Step Reference

Every trigger and step available in the flow editor, with what it takes and what it returns. For how to put them together, see the Projects & Flows guide.

Reading this page:

• Configuration lists the fields you fill in on the node. A field marked required must be set before the flow validates.
• Outputs are what the step publishes into the run context. A step's outputs are namespaced under the output name you give the node, so a Smart Agent named triage publishes {{triage.response}}.
• How it behaves appears on the steps whose real configuration is not obvious from the field list — branching, loops and agents especially.

Fields whose notes say Shown when … only appear once the field they depend on has that value.

---

Triggers

Peer Message

Key: peer.message

Triggered when a message is received from a peer

Outputs

Output	Type	Description
message	text	The message content from the user
attachments	array	Any files or media attached to the message
messageId	string	Identifier of the user message that triggered the flow
responseMessageId	string	Identifier of the assistant's response message when available
history	array	Previous messages in the conversation
conversationId	string	Identifier of the conversation
userId	string	Identifier of the user

Peer Action

Key: peer.action

Triggered when an action is received from a peer

Configuration

Field	Type	Notes
inputs	list	Each entry: name, type, description.

API

Key: api

Triggered when an API request is received

Form

Key: form

Triggered when a form submission UI is received

Configuration

Field	Type	Notes
inputs	list	Each entry: name, type, description.

Task

Key: task

Triggered when a Task in this flow's project is run. Give a flow this trigger to make it selectable when creating a task; the task's name, detail and attachments arrive as trigger outputs.

Outputs

Output	Type	Description
taskId	string	Identifier of the task that started this run
name	string	The task's name, as typed by whoever created it
detail	long-text	The task's free-text detail / instructions
attachments	array	Files attached to the task
priority	string	low | normal | high
flowProjectId	string	The project the task belongs to
task	object	Every field above in one object — handy for passing the task to an agent prompt or a subflow in one piece, e.g. {{task}}.

Schedule

Key: schedule

Triggered automatically based on a schedule defined by a cron expression

Configuration

Field	Type	Notes
cron	text, required	A cron expression to define the schedule (e.g., '0 * * * *' for every hour)
inputs	code	Default inputs to provide when triggered by schedule

Steps

AI & Language Models

Speech to Text

Key: speech-to-text

Transcribe audio (base64) to text

Configuration

Field	Type	Notes
base64Key	select	Key in context containing base64 audio (data URL or raw base64)
base64	long-text	Provide base64 directly if not using a key

Outputs

Output	Type	Description
language	string	—
duration	number	—
text	long-text	—

Agent

Key: agent

Use ReAct agent pattern to solve complex tasks with actions.

Configuration

Field	Type	Notes
modelId	model-select	If left empty, the default workspace agent model will be used.
includeChatHistory	checkbox	Include previous conversation messages in the agent context. This allows the agent to maintain conversation continuity and reference past interactions. Default false.
systemPrompt	long-text	Instructions for guiding the agent's behavior and capabilities. Default "You are Cognipeer Agent, a focused AI that solves tasks step-by-step and uses integrated tools when necessary.".
userMessage	long-text, required	The user's message or task for the agent to process.
actions	multiple-select	List of actions the agent can use to complete the task.
media	list	Each entry: type, url.
output_type	select	Choose how the agent should format its response. Default "text". One of text, structured_output.
jsonSchema	json-schema	Define the structure of the JSON output when using Structured Output. Shown when output_type is structured_output.
enableMemoryTools	checkbox	Let the agent read, search, save and delete Flow Project memory on its own via tool calls. Default false.
memoryScope	select	Default "project". One of project, flow, custom. Shown when enableMemoryTools is true.
memoryFlowId	text	Only used when Memory Scope is Flow. Leave as default for the current flow. Default "{{flow.id}}". Shown when memoryScope is flow.
memoryCustomKey	text	Only used when Memory Scope is Custom — any identifier to bucket memory by, e.g. {{trigger.userId}}. Shown when memoryScope is custom.
memoryOperations	multiple-select	Default ["read","search","put","delete"]. One of read, search, put, delete. Shown when enableMemoryTools is true.
enableSandboxTools	checkbox	Let the agent run shell commands, code, and file operations inside a sandbox via tool calls. Default false.
sandboxRef	text	Sandbox Record ID from the project's Sandbox tab (pick one already created), or a template expression resolved at runtime (e.g. from a Create Sandbox node's output). Shown when enableSandboxTools is true.
sandboxOperations	multiple-select	Default ["exec","run_code","list_files","read_file","write_file","mkdir","delete","move"]. One of exec, run_code, list_files, read_file, write_file, mkdir, delete, move. Shown when enableSandboxTools is true.
injectSecretNames	checkbox	Append the names of this Flow Project's secrets to the system prompt so the agent knows which credentials are available as environment variables. Values are never included. Default true.

Outputs

Output	Type	Description
response	long-text	—
reasoning	long-text	—
data	object	Parsed JSON data when using structured output.
toolCalls	array	List of actions called by the agent during execution.

