Boomi Professional cheat sheet

Process shapes: Start, Data Passthrough, Map, Decision, Branch, Route, Try/Catch, Notify, Stop, Return Documents, Message, Process Call, Process Route, Cleanse, Combine, Split, Find Changes, No Data, Set Properties, Dynamic Process Property, Document Cache, Flow Control, Business Rules, Custom Scripting.

Process Call runs a child process synchronously and returns documents to the parent; Process Route invokes a process asynchronously and returns no documents to the caller.

Try/Catch encloses shapes; on error the Catch path runs, and Try/Catch properties can rethrow, stop, or continue based on configured behavior.

Decision shape evaluates one or more conditions in order; the first true branch is taken. Route shape evaluates all configured routes and can send a single document to multiple branches.

Branch shape routes a document to exactly one path based on the first matching condition; subsequent conditions are not evaluated.

Combine concatenates documents into one document; Split divides a single document into multiple documents. The Combine operation is the inverse of Split.

Flow Control limits documents per execution and can set a maximum document count; documents beyond the limit are held or failed per configuration.

A process property is scoped to a single process execution; a dynamic process property is persisted at the account level and survives across executions.

Scheduling options include interval, daily, weekly, monthly, and custom cron; a listener process (such as a trading partner or connector listener) starts on a document/event arrival instead of a schedule.

A profile is a JSON-schema-like definition of a document structure; profiles are the source and target of maps and are used to validate documents.

A map transforms a source profile to a target profile; mappings can be direct, use functions, or use user-defined functions (UDFs) written in JavaScript or Groovy.

Map functions include string (concat, substring, replace), numeric (add, multiply, round), date (format, parse, add days), lookup (database, cross-reference, document cache), and logic (if, choose, equals).

A cross-reference table maps a source value to a target value (a lookup table); it is maintained as a separate component and referenced from a map.

A document cache stores key/value pairs for the duration of a process execution and is commonly used to avoid repeated lookups within a loop.

Standard profile formats include flat file (delimited, fixed-width), XML, JSON, and EDI; flat-file profiles require configured delimiters or column positions and a record/segment structure.

Cleanse shape standardizes data (for example, trimming whitespace, normalizing case, and applying replacement rules) before mapping or persistence.

Find Changes compares a new document against a stored version and emits only changed documents; it requires a key field and a stored baseline.

Profile validation can be enforced on a connector operation so malformed documents fail before mapping, rather than producing silent nulls in the target.

An API component exposes a process (or a set of processes) as a REST endpoint; each API has one or more resources, each with a path, HTTP method, and an underlying process.

API authentication options include API key, OAuth 2.0 (client credentials and authorization code), basic authentication, and JWT; the chosen scheme is configured on the API component.

An API gateway enforces policies at the edge: rate limiting, quota, throttling, authentication, CORS, and request/response transformation.

Rate limiting restricts requests per time window per client; quota restricts total requests over a longer period (for example, per day or per month).

API versions are managed by publishing a new version and deprecating the old one; consumers are pointed at a specific version path.

The API developer portal publishes documentation and lets consumers register applications and obtain credentials for API access.

A private API is reachable only from within the account's network or via a configured gateway; a public API is reachable from the internet.

HTTP status conventions returned by API resources: 200 success, 201 created, 400 bad request, 401 unauthenticated, 403 forbidden, 404 not found, 429 rate limited, 500 server error.

Boomi AI features are configured at the account level; model connections (for example, to an LLM provider) are stored as reusable connector or service configurations.

An AI-powered mapping suggestion proposes source-to-target field mappings from profile names, descriptions, and sample data; proposed mappings require human review before activation.

Boomi Suggest (AI-assisted process design) generates process steps from a natural-language description; the generated process is editable and must be validated before deployment.

Document Intelligence extracts structured data from unstructured documents (such as PDFs and invoices) using configured extraction models, then feeds the result into a profile for mapping.

A vector store or embedding index is used for retrieval-augmented generation; documents are chunked, embedded, and retrieved by similarity at query time.

Prompt components and LLM connector operations are invoked from a process like any other connector operation; the response is parsed into a document for downstream mapping.

Data sent to an external LLM provider leaves the Boomi runtime; data residency, retention, and PII masking must be assessed before enabling the operation.

Guardrails for AI features include input validation, output schema validation, and fallback paths when the model call fails or times out.

