/docs/agents/flow-builder).

## Settings every step has [#settings-every-step-has]

Whatever its type, a selected step's settings start with a **Step** section:

| Setting | What it does |
| --- | --- |
| **Name** | What the step is called on the canvas, in runs and in tests. |
| **Reference name** | The short name later steps use to read this step's results, as `{{steps.<name>.…}}`. Lowercase letters, numbers and underscores, starting with a letter, and unique in the flow. Renaming it doesn't update steps that already use the old name. Triggers don't have one. |
| **Notes** | For teammates: why the step is here. Runs ignore it. |
| **Run this step** | Off: runs pass over the step as if it weren't there, and its card shows **Off**. On an If / else or Switch, every run skips the condition and all its paths. Triggers can't be switched off. |

![The Step section for Extract lead details: Name, Reference name extract with the hint Later steps read its output as steps.extract, Notes, and the Run this step switch](/screenshots/flow-nodes/08-step-basics.webp)

Below the step's own settings:

- **If it fails** (on steps that call out or change data: AI, message, data, integration and app steps, and **Run another flow**): choose **Stop the run and flag it** (the default), **Retry, then stop** (with **Tries**: 1, 2, 3 or 5 more, and **First wait**: 5s, 30s, 1m or 5m) or **Carry on to the next step**. The agent's run log, where failed runs will show, is [coming soon](/docs/agents/agent-runs). When a step carries on after failing, later steps see its outputs as empty, so check them with an If / else.
- **Last 7 days:** how the step did in published runs, such as "142 passed · 18 held", or "New in this draft. It runs once you publish."
- **Outputs:** the values later steps can read from this step, with sample values.
- **Move up**, **Move down**, **Move into a path** and **Remove this step**. See [Select, move and remove steps](/docs/agents/flow-builder#select-move-and-remove-steps).

![The Outputs section of Extract lead details: Budget they mention, How many people will use it and An email address in the message, each with its steps.extract variable and a sample value](/screenshots/flow-nodes/10-outputs.webp)

### Use values from earlier steps [#use-values-from-earlier-steps]

Text settings that take values from the run (a message, a URL, a payload, a field value) have a **\{ \}** button beside their label. Click it to pick a variable: the trigger's data, the contact and deal, and the outputs of every step above this one. It's inserted where your cursor is, as `{{contact.name}}`, `{{steps.classify.intent}}` and so on. Under the field, **Preview** shows the text with sample values filled in.

If a setting uses a variable this step can't see (it comes from a later step, a step on another path, or a step that was removed or renamed), the field turns red with "… isn't available at this step", and the flow can't publish until you fix it.

## Triggers [#triggers]

Every flow starts with exactly one trigger. Picking another one replaces it. See [Choose a trigger](/docs/agents/flow-builder#choose-a-trigger).

| Block | Starts a run when | Settings |
| --- | --- | --- |
| **Any channel message** | A customer writes on any connected channel. The step is called **Inbound message received**. | **Listens on**: the channels to listen to (at least one). **Skip the run when**: **Never skip**, or a teammate replied in the last hour, 24 hours (the default) or 3 days. **Start one run per**: **Conversation** (the default) or **Message**. |
| **Instagram comment** | Someone comments on or mentions your Instagram account. | **Listens on** Instagram, plus the same **Skip the run when** and **Start one run per** settings. |
| **Deal stage change** | A deal moves from one stage to another. Later steps read the stages as `{{trigger.from_stage}}` and `{{trigger.to_stage}}`; add a [Filter](/docs/agents/flow-conditions#stop-runs-with-a-filter) to act on one stage only. | **Skip the run when**, **Start one run per**. |
| **Schedule** | Every weekday at 09:00. Later steps read the time as `{{trigger.scheduled_at}}`. | **Skip the run when**, **Start one run per**. The timetable itself can't be changed. |
| **Webhook received** | **Coming soon.** Another app (n8n, Zapier, a form, your backend) will post to a Mokanata webhook URL. Flows don't run on the server yet and the webhook address doesn't exist yet, so this trigger can't start a flow today. | See [Start a flow from a webhook](/docs/agents/flow-connections#start-a-flow-from-a-webhook). |
| **Payment received** | A payment link is paid, fails or is refunded. | **When a payment link is**: **Paid**, **Failed** or **Refunded**. Covers every payment link and gateway on the [Payments](/docs/payments/payments) page. |
| **Record created or updated** | A contact or a custom object record changes. | **Object**, and **When a record is**: **Created**, **Updated** or **Created or updated**. Changes made by this flow don't start it again, so it can't loop. For a table synced from your database, changes made in the database itself start it too. |
| **Tag added** | A contact gets a tag, from a person, an import or another flow. | **Tag**. |