Ask to LLM

Key: ask-to-llm

Send a prompt to the llm and get a response.

Configuration

Field	Type	Notes
modelId	model-select, required	—
prompt	long-text, required	—
media	list	Each entry: type, url.

Outputs

Output	Type	Description
response	long-text	—

Ask to Peer

Key: ask-to-peer

Send a prompt to the peer and get a response.

Configuration

Field	Type	Notes
peerId	select, required	—
prompt	long-text, required	—

Outputs

Output	Type	Description
response	long-text	—

Guardrail

Key: guardrail

Validate content against guardrail policies for PII detection, prompt shielding, and moderation.

Configuration

Field	Type	Notes
content	long-text, required	Content or template (supports {{ }}) that will be evaluated against the guardrail policy.
policy	guardrail-policy	Configure PII detection, PromptShield, and moderation checks applied to this step.

Outputs

Output	Type	Description
status	string	Overall evaluation status returned by the guardrail step (passed or blocked).
passed	boolean	True when the content passes all configured guardrail checks without blocking issues.
errors	array	List of guardrail violations detailing detected issues, severities, and recommended actions.

How it behaves

• The pass path is the step's normal nextStepId.
• The blocked path is data.failNextStepId — set it with connect_steps(branch: 'fail'). If it is left empty, blocked content continues down the normal path.

Extract Data

Key: extract-data

Extract structured data from text using an LLM and JSON schema.

Configuration

Field	Type	Notes
modelId	model-select, required	—
prompt	long-text	Optional prompt to guide the extraction. If empty, the entire context will be used.
jsonSchema	json-schema, required	The JSON schema defining the structure of data to extract
media	list	Each entry: type, url.

Outputs

Output	Type	Description
data	object	—

Intent Parser

Key: intent-parser

Analyze text to detect user intent and extract parameters.

Configuration

Field	Type	Notes
modelId	model-select, required	—
prompt	long-text	—
media	list	Each entry: type, url.
intents	list, required	Each entry: name, description.

Outputs

Output	Type	Description
name	string	—

Smart Agent

Key: smart-agent

Run a long-running autonomous agent (Cognipeer Agent SDK) with planning, context summarization and tools. Compose several Smart Agents to accomplish complex work.

Configuration

Field	Type	Notes
modelId	model-select	If left empty, the default workspace agent model will be used.
systemPrompt	long-text	Instructions guiding the agent's behavior. Supports {{template}} variables. Default "You are a Cognipeer Smart Agent. Plan your work, use the available tools, and produce a thorough, correct result.".
userMessage	long-text, required	The task/instructions for the agent to accomplish. Supports {{template}} variables.
planningMode	select	How the agent plans its work. Todo mode maintains a live checklist of steps. Default "todo". One of todo, planner_executor, reasoning_then_tools, off.
enableSummarization	checkbox	Automatically compact long context/tool outputs so the agent can run for a long time without exceeding the token window. Default true.
maxToolCalls	number	Safety limit on the total number of tool calls the agent may make. Default 30.
toolApprovalMode	select	Pause the whole flow before the agent runs a gated tool, and wait for a human decision. In a Task the pause shows up as 'Waiting approval' with Approve/Reject; the flow continues from where it stopped. Default "off". One of off, all, selected.
approvalTools	text	Comma-separated tool names, e.g. sandbox_exec, send_email. Names must match the tool names the agent sees. Shown when toolApprovalMode is selected.
enableAskUser	checkbox	Gives the agent an ask_user tool so it can stop and ask instead of guessing. The flow pauses the same way an approval does, and resumes with the answer. Default false.
actions	multiple-select	Tools/actions the agent can use to complete the task.
enableWebSearch	checkbox	Give the agent a web_search tool so it can research online. Requires a Tavily API key on the server; without one the agent simply gets no search tool. Default false.
connectedApps	multiple-select	Other flows the agent may invoke as tools (each becomes an execute_<flow> tool).
connectedPeers	multiple-select	Peers the agent may delegate to as sub-agents (each becomes an ask_<peer> tool).
output_type	select	Choose how the agent should format its final response. Default "text". One of text, structured_output.
jsonSchema	json-schema	Define the structure of the JSON output when using Structured Output. Shown when output_type is structured_output.
skillMode	select	Give the agent access to workspace skills (Settings → Skills). Each skill is disclosed as a one-line header; its full prompt and tools load only when the agent opens it. Default "off". One of off, all, selected.
skillIds	skill-select	Skills this agent may open. Shown when skillMode is selected.
enableMemoryTools	checkbox	Let the agent read, search, save and delete Flow Project memory on its own via tool calls. Default false.
memoryScope	select	Default "project". One of project, flow, custom. Shown when enableMemoryTools is true.
memoryFlowId	text	Only used when Memory Scope is Flow. Leave as default for the current flow. Default "{{flow.id}}". Shown when memoryScope is flow.
memoryCustomKey	text	Only used when Memory Scope is Custom — any identifier to bucket memory by, e.g. {{trigger.userId}}. Shown when memoryScope is custom.
memoryOperations	multiple-select	Default ["read","search","put","delete"]. One of read, search, put, delete. Shown when enableMemoryTools is true.
enableSandboxTools	checkbox	Let the agent run shell commands, code, and file operations inside a sandbox via tool calls. Default false.
sandboxRef	text	Sandbox Record ID from the project's Sandbox tab (pick one already created), or a template expression resolved at runtime (e.g. from a Create Sandbox node's output). Shown when enableSandboxTools is true.
sandboxOperations	multiple-select	Default ["exec","run_code","list_files","read_file","write_file","mkdir","delete","move"]. One of exec, run_code, list_files, read_file, write_file, mkdir, delete, move. Shown when enableSandboxTools is true.
injectSecretNames	checkbox	Append the names of this Flow Project's secrets to the system prompt so the agent knows which credentials are available as environment variables. Values are never included. Default true.

