Activity Catalog

An activity is a unit of work a workflow performs outside itself: an HTTP call, a database query, a shell command, an LLM completion. Activities are grouped into providers, and an activity's type is <provider>.<name> — http.request, builtin.state.set_state, llama_index.query.

This page indexes every activity the platform ships. Each provider has its own reference page with the input and output contract of each of its activities, and worked examples.

For what an activity is and how to write one into a workflow, see Activity System and the activity statement.

How to read this reference

Every activity is invoked the same way, so the statement-level fields — type, input_data, config_data, output_name, output_data, retry_policy, execute_locally, async_mode, enable_cache — are documented once in Activity System rather than repeated on each page. The provider pages cover only what is specific to an activity:

• Description — what it does and when to reach for it.
• Input — the fields of input_data, with types, defaults and which are required.
• Output — the shape of the result, which output_name captures and output_data transforms.
• Examples — real YAML, drawn from moco-examples/ wherever an example exists.

Conventions across providers

Secrets are named, not inlined

A field ending _secret_key holds the name of a secret, never the secret itself: apikey_secret_key, connection_string_secret_key, password_secret_key, auth.credentials_secret_key. The activity resolves and decrypts it internally, so plaintext never enters workflow context. A bare NAME resolves a user-scoped secret; global/NAME resolves a global one. See Secret Activities.

Four exceptions are worth knowing:

• http.request takes encrypted_auth_token — the still-encrypted blob from builtin.secret.get — rather than a secret name.
• graphql.subscribe, websocket.subscribe and mcp.call_tool authenticate with a plain headers dict, which passes through workflow context in plaintext.
• langfuse.* takes raw credentials inline; prefer configuring them as environment variables on the worker.
• k8s.* names its token and client key through token_secret_key and client_key_secret_key as usual, but also accepts an inline ca_cert — a CA certificate is public material, so routing it through the secret store buys nothing.

Retry and timeout defaults

The platform default is 60 seconds and 3 attempts. Activities whose work is not idempotent — writes, sends, spend — ship max_attempts: 1 instead, and the exceptions are called out on each page. Override either per call with retry_policy.

Only the Temporal runtime honours retry, timeout and heartbeat settings; the in-memory runtime ignores them.

Long-running activities and the relay pattern

Six activities run for as long as their source keeps producing, and deliver what they receive as workflow events rather than as a return value:

Activity	Delivers
kafka.consume	Each Kafka message
rabbit.receive	Each RabbitMQ message
graphql.subscribe	Each subscription payload
websocket.subscribe	Each inbound frame
mcp.call_tool	Each progress notification
claude_agent.query	Agent progress

They all share the same three fields — relay_topic, relay_event_type and target_workflow_id — and the workflow consumes what they publish with wait_for.

Two things follow. First, start them asynchronously: set async_mode: true or put them in a parallel branch, or they block the workflow for their whole timeout, which is usually a day. Only rabbit.receive defaults to async mode. Second, on the Temporal runtime each relayed payload is a workflow signal and a history entry, and the engine retains at most 1000 unmatched events per topic — so these suit low-rate control streams, not high-throughput data feeds.

k8s.wait also runs long and heartbeats, but is not part of this pattern: it polls until its condition holds and then returns normally, rather than relaying events. Its result is the point, so it blocks by design.

Where activities run

Every activity runs on the base worker (task queue default) except claude_agent.query, which runs on the agent worker (task queue agent). Routing is automatic; nothing in the workflowspec changes. An activity served by a different worker type cannot run locally, so execute_locally has no effect on it.

Conversely, the browser activities (Playwright, Selenium) and the short built-ins builtin.now and builtin.delay default to execute_locally: true. For the browser activities this is load-bearing — it pins a browser session to one worker — and must not be overridden.

---

Providers

Provider	Activities	What it's for
Built-in Core	3	Clock, delay, and running a workflow inside an activity
State Store	8	Durable key/value storage that outlives a run
Secrets	4	The secret store behind every *_secret_key field
Events & Metrics	2	Debug events and metrics for observability
HTTP	1	Calling any HTTP service
Shell	1	Running a command on the worker
SQL	2	Parameterized queries and writes against PostgreSQL
Email	1	Sending mail over SMTP
Google Drive	6	Reading and writing Drive files
Kubernetes	8	Applying, inspecting and operating cluster resources
Kafka	2	Publishing to and consuming from Kafka
RabbitMQ	2	Publishing to and subscribing to RabbitMQ
GraphQL	1	GraphQL subscriptions
WebSocket	1	Generic WebSocket feeds
MCP	1	Calling a remote MCP server's tools
OpenAI	1	Chat completions against any OpenAI-compatible endpoint
Claude Agent	1	An autonomous agent loop with granted tools
LlamaIndex	7	Building and querying a vector index — RAG
Langfuse	2	Online and offline LLM evaluation
Playwright	26	Browser automation
Selenium	23	Browser automation via WebDriver
Authorization	4	Evaluating authorization policy from a workflow
Deployment & Admin	75	The control plane: namespaces, packages, deployments, users, RBAC