![The settings of the Inbound message received trigger: Listens on, with every channel ticked, then Skip the run when set to A teammate replied in the last 24h and Start one run per set to Conversation](/screenshots/flow-nodes/02-message-trigger.webp)

## Agent actions [#agent-actions]

The agent's own work on a conversation. These steps use the agent's model, voice, knowledge and channels from its setup.

| Block | What it does | Settings |
| --- | --- | --- |
| **Draft reply** | The agent writes a reply from your sources. The step is called "*Agent* drafts reply". | **Drafting**: **Model** and **Temperature** (lower keeps replies close to the sources, higher varies the wording). **Voice**: **Tone** and **Instructions** (saved when you leave the field). |
| **Send reply** | Sends the draft on the customer's channel and logs it to the contact's timeline. The step is called **Send automatically**. | **Wait out quiet hours**: holds messages during the agent's quiet hours, in each contact's time. Change the hours on the agent's [Settings](/docs/agents/agent-settings) tab. |
| **Hand to a person** | A teammate takes the conversation. The customer sees nothing until a person replies. | **Goes to**: **The deal owner**, you, or **The revenue team**. **If no one acts within**: 30m, 2h, 4h or 1d, after which it moves to you (or, if it's already yours, shows as overdue). |
| **Book meeting** | Finds time on both calendars and books a meeting. | Only the settings every step has. |

A **Send reply** must come after a **Draft reply**, and behind a guardrail gate or an approval step, or the flow can't publish. See [Guardrails](#guardrails).

![The Drafting and Voice sections of Closer Agent drafts reply: model mokanata-sales-v3, temperature 0.3 precise, tone Direct, warm, concise, and the drafting instructions](/screenshots/flow-nodes/04-draft-reply.webp)

## AI [#ai]

Steps that read the message or the record and hand their results to later steps. Each has a **Model** setting.

| Block | What it does | Settings | Outputs |
| --- | --- | --- | --- |
| **Classify message** | Sorts the message into one of the labels you choose. It also detects the language and urgency. | **Sort each message into one of**: labels such as pricing, support, partnership, spam (**Add a label**). | Intent, Confidence, Language, Urgency |
| **Extract fields** | Pulls values such as a budget, an email or a date out of the text. Fields it can't find come back empty, never guessed. | **Fields to extract**: for each, a name, a type (text, number, yes / no, date) and what to look for. **Add a field**. | One per field |
| **Generate text** | Runs your own prompt, for example a two-sentence summary for the deal owner. | **Instructions** (the prompt, with variables), **Model**, **Temperature**. | Generated text |
| **Score lead** | A 0 to 1 fit score from the record and the thread. | **Model**. | Score, Reason |
| **Enrich contact** | Company, size and role from public sources. | **Model**. | Company, Industry, Employees, Role |
| **Route to owner** | Assigns the right teammate or team. | **When it can't tell, assign to**: **The deal owner**, **The revenue team, in turn**, or you. **Model**. | Owner, Team |