Outputs

Output	Type	Description
response	long-text	—
data	object	Parsed JSON data when using structured output.
plan	object	The agent's final plan / todo list with step statuses.
toolCalls	array	Tools the agent invoked during the run.
summaries	array	Summarization events emitted while compacting context.
usage	object	Token usage reported for the run.
events	array	Ordered log of plan/tool/summarization events for detailed inspection.

How it behaves

• userMessage and systemPrompt are rendered as templates, so this is where you feed the agent its input: {{detail}} from a task trigger, {{research.response}} from an earlier agent.
• Reading the result: free text is at <output>.response. For anything a later step must branch on or iterate over, set output_type to "structured_output" with a jsonSchema — the parsed object is then at <output>.data.<field>, e.g. {{outline.data.sections}}.
• An agent with no actions has NO tools: it can only reason over what you put in its prompt. It cannot search the web, read a URL or call an API unless you give it actions. Call list_workspace_resources to see which tool-actions this workspace actually has, and pass them by id or name.
• enableWebSearch gives the agent a web_search tool without needing a tool-action, but it only works when the server has a Tavily API key configured; without one the agent gets no search tool.
• maxToolCalls (default 30) is a hard cap — a research agent that must read many sources needs it raised.
• toolApprovalMode/enableAskUser pause the whole flow waiting for a human. Only use them when the user asked for a human in the loop.

Workflow Control

Final

Key: final

Final step that contains outputs for the flow.

Configuration

Field	Type	Notes
outputType	select	Choose how the final output should be structured. Default "structured". One of structured, message, text.

How it behaves

• A final step ends its execution path — it has no nextStepId.
• Every path through the flow must reach a final step.
• Each entry's output is a template string rendered against the run context, and name becomes the key it is published under. For a markdown deliverable use type 'markdown' and put the whole document in output, e.g. "{{article}}".
• A final step declares the run's RESULT. It does not create a downloadable file — use publish-file for that.

Condition

Key: condition

Condition step to control the app.

How it behaves

• rule is evaluated as a raw JavaScript expression with the run context's variables in scope — use bare variable names (riskScore > 7), NOT {{template}} syntax.
• Branches are tried top to bottom; the first rule that evaluates truthy wins.
• If no rule matches, execution falls through to the condition step's own nextStepId — always wire that as the default/else path.
• Set each branch's targetStepId with connect_steps(branch: 'condition', conditionKey: <branch id>), not by editing targetStepId directly.
• A rule that names a variable which does not exist yet throws a ReferenceError and FAILS the run. Any variable a rule tests must be created by an earlier step (define-variable), or guarded with typeof.
• A branch may target a step EARLIER in the flow — that backwards edge is how a revise-until-approved loop is built. Always bound it in the rule (e.g. approved !== true && revision < 2); a run is aborted after ~500 step executions.

Classifier

Key: classifier

Use a model to classify input into one of multiple branches.

Configuration

Field	Type	Notes
modelId	model-select, required	—
prompt	long-text, required	Describe what the model should classify. Template variables are supported.

Outputs

Output	Type	Description
key	string	—
reason	string	—

How it behaves

• The model picks one category by matching its description, so descriptions must be mutually exclusive.
• Wire each category with connect_steps(branch: 'classifier', conditionKey: <category id>).
• If the model returns no category, execution falls through to the step's own nextStepId.

User Approval

Key: user-approval

Pause execution and request user approval before continuing. Execution resumes based on user's decision.

Configuration

Field	Type	Notes
message	long-text, required	Message to display when requesting approval from the user

Outputs

Output	Type	Description
decision	string	The user's decision (approve or reject)
decidedAt	date	When the decision was made
decidedBy	string	User who made the decision

How it behaves

• The approve path is the step's normal nextStepId.
• The reject path is the top-level rejectNextStepId field — set it with connect_steps(branch: 'reject'). If it is left empty, rejecting continues down the SAME path as approving.

Client Tool

Key: client-tool

Pause execution and request client-side tool execution. The client executes the tool locally and provides the result to continue the flow.

Configuration

Field	Type	Notes
toolName	text, required	Name of the client-side tool to execute
parameters	list	Parameters to pass to the tool. Each parameter supports template variables. Each entry: key, value.