An agent is an LLM-driven component that selects and invokes tools (processes, APIs, or connector operations) to accomplish a goal, rather than following a fixed process path.

A tool is a callable capability exposed to an agent, typically a process or API with a defined input and output schema; the agent chooses which tool to call and with what arguments.

Agent reasoning loops iterate: observe, decide, act (call tool), and observe the result; a maximum iteration or timeout bound prevents runaway loops.

Agent memory can be short-term (the current conversation or execution context) or long-term (a persisted store such as a vector database or document store).

Multi-agent patterns include orchestrator/worker (a planner delegates to specialist agents) and handoff (control transfers from one agent to another).

Human-in-the-loop checkpoints pause an agent before a high-impact action (for example, a write or a payment) and require approval to continue.

Agent observability requires logging each tool call, its arguments, its result, and the token/latency cost so that failures can be traced and replayed.

Guardrails for agents include allow-listed tools, scoped credentials per tool, input/output schema validation, and deterministic fallback when the model is unavailable.

Process Call is synchronous and returns documents to the caller; Process Route is asynchronous and returns nothing to the caller.

Decision evaluates conditions in order and takes the first true branch; Route can send one document to multiple matching branches.

Try/Catch encloses shapes and defines the error path; the catch branch can stop the process, continue, or rethrow.

A process property lives for one execution; a dynamic process property persists at the account level across executions.

Combine merges multiple documents into one; Split divides one document into many.

A listener process starts on an event or document arrival, whereas a scheduled process starts on a configured interval or cron.

Flow Control caps the number of documents processed in a single execution.

A connector operation is configured with an action (get, query, create, update, delete, upsert) and an object type, and it produces or consumes a profile-shaped document.

A profile defines document structure and is the source or target of every map; profiles can be flat file, XML, JSON, or EDI.

A map transforms source profile fields to target profile fields, optionally using functions or user-defined functions.

A cross-reference table is a reusable lookup that translates a source value into a target value inside a map.

A document cache holds key/value pairs for the duration of one process execution and reduces repeated lookups.

The Cleanse shape standardizes values (whitespace, case, replacements) before mapping or persistence.

Find Changes compares an incoming document with a stored baseline and emits only the changed documents, keyed on a defined field.

Map functions are grouped as string, numeric, date, lookup, and logic functions; logic functions include if/choose/equals.

Flat-file profiles require configured delimiters or fixed column positions plus a record structure; malformed input fails profile validation.

An API component exposes processes as REST resources; each resource has a path, an HTTP method, and an underlying process.

Authentication schemes for an API include API key, OAuth 2.0, basic authentication, and JWT.

Rate limiting caps requests per time window per client; quota caps total requests over a longer period.

Gateway policies include authentication, rate limiting, quota, CORS, and request/response transformation.

API versions are published and deprecated explicitly; consumers target a specific version path.

A developer portal publishes API documentation and lets consumers register applications and obtain credentials.

A private API is reachable only from within the account's network or a configured gateway; a public API is internet-reachable.

Common API response codes: 200 success, 201 created, 400 bad request, 401 unauthenticated, 403 forbidden, 404 not found, 429 rate limited, 500 server error.

AI model connections are configured at the account level and reused across processes as connector or service configurations.

AI-assisted mapping suggests source-to-target field mappings from profile names, descriptions, and sample data; suggestions require human review before activation.

AI-assisted process generation produces editable process steps from a natural-language description; the result must be validated before deployment.

Document Intelligence extracts structured fields from unstructured documents such as PDFs and invoices and feeds them into a profile for mapping.

Retrieval-augmented generation chunks and embeds documents into a vector store, then retrieves similar chunks at query time to ground the model response.

LLM connector operations are invoked from a process like any other operation; the response is parsed into a document for downstream mapping.

Data sent to an external LLM provider leaves the Boomi runtime; data residency, retention, and PII masking must be assessed first.

AI guardrails include input validation, output schema validation, and a fallback path when the model call fails or times out.

An agent is an LLM-driven component that selects and invokes tools to reach a goal instead of following a fixed process path.

A tool is a callable capability, usually a process or API, with a defined input and output schema that the agent can invoke.

The agent reasoning loop is observe, decide, act, and observe the result; a maximum iteration count or timeout bounds the loop.