![Fields to extract on Extract lead details: budget (number), team_size (number) and email (text), each with a description, Add a field, and the If it fails section set to Stop the run and flag it](/screenshots/flow-nodes/09-extract-fields.webp)

## Messages [#messages]

| Block | What it does | Settings |
| --- | --- | --- |
| **Send template** | Sends an approved WhatsApp template, any time of day, including outside WhatsApp's 24-hour window. | **WhatsApp template**, then a value for each of its placeholders. Only approved templates send; the step says when the one you chose isn't. See [WhatsApp message templates](/docs/campaigns/message-templates). |
| **Send payment link** | Creates a payment link with your default gateway and sends it (payment gateways are coming soon). A paid link starts any **Payment received** flows. | **Amount** (a number or a variable such as `{{deal.value}}`), **Currency**, **What it's for**. See [Payments](/docs/payments/payments). |
| **Notify a teammate** | Tells a teammate something. The customer never sees it. | **Notify**: **The deal owner**, **The revenue team** or you. **Through**: **Mokanata inbox**, **Email** or **Mobile push**. **Message**. |

## Logic [#logic]

| Block | What it does | Settings |
| --- | --- | --- |
| **If / else** | Two paths: one when its conditions match, one when they don't. | **Paths** and **Try it**. See [Conditions and paths](/docs/agents/flow-conditions). |
| **Switch** | One path per value, plus a fallback. After a **Classify message** step it starts with a path per label. | **Paths** and **Try it**. |
| **Filter** | Stops the run unless its conditions match. Stopped runs count as filtered, not failed. | **Continue only when** and **Try it**. |
| **Wait** | Pauses the run. A new message from the contact doesn't start a second run meanwhile. | What to wait for: **A while** (**Wait for** 5m to 7d), **A reply** (**Give up after** 5m to 7d, then carry on with "They replied" set to no) or **A time** (**Wait until**, such as "09:00 next business day", or a date variable). The step's name follows these settings until you rename it. |
| **Set variables** | Names values for later steps, read as `{{vars.name}}`. Setting one again replaces it. | **Variables**: a name and a value for each (**Add a variable**). |
| **Run another flow** | Hands the run's contact and variables to another flow of the same agent. | **Run** (the flow), **Wait for it to finish** (off: start it and carry on straight away). |
| **Stop** | Ends the run here. | **End the run as**: **Completed**, or **Failed, so it shows under Runs**. **Reason**. |

![The Wait step's settings: A while, A reply and A time, with Wait for set to 1h and the note that a new message from the contact doesn't start a second run](/screenshots/flow-nodes/14-wait.webp)

## Data [#data]

Steps that change records. Fields marked read-only in the data model are greyed out; allow agents to write them in [Objects](/docs/contacts/objects).

| Block | What it does | Settings |
| --- | --- | --- |
| **Update CRM record** | Writes values onto the run's contact and its deal, or onto a record of another object. | **Update the**: the object. For objects other than contacts, **Find it by** and **Matching** pick the record. **Field mapping**: a field and a value for each (**Map a field**). |
| **Create record** | Adds a record to any object, such as an Order. | **Create a**: the object (objects flows can't create are marked). **Field mapping**; required fields are marked with an asterisk. |
| **Find record** | Looks up a record for later steps. | **Find a**: the object. **Where**: the field to match. **Matches**: the value. **Create one when nothing matches** (off: "Found" is no, and later steps can branch on it). |
| **Change deal stage** | Moves the contact's deal to a stage. It never moves a deal backwards. | **Move the deal to**: Lead, Qualified, Discovery, Negotiation or Closed won. |
| **Add or remove tags** | Tags the contact, or removes tags. | **Add** or **Remove**, then **Tags**. |
| **Map data** | Reshapes values into the field names another step needs, such as a webhook payload. | **Output fields**: a name and a value for each. Later steps read them as `{{steps.<name>.field_name}}`. |