Outputs

Output	Type	Description
result	any	The output returned by the client after executing the tool
error	boolean	Whether the tool execution failed
executionId	string	Unique identifier for this tool execution

Each

Key: each

Iterate over an array and execute nested steps for every item.

Configuration

Field	Type	Notes
arrayVariable	select, required	Select the array to iterate over
maxItems	number	Optional safety limit for processed items (leave empty for all)

Outputs

Output	Type	Description
item	array	All items that were processed during the loop
results	array	Detailed output for each iteration
errors	array	Errors that occurred during processing
successfulItems	number	Number of iterations completed without error

How it behaves

• The loop body lives in data.steps[]. Add steps to it with add_step(parentStepId: <this step id>), never by writing data.steps directly.
• Inside a loop body steps run in array order when nextStepId is empty, so body steps do not each need an explicit connection.
• arrayVariable is a BARE context path, not a template — write "outline.data.sections", never "{{outline.data.sections}}". If the path does not resolve to an array the step does nothing and reports an error in its result instead of failing the run.
• item and index are available as context variables inside the body.
• A body step's output variable is overwritten on every iteration, so after the loop it holds the LAST iteration only. To keep every iteration's work, create a variable with define-variable BEFORE the loop and grow it inside the body — set-variable for a concatenated string, add-item-to-array for a list.

While

Key: while

Repeat nested steps while the condition evaluates to true.

Configuration

Field	Type	Notes
condition	long-text, required	JavaScript expression evaluated against the current context (e.g. score > 0)
maxIterations	number	Safety limit to prevent infinite loops (default 200)

Outputs

Output	Type	Description
results	array	Outputs returned from each loop run
iterations	number	Number of completed iterations
lastIteration	object	Outputs from the last iteration that ran
errors	array	Errors captured while evaluating the loop

How it behaves

• condition is a raw JavaScript expression over context variables, same as a condition step's rule — bare names, no {{templates}}.
• The loop body lives in data.steps[] — add to it with add_step(parentStepId: <this step id>).
• Every variable the condition tests MUST already exist when the step is first reached. A condition naming an undefined variable throws, and the step swallows the error and runs ZERO iterations — a silent no-op that looks like a working flow. Initialise the loop state with define-variable steps beforehand.
• The body must change something the condition tests, and maxIterations (default 200) is the only other backstop.

End Session

Key: end-session

End the current conversation session. When triggered, the active session is marked as ended. A new session will automatically start when the user sends a new message.

Configuration

Field	Type	Notes
reason	text	The reason for ending the session (e.g., 'flow_completed', 'goal_achieved', 'task_finished'). Defaults to 'flow_completed' if not specified.

Outputs

Output	Type	Description
success	boolean	Whether the session was successfully ended
sessionId	string	ID of the ended session
endReason	string	The reason the session was ended
endedAt	date	Timestamp when the session was ended
error	string	Error message if the session could not be ended

Subflow

Key: subflow

Run another flow as a child and use its outputs. Enables reusable, composable flows that call each other.

Configuration

Field	Type	Notes
appId	select, required	The flow to run as a subflow.
version	select	Which version of the target flow to execute. Default "latest". One of latest, draft.
inputs	list	Values mapped into the subflow's trigger inputs. Each value supports {{template}} variables from the current flow. Each entry: key, value.

Outputs

Output	Type	Description
output	object	The final outputs produced by the subflow.
response	long-text	Stringified subflow output, convenient for prompts.
appRunId	string	The subflow's execution run id.
status	string	—

How it behaves

• inputs is a list of { key, value } pairs; value supports {{template}} variables from the current flow.

User Interaction

Show Form

Key: show-form

Display a form to collect user input. Supports both predefined forms (from form library) and inline form definitions with cascading/conditional fields.

Configuration

Field	Type	Notes
formId	select	Select a predefined form from your form library. Leave empty to use inline form definition.
inlineForm	object	Define form fields directly in the step. Use this for flow-specific forms or cascading scenarios.
title	text	Override the form title (optional).
description	text	Override the form description (optional).
waitForSubmission	boolean	If true, the flow will pause until the form is submitted. Default true.

Outputs

Output	Type	Description
messageWidgetId	string	ID of the created form widget instance
formId	string	ID of the form (null for inline forms)
formTitle	string	Title of the displayed form
waitingForSubmission	boolean	Whether the flow is waiting for user to submit the form

Data & Databases

Search In Knowledgebase

Key: search-in-knowledgebase

Semantic search in the knowledgebase.

Configuration

Field	Type	Notes
datasourceIds	multiple-select, required	—
prompt	long-text, required	—
resultsCount	number	—
saveSourcesToMessage	checkbox	—

Outputs

Output	Type	Description
results	array	—

How it behaves

• datasourceIds are workspace datasource ids — get them from list_workspace_resources. Do not invent them.

Web & Internet

Http Request

Key: http-request

Send a http request and get a response.