Agent memory is short-term (current execution context) or long-term (a persisted store such as a vector database).

Multi-agent patterns include orchestrator/worker, where a planner delegates to specialist agents, and handoff, where control transfers between agents.

A human-in-the-loop checkpoint pauses the agent before a high-impact action and requires approval to continue.

Agent observability logs each tool call, its arguments, its result, and token/latency cost so failures can be traced and replayed.

Agent guardrails include allow-listed tools, per-tool scoped credentials, input/output schema validation, and a deterministic fallback when the model is unavailable.

Process Call runs a child process synchronously and returns documents to the parent; Process Route invokes a process asynchronously and returns no documents.

Try/Catch encloses shapes and defines the error path; the catch branch can stop the process, continue, or rethrow.

Decision takes the first true branch; Route evaluates all routes and can send one document to multiple branches.

A process property is scoped to one execution; a dynamic process property persists at the account level across executions.

Combine merges documents into one; Split divides one document into many.

A listener process starts on an event or document arrival; a scheduled process starts on an interval, daily/weekly/monthly schedule, or cron.

Flow Control caps the number of documents processed in one execution.

A connector operation is configured with an action (get, query, create, update, delete, upsert) and an object type, and it consumes or produces a profile-shaped document.

A profile defines document structure and is the source or target of every map; formats include flat file, XML, JSON, and EDI.

A map transforms source profile fields to target profile fields, optionally using built-in functions or user-defined functions.

A cross-reference table is a reusable lookup that translates a source value to a target value inside a map.

A document cache holds key/value pairs for one process execution and reduces repeated lookups.

The Cleanse shape standardizes values such as whitespace, case, and replacements before mapping or persistence.

Find Changes compares an incoming document with a stored baseline and emits only changed documents, keyed on a defined field.

Map functions are grouped as string, numeric, date, lookup, and logic functions; logic functions include if, choose, and equals.

Flat-file profiles require configured delimiters or fixed column positions plus a record structure; malformed input fails profile validation.

An API component exposes processes as REST resources; each resource has a path, an HTTP method, and an underlying process.

API authentication schemes include API key, OAuth 2.0, basic authentication, and JWT.

Rate limiting caps requests per time window per client; quota caps total requests over a longer period.

Gateway policies include authentication, rate limiting, quota, CORS, and request/response transformation.

API versions are published and deprecated explicitly; consumers target a specific version path.

A developer portal publishes API documentation and lets consumers register applications and obtain credentials.

A private API is reachable only from within the account's network or a configured gateway; a public API is internet-reachable.

Common API response codes: 200 success, 201 created, 400 bad request, 401 unauthenticated, 403 forbidden, 404 not found, 429 rate limited, 500 server error.

AI model connections are configured at the account level and reused across processes as connector or service configurations.

AI-assisted mapping suggests source-to-target field mappings from profile names, descriptions, and sample data; suggestions require human review before activation.

AI-assisted process generation produces editable process steps from a natural-language description; the result must be validated before deployment.

Document Intelligence extracts structured fields from unstructured documents such as PDFs and invoices and feeds them into a profile for mapping.

Retrieval-augmented generation chunks and embeds documents into a vector store, then retrieves similar chunks at query time to ground the model response.

LLM connector operations are invoked from a process like any other operation; the response is parsed into a document for downstream mapping.

Data sent to an external LLM provider leaves the Boomi runtime; data residency, retention, and PII masking must be assessed first.

AI guardrails include input validation, output schema validation, and a fallback path when the model call fails or times out.

An agent is an LLM-driven component that selects and invokes tools to reach a goal instead of following a fixed process path.

A tool is a callable capability, usually a process or API, with a defined input and output schema that the agent can invoke.

The agent reasoning loop is observe, decide, act, and observe the result; a maximum iteration count or timeout bounds the loop.

Agent memory is short-term (current execution context) or long-term (a persisted store such as a vector database).

Multi-agent patterns include orchestrator/worker, where a planner delegates to specialist agents, and handoff, where control transfers between agents.

A human-in-the-loop checkpoint pauses the agent before a high-impact action and requires approval to continue.

Agent observability logs each tool call, its arguments, its result, and token/latency cost so failures can be traced and replayed.

Agent guardrails include allow-listed tools, per-tool scoped credentials, input/output schema validation, and a deterministic fallback when the model is unavailable.