![Update CRM record on Qualify the deal: Update the Contacts, then a field mapping that sets Deal stage to qualified and Deal value to the extracted budget, with Map a field](/screenshots/flow-nodes/11-update-record.webp)

## Integrations [#integrations]

| Block | What it does | Settings |
| --- | --- | --- |
| **HTTP request** | Calls any REST API. | **Request**: **Method** (GET, POST, PUT, PATCH, DELETE), **URL** (https, or a variable) and **Authentication** (**None**, or a saved API connection; credentials stay with the connection, never in the flow). **Parameters**: **Query**, **Headers** and **Body** tabs (**Send as** JSON, **Form fields** or **No body**). **Response**: a **Sample response** whose fields become variables before the request ever runs, and **Time out after** (10 to 120 seconds). |
| **Send webhook** | Posts the run's data to a saved destination, signed with `X-Mokanata-Signature` and retried with backoff. | **Send to**: one of your outgoing destinations. **Payload** (JSON, with variables). **Send a test delivery**. See [Post data to another app](/docs/agents/flow-connections#post-data-to-another-app). |

![The Request and Parameters sections of an HTTP request: POST to the Helpdesk API tickets URL, authenticated with the Helpdesk API connection, and a JSON body built from the contact's name, the classified intent and the contact's email](/screenshots/flow-nodes/12-http-request.webp)

**Send a test request** at the bottom of an HTTP request shows **200 OK** with the sample response and how long it took, marked "sample data, nothing was changed". It needs the **Test agents** permission.

![The Response section after a test: the sample JSON response, Time out after 30 seconds, and a 200 OK result marked sample data, nothing was changed](/screenshots/flow-nodes/13-http-response.webp)

## Apps [#apps]

One block for each app connected in [Integrations](/docs/agents/flow-connections), such as n8n, Google Sheets, HubSpot or Stripe, with its account under the name. An app whose sign-in expired says **Needs reconnecting**. **Manage** opens Integrations. With nothing connected, the group says "No apps connected. Connect n8n, Sheets, HubSpot and more in Integrations."

![The Apps and Guardrails groups at the bottom of Blocks: n8n, Google Sheets, HubSpot marked Needs reconnecting and Stripe, then Guardrail checks, Value threshold, Approved language, Citation required, Human approval gate and WhatsApp 24h window](/screenshots/flow-nodes/01-apps-guardrails.webp)

An app step's settings:

| Setting | What it does |
| --- | --- |
| **Account** | Which connection of that app to use. |
| **Action** | What to do, depending on the app: for example **Add a row** or **Look up a row** in Google Sheets, **Run a workflow** in n8n, or **Create or update a contact** and **Create a deal** in HubSpot. |
| **Fields** | The action's own fields; required ones turn red while empty. |

If the app isn't connected any more, the step says so with a **Connect it** button; if its sign-in expired, reconnect it in **Manage**.

![A Google Sheets step: Account Google Sheets, Action Add a row (Appends one row; each column takes a value or a variable), then Spreadsheet, Sheet and Columns fields](/screenshots/flow-nodes/15-app-step.webp)

## Guardrails [#guardrails]

A guardrail gate checks every draft against rules before anything sends. Without one (or a **Human approval gate**), a flow with a send step can't publish.

| Block | What it adds |
| --- | --- |
| **Guardrail checks** | A gate with all of the agent's rules, placed before the first send or handoff. |
| **Value threshold** | The rule "Deal value under $100,000": holds any deal of $100,000 or more. |
| **Approved language** | The rule "Uses approved language only": warns on wording outside the approved list. |
| **Citation required** | The rule "Claims cite product knowledge": every claim must cite a knowledge source. |
| **Human approval gate** | The rule "Every reply waits for approval". |
| **WhatsApp 24h window** | The rule "WhatsApp session window open": outside WhatsApp's 24-hour window, it sends an approved template instead (a hard limit, WhatsApp only). |

The rule blocks go into the selected gate (or the flow's first gate). If the flow has no gate yet, they make one. If the gate already checks that rule, a toast says so; its switch is in the step's settings.