Configuration

Field	Type	Notes
url	text, required	—
method	select, required	One of GET, POST, PUT, PATCH, DELETE.
headers	long-text	—
body	code	—

Outputs

Output	Type	Description
response	object	—

Scrape Web Page

Key: scrape-web-page

Scrape web page using puppeteer.

Configuration

Field	Type	Notes
url	long-text, required	—

Outputs

Output	Type	Description
url	text	—
title	text	—
description	long-text	—
body	long-text	—

Variable Manipulation

Define Variable

Key: define-variable

Define a new variable with a specific value.

Configuration

Field	Type	Notes
name	text, required	Name of the variable to define
type	select, required	Type of the variable to define One of string, number, boolean, array, object.
value	long-text	Initial value for the variable

Outputs

Output	Type	Description
value	text	The value of the defined variable

How it behaves

• name is the bare context key the variable is created under — read it back as {{name}}, and reference it in condition rules as a bare identifier. Do NOT also set the step's output field for this purpose.
• value is rendered as a template, so a variable can be initialised from earlier results.
• Use this to initialise loop state (counters, accumulators) BEFORE any condition/while rule tests it — an undefined variable in a rule is a ReferenceError.
• For an accumulator, type 'string' with an empty value (concatenate with set-variable) or type 'array' with an empty value (append with add-item-to-array).

Set Variable

Key: set-variable

Set an existing variable to a new value.

Configuration

Field	Type	Notes
name	select, required	Name of the variable to set
value	long-text, required	New value for the variable

Outputs

Output	Type	Description
value	text	The new value of the variable

How it behaves

• value is rendered as a template, which is what makes accumulation work: value "{{article}}\n\n{{sectionDraft.response}}" appends to the variable named article.
• The variable must already exist (create it with define-variable); the new value is coerced to the existing variable's type.

Set State

Key: set-state

Set a state variable that persists across flow executions within the same conversation. State is stored in context.state and in the conversation document.

Configuration

Field	Type	Notes
name	text, required	Name of the state variable to set (accessible via state.variableName)
value	long-text, required	Value to store in the state variable

Outputs

Output	Type	Description
value	text	The value that was set in the state

Add Item to Array

Key: add-item-to-array

Add an item to an array variable.

Configuration

Field	Type	Notes
arrayName	select, required	Name of the array variable
item	list, required	Item to add to the array Each entry: name, type, value.

Outputs

Output	Type	Description
array	array	The array after adding the item
length	number	The new length of the array

How it behaves

• arrayName is a BARE context variable name, not a template.
• item is a LIST of field descriptors — [{ name, type, value }] — and each value is rendered as a template. The step pushes one object built from those fields, so item [{name:"content",type:"string",value:"{{draft.response}}"}] appends { content: "..." }.
• The array is created if it does not exist, but initialising it with define-variable first is clearer and is required if a condition rule tests it.

Add Items to Array

Key: add-items-to-array

Add multiple items from context to an existing array variable.

Configuration

Field	Type	Notes
arrayVariableName	select, required	Name of the array variable to add items to
itemsKey	text, required	Key from context containing items to add to the array

Outputs

Output	Type	Description
array	array	The array after adding the items
addedCount	number	Number of items added to the array
totalLength	number	The new total length of the array
success	boolean	Whether the operation was successful

Increment Variable

Key: increment-variable

Increment a numeric variable by a specific amount.

Configuration

Field	Type	Notes
variableName	select, required	Name of the numeric variable to increment
amount	number	Amount to increment by (default: 1) Default 1.

Outputs

Output	Type	Description
value	number	The variable after incrementing

Decrement Variable

Key: decrement-variable

Decrement a numeric variable by a specific amount.

Configuration

Field	Type	Notes
variableName	select, required	Name of the numeric variable to decrement
amount	number	Amount to decrement by (default: 1) Default 1.

Outputs

Output	Type	Description
value	number	The variable after decrementing

Memory Management

Save User Memory

Key: save-user-memory

Save information to user memory for later retrieval.

Configuration

Field	Type	Notes
peerId	text, required	ID of the peer to save memory for
key	text, required	Unique key to identify this memory
value	long-text, required	Value to store in memory
description	text, required	Description for this memory (required for better organization)
metadata	long-text	Additional metadata as JSON object

Outputs

Output	Type	Description
success	boolean	Whether the memory was saved successfully
memoryId	text	ID of the saved memory record
key	text	The key used to save the memory
saved	boolean	Whether the memory was saved

Search User Memory

Key: search-user-memory

Search for user memories using text query across keys, values, and descriptions.

Configuration

Field	Type	Notes
peerId	text, required	ID of the peer to search memory for
query	text	Text to search for in memory (leave empty to get all records)
limit	number	Maximum number of results (default: 10)

Outputs

Output	Type	Description
success	boolean	Whether the search was successful
found	boolean	Whether any memories were found
count	number	Number of memories found
results	array	Array of found memories

Delete User Memory

Key: delete-user-memory

Delete a specific user memory record using its key.

Configuration

Field	Type	Notes
peerId	text, required	ID of the peer to delete memory from
key	text, required	Key of the memory to delete
confirmDelete	boolean, required	Safety confirmation for deletion (must be true)

Outputs

Output	Type	Description
success	boolean	Whether the deletion was successful
deleted	boolean	Whether the memory was deleted
found	boolean	Whether the memory was found before deletion
key	text	The key that was deleted
deletedValue	text	The value that was deleted

Memory: Read

Key: flow-memory-read

Read entries from Flow Project memory (Project, Flow, or Custom scope), newest/oldest first.

Configuration

Field	Type	Notes
flowProjectId	text	Leave as default to use the current flow's project. Default "{{flow.flowProjectId}}".
scope	select, required	Default "project". One of project, flow, custom.
flowId	text	Only used when Scope is Flow. Leave as default for the current flow. Default "{{flow.id}}". Shown when scope is flow.
customKey	text	Only used when Scope is Custom — any identifier you want to bucket memory by. Shown when scope is custom.
limit	number	Maximum number of items to return. Default 20.
order	select	Default "newest". One of newest, oldest.

Outputs

Output	Type	Description
items	array	—
count	number	—
memoryStoreId	text	—

Memory: Search

Key: flow-memory-search

Search Flow Project memory (Project, Flow, or Custom scope) by text across key and value.

Configuration

Field	Type	Notes
flowProjectId	text	Leave as default to use the current flow's project. Default "{{flow.flowProjectId}}".
scope	select, required	Default "project". One of project, flow, custom.
flowId	text	Only used when Scope is Flow. Leave as default for the current flow. Default "{{flow.id}}". Shown when scope is flow.
customKey	text	Only used when Scope is Custom — any identifier you want to bucket memory by. Shown when scope is custom.
query	text	Text to search for across keys and values (leave empty to get all records).
limit	number	Default 20.

Outputs

Output	Type	Description
items	array	—
count	number	—
found	boolean	—
memoryStoreId	text	—

Memory: Put

Key: flow-memory-put

Save (create or update) a key/value entry in Flow Project memory (Project, Flow, or Custom scope).

Configuration

Field	Type	Notes
flowProjectId	text	Leave as default to use the current flow's project. Default "{{flow.flowProjectId}}".
scope	select, required	Default "project". One of project, flow, custom.
flowId	text	Only used when Scope is Flow. Leave as default for the current flow. Default "{{flow.id}}". Shown when scope is flow.
customKey	text	Only used when Scope is Custom — any identifier you want to bucket memory by. Created automatically on first write. Shown when scope is custom.
key	text, required	Unique key to identify this memory entry within its scope.
value	long-text, required	Value to store.

Outputs

Output	Type	Description
success	boolean	—
memoryStoreId	text	—
key	text	—

Memory: Delete

Key: flow-memory-delete

Delete a specific entry from Flow Project memory using its key.

Configuration

Field	Type	Notes
flowProjectId	text	Leave as default to use the current flow's project. Default "{{flow.flowProjectId}}".
scope	select, required	Default "project". One of project, flow, custom.
flowId	text	Only used when Scope is Flow. Leave as default for the current flow. Default "{{flow.id}}". Shown when scope is flow.
customKey	text	Only used when Scope is Custom — any identifier you want to bucket memory by. Shown when scope is custom.
key	text, required	Key of the memory entry to delete.
confirmDelete	boolean, required	Safety confirmation for deletion (must be true).

Outputs

Output	Type	Description
success	boolean	—
deleted	boolean	—
found	boolean	—
key	text	—
deletedValue	text	—

Sandbox

Sandbox: Create

Key: flow-sandbox-create

Create a new persistent sandbox and add it to the Flow Project's Sandbox list.

Configuration

Field	Type	Notes
flowProjectId	text	Leave as default to use the current flow's project. Default "{{flow.flowProjectId}}".
label	text	A friendly name shown in the project's Sandbox tab.
template	text	Leave empty to use the workspace default sandbox template.
secretMode	select	Which of the Flow Project's secrets are set as environment variables inside this sandbox. Default "all". One of all, selected, none.
secretKeys	text	Comma-separated secret names to inject, e.g. GITHUB_TOKEN, STRIPE_KEY. Only used when Project Secrets is set to "selected". Shown when secretMode is selected.
env	code	Additional non-secret environment variables as a JSON object, e.g. {"NODE_ENV": "production"}. These override project secrets on a name clash.

Outputs

Output	Type	Description
sandboxRecordId	text	—
sandboxId	text	—
status	text	—

Sandbox: Run

Key: flow-sandbox-run

Run an operation inside an existing sandbox — shell command, code execution, or a file operation.