A **Guardrail checks** step's settings:

- **Rules:** each rule with its switch. Use the **…** menu to **Edit rule** or **Delete rule** (which only takes it off these checks; it stays on the agent's [Guardrails](/docs/agents/agent-guardrails) tab). **Add rule** writes a new one or starts from a template.
- **Grounded in:** the knowledge sources claims are checked against (**Edit sources** jumps to the knowledge step) and **Require inline citation**: every number and claim links to the passage it came from.
- **Test these checks on the last 20 drafts:** replays the rules on recent drafts and shows how many would have passed, been held, handed off or blocked, rule by rule. It needs the **Test agents** permission. For an agent in shadow, held drafts only show in its digest.

![The Rules section of Guardrail checks with Add rule, and Price within approved band: holds any price more than 15% below the tier price, with its condition and its Holds outcome](/screenshots/flow-nodes/05-guardrail-rules.webp)

![Grounded in: Pricing & packaging, Security & compliance, Product docs v4.2 and Objection playbook, with Require inline citation on](/screenshots/flow-nodes/06-grounded-in.webp)

![The result of a gate test: Last 20 drafts: 17 passed (1 with a warning), 3 held, with 2 held by the price rule, 1 held by the citation rule and 1 warning on tone](/screenshots/flow-nodes/07-gate-test.webp)

For what each rule outcome means and how to write a rule, see [Guardrails and rules](/docs/agents/agent-guardrails).

## Steps from the agent's setup [#steps-from-the-agents-setup]

Flows built with the agent also contain steps that aren't in Blocks. You can open, rename, move, switch off and remove them like any other step.

| Step | What it does | Settings |
| --- | --- | --- |
| **Retrieve product knowledge** | Finds the passages the draft answers from. | **Sources**: switch each of the agent's sources on or off for this flow. **Retrieval**: **Passages per answer** (4, 8, 12 or 16), **Match strictness** (higher cites fewer, closer passages; lower answers more often) and **Use docs updated in the last** 30 days, 90 days, 1 year or any age. |
| **Hold for approval** | The draft waits for a teammate to approve it. | **Goes to**, **If no one acts within**, as for **Hand to a person**. |
| **Every rule passes** / **A rule stops it** | The two branches after a guardrail gate: when every rule passes, the next step (usually **Send automatically**) runs; when a rule stops the draft, the other one (usually **Hold for approval**) does. Moving or removing one branch moves or removes both. | Each branch's step has its own settings. |
| **Write to CRM and timeline** | Writes a summary, deal fields and an event after every outcome, once per message even on a retry. It stays at the end of the flow. | None. |
| Signals, CRM read, Match, Merge, Billing and similar | The agent's own pipeline steps, for example reading product usage or merging duplicate records. | Only the settings every step has. |

![The Sources and Retrieval sections of Retrieve product knowledge: four sources with their sync state and switches, 8 passages per answer and match strictness 0.72](/screenshots/flow-nodes/03-knowledge.webp)

At the bottom of Blocks, the agent's knowledge (such as **Closer Agent's knowledge**) lists the agent's sources with their document counts and sync state, and how many are ready. **Sync now** syncs them; it needs the **Edit agents** permission. To add or remove sources, use the agent's [Knowledge](/docs/agents/agent-knowledge) tab.

## Next steps [#next-steps]

<Cards>
<Card title="Conditions and paths" href="/docs/agents/flow-conditions" description="Branch the flow and compare values from the run." />
<Card title="Apps and webhooks in flows" href="/docs/agents/flow-connections" description="Connect the apps and webhooks these blocks use." />
<Card title="Edit a flow as code" href="/docs/agents/flow-code" description="Every block has a use value you can write in YAML." />
</Cards>