Configuration

Field	Type	Notes
sandboxRef	text, required	Sandbox Record ID from the project's Sandbox tab (or a Create Sandbox node's output), or a template expression resolved at runtime.
operation	select, required	Default "exec". One of exec, run_code, list_files, read_file, write_file, mkdir, delete, move.
command	long-text	Shown when operation is exec.
timeoutSec	number	Default 120. Shown when operation is exec.
language	select	Default "python". One of python, javascript. Shown when operation is run_code.
code	long-text	Shown when operation is run_code.
path	text	Sandbox-relative path, e.g. /workspace/data.csv Shown when operation is list_files / read_file / write_file / mkdir / delete / move.
destination	text	Shown when operation is move.
content	long-text	Shown when operation is write_file.
encoding	select	Default "utf8". One of utf8, base64. Shown when operation is read_file / write_file.
recursive	checkbox	Default true. Shown when operation is delete.

Outputs

Output	Type	Description
success	boolean	—
stdout	long-text	—
stderr	long-text	—
exitCode	number	—
items	array	—
content	long-text	—

Sandbox: Start

Key: flow-sandbox-start

Start a stopped sandbox.

Configuration

Field	Type	Notes
sandboxRef	text, required	Sandbox Record ID from the project's Sandbox tab, or a template expression.

Outputs

Output	Type	Description
status	text	—

Sandbox: Wait Until Running

Key: flow-sandbox-wait-until-running

Pass through immediately if the sandbox is already running; otherwise poll every 500ms until it is.

Configuration

Field	Type	Notes
sandboxRef	text, required	Sandbox Record ID from the project's Sandbox tab, or a template expression.
timeoutSec	number	Give up and fail the step if the sandbox hasn't reached running within this time. Default 120.

Outputs

Output	Type	Description
status	text	—
success	boolean	—

Sandbox: Stop

Key: flow-sandbox-stop

Stop (suspend) a running sandbox. Data is kept.

Configuration

Field	Type	Notes
sandboxRef	text, required	Sandbox Record ID from the project's Sandbox tab, or a template expression.

Outputs

Output	Type	Description
status	text	—

Sandbox: Delete

Key: flow-sandbox-delete

Permanently delete a sandbox and remove it from the project's Sandbox list.

Configuration

Field	Type	Notes
sandboxRef	text, required	Sandbox Record ID from the project's Sandbox tab, or a template expression.
confirmDelete	boolean, required	Safety confirmation for deletion (must be true).

Outputs

Output	Type	Description
success	boolean	—
deleted	boolean	—

Files

Publish File

Key: publish-file

Publish a file as a downloadable output of this run. Takes a file already inside a sandbox (by path) or base64 content, stores it, and attaches it to the run — where the Task detail view lists it under Output files.

Configuration

Field	Type	Notes
source	select, required	Default "sandbox". One of sandbox, text, base64.
text	long-text	Text to publish as a file, e.g. {{article}} for a markdown document the flow assembled. Set File Name with a matching extension. Shown when source is text.
sandboxRef	text	Sandbox Record ID from the project's Sandbox tab, or a template expression. Shown when source is sandbox.
path	text	Absolute path inside the sandbox, e.g. /workspace/report.pdf Shown when source is sandbox.
base64	text	Template expression resolving to base64 content, e.g. {{steps.x.data}} Shown when source is base64.
fileName	text	Name shown to the user. Defaults to the sandbox file's own name.

Outputs

Output	Type	Description
success	boolean	—
key	text	—
fileName	text	—
downloadPath	text	—
downloadLink	text	—

How it behaves

• This is the only step that attaches a real file to the run — it is what appears under a Task's "Output files" for download. save-text-file merely returns a temporary link.
• source 'text' is the simplest form: put the rendered document in text (e.g. "{{article}}") and set a fileName with the right extension (report.md). source 'sandbox' reads a path from a sandbox; source 'base64' takes already-encoded content.

Reporting & Analytics

Track Event

Key: track-event

Record a structured, durable business event for reporting (e.g. 'this email was routed to support'). Unlike debug logs, this always persists regardless of the flow's debug setting, and is meant to be queried/aggregated later (counts, breakdowns by dimension, trends). Use this after a decision/classification/output point whenever the flow author wants durable domain reporting, not raw technical tracing.

Configuration

Field	Type	Notes
eventName	text, required	Stable identifier for this kind of event, e.g. 'email_processed'.
dimensions	list	Categorical labels to group/filter by later (e.g. category, destination). Each entry: key, value.
measures	list	Numeric values to count/sum later (e.g. count, amount). Each entry: key, value.

Outputs

Output	Type	Description
success	boolean	—
eventId	text	ID of the persisted event record.

Utilities

Execute NodeJS

Key: execute-nodejs

Execute NodeJS code.

Configuration

Field	Type	Notes
code	code	—

Outputs

Output	Type	Description
response	object	—

Check Regex

Key: check-regex

Check if a string matches a regular expression pattern.

Configuration

Field	Type	Notes
input	long-text, required	The string to check against the regex pattern
pattern	text, required	Regular expression pattern to test
flags	text	Optional regex flags (e.g., 'g', 'i', 'm')

Outputs

Output	Type	Description
isMatch	boolean	Whether the input matches the pattern
matches	array	Array of matched substrings

Read File

Key: read-file

Read File from url or file key in Cognipeer with context.

Configuration

Field	Type	Notes
fileKey	select, required	—

Outputs

Output	Type	Description
base64	text	—

Read File From URL

Key: read-file-from-url

Read File from url

Configuration

Field	Type	Notes
url	text	—

Outputs

Output	Type	Description
base64	text	—

Convert To Markdown

Key: convert-to-markdown

Convert base64 content to markdown.

Configuration

Field	Type	Notes
key	select, required	—

Outputs

Output	Type	Description
markdown	long-text	—

Markdown To PDF

Key: markdown-to-pdf

Convert markdown content to a PDF file and return it as base64.

Configuration

Field	Type	Notes
markdownKey	select, required	Key from context containing markdown content to convert
fileName	text	Optional file name to expose in outputs (defaults to document.pdf)
title	text	Optional document title used in PDF metadata

Outputs

Output	Type	Description
base64	text	PDF file content as base64 string
mimeType	text	MIME type of the generated PDF
fileName	text	Suggested PDF file name
size	number	Size of the generated PDF in bytes

Chunk Content

Key: chunk-content

Split content into chunks based on specified parameters.

Configuration

Field	Type	Notes
content	long-text, required	The content to be chunked
chunkType	select, required	The type of chunking method to use One of sentenceSplitter, tokenizer, recursiveCharacterWithSpecialSeparators, paragraphSplitter, regexSplitter.
chunkSize	number	Maximum size of each chunk (for recursive character and paragraph splitters)
chunkOverlap	number	Number of characters to overlap between chunks
maxTokens	number	Maximum number of tokens per chunk (for tokenizer)
maxChunkSize	number	Maximum size of each chunk (for paragraph and regex splitters)
separators	long-text	JSON array of separators for recursive character splitter (e.g., ["\n\n", "\n", " "])
regexPattern	text	Regular expression pattern for regex splitter

Outputs

Output	Type	Description
chunks	array	Array of chunked content strings
chunkCount	number	Total number of chunks generated

JSON to Excel

Key: json-to-excel

Convert JSON data to Excel file and return as base64.

Configuration

Field	Type	Notes
jsonDataKey	select, required	Key from context containing JSON data to convert to Excel
worksheetName	text	Name for the worksheet (default: Sheet1)

Outputs

Output	Type	Description
base64	text	Excel file content as base64 string
fileName	text	Generated file name
mimeType	text	MIME type of the Excel file
size	number	Size of the file in bytes

Save File

Key: save-file

Save base64 content as a file and generate download link.

Configuration

Field	Type	Notes
base64Key	select, required	Key from context containing base64 encoded file content to save
fileName	text, required	Name for the file including extension
expiryMinutes	number	How long the download link should be valid (default: 1440 minutes / 24 hours)

Outputs

Output	Type	Description
success	boolean	Whether the file was saved successfully
fileKey	text	Unique key for the saved file
fileName	text	Name of the saved file
downloadLink	text	Temporary download link for the file
expiryMinutes	number	How long the download link is valid
message	text	Status message
error	text	Error message if operation failed

Save Text File

Key: save-text-file

Save plain text content as a file and generate a direct download link.

Configuration

Field	Type	Notes
textKey	select	Key in context that holds the text content. If not set, uses the inline Text field.
text	long-text	Inline text content to save. If both textKey and text are provided, inline text is used.
fileName	text, required	Name for the file including extension (e.g., notes.txt)
expiryMinutes	number	How long the download link should be valid (default: 1440 minutes / 24 hours)

Outputs

Output	Type	Description
success	boolean	Whether the file was saved successfully
fileKey	text	Unique key for the saved file
fileName	text	Name of the saved file
downloadLink	text	Temporary download link for the file
expiryMinutes	number	How long the download link is valid
message	text	Status message
error	text	Error message if operation failed

How it behaves

• textKey is a BARE context path (e.g. "article"), while text is a template string. If both are set, text wins.
• This does NOT attach the file to the run — use publish-file when the user should be able to download the result from the task.

String to JSON

Key: string-to-json

Parse a string and convert it to JSON object.

Configuration

Field	Type	Notes
stringValue	long-text, required	The string content to parse as JSON

Outputs

Output	Type	Description
jsonObject	object	Parsed JSON object from the string
success	boolean	Whether the parsing was successful
error	text	Error message if parsing failed

JSON to String

Key: json-to-string

Convert JSON object to string representation.

Configuration

Field	Type	Notes
jsonKey	select, required	Key from context containing JSON data to convert to string
prettyFormat	boolean	Format the JSON string with indentation for readability

Outputs

Output	Type	Description
stringValue	text	JSON object converted to string format
success	boolean	Whether the conversion was successful
error	text	Error message if conversion failed

Widgets

Add Widget to Message

Key: add-widget-to-message

Attach a widget to the assistant's response message and prefill its initial state.

Outputs

Output	Type	Description
messageWidgetId	string	Identifier of the created message widget record
widgetId	string	Identifier of the attached widget
widgetName	string	Name of the attached widget
initialState	object	Initial state passed to the widget