---

All activity types

Built-in core

Activity	Description
builtin.now	Current timestamp on the worker
builtin.delay	Pause for a duration
builtin.execute_workflow	Run a whole workflow in memory inside one activity

State store

Activity	Description
builtin.state.set_state	Write a value
builtin.state.get_state	Read a value
builtin.state.get_state_with_ts	Read a value with its last-update time
builtin.state.del_state	Delete one entry
builtin.state.list_states	List keys in a namespace, filtered and paged
builtin.state.list_namespaces	List namespaces holding data
builtin.state.update_topic	Retag an entry without rewriting it
builtin.state.delete_by_topic	Delete every entry matching a topic pattern

Secrets

Activity	Description
builtin.secret.upload	Store an encrypted secret
builtin.secret.get	Fetch a secret, still encrypted and short-lived
builtin.secret.list	List secret names
builtin.secret.delete	Remove a secret

Events and metrics

Activity	Description
builtin.event.emit_debug_event	Publish a debug event, streamed live to the client
builtin.event.emit_metric_event	Publish a metric to Kafka

HTTP, shell, SQL and email

Activity	Description
http.request	Make an HTTP request
shell.run	Execute a command on the worker
sql.query	Run a SELECT and return the rows
sql.execute	Run a write or DDL statement
email.send	Send an email over SMTP

Google Drive

Activity	Description
gdrive.list	List files, by folder, name or Drive query
gdrive.download	Download a file, inline or to disk
gdrive.upload	Upload or replace a file
gdrive.get_metadata	Read one file's metadata
gdrive.create_folder	Create a folder, optionally reusing an existing one
gdrive.delete	Trash or permanently delete a file

Kubernetes

Activity	Description
k8s.apply	Server-side apply one or more manifests
k8s.get	Fetch a single resource
k8s.list	List resources by label or field selector
k8s.delete	Delete a resource, or a set of them
k8s.scale	Set a workload's replica count
k8s.logs	Read a bounded tail of a pod's log
k8s.wait	Poll until a condition holds, or the resource is gone
k8s.exec	Run a command in a container

Messaging and streaming

Activity	Description
kafka.publish	Publish messages to a Kafka topic
kafka.consume	Consume a topic, relaying each message as an event
rabbit.publish	Publish a message to a RabbitMQ topic
rabbit.receive	Subscribe to a topic, relaying each message as an event
graphql.subscribe	Hold a GraphQL subscription, relaying each payload
websocket.subscribe	Hold a WebSocket connection, relaying each frame
mcp.call_tool	Call a remote MCP tool, relaying its progress

AI

Activity	Description
openai.chat.completions	One chat completion, with tools and structured output
claude_agent.query	An autonomous multi-turn agent loop
llama_index.index_web	Index a list of URLs
llama_index.index_site	Index a site from one seed — sitemap, feed or crawl
llama_index.index_github	Index a GitHub repository
llama_index.index_gdrive	Index a Google Drive folder or file list
llama_index.index_files	Index files on the worker's disk
llama_index.index_docs	Deprecated — use llama_index.index_web
llama_index.query	Semantic search, retrieving chunks or synthesizing an answer
langfuse.run_experiment	Offline evaluation over a Langfuse dataset
langfuse.create_score	Attach scores to the running workflow's trace

Browser automation — Playwright

Activity	Description
playwright.browser.create	Launch a browser and get a session id
playwright.browser.close	Close the session
playwright.browser.get_info	Report on a live session
playwright.page.goto	Navigate to a URL
playwright.page.back	Go back in history
playwright.page.forward	Go forward in history
playwright.page.reload	Reload the page
playwright.element.click	Click an element
playwright.element.fill	Set an input's value
playwright.element.type	Type text with keyboard events
playwright.element.clear	Empty an input
playwright.element.select	Choose a dropdown option
playwright.element.get_text	Read an element's text
playwright.element.get_attribute	Read an element attribute
playwright.element.is_visible	Whether an element is visible
playwright.element.is_enabled	Whether an element is enabled
playwright.element.query_selector	Whether one element matches
playwright.element.query_selector_all	Count matching elements
playwright.page.content	Get the rendered HTML
playwright.page.title	Get the page title
playwright.page.url	Get the current URL
playwright.page.screenshot	Capture a screenshot
playwright.page.evaluate	Run JavaScript in the page
playwright.page.wait_for_selector	Wait for an element to reach a state
playwright.page.wait_for_url	Wait for the URL to match
playwright.page.wait_for_timeout	Wait a fixed time

Browser automation — Selenium

Activity	Description
selenium.browser.create	Launch Chrome and get a session id
selenium.browser.close	Close the session
selenium.browser.get_info	Report on a live session
selenium.nav.goto	Navigate to a URL
selenium.nav.back	Go back in history
selenium.nav.forward	Go forward in history
selenium.nav.refresh	Reload the page
selenium.element.click	Click an element
selenium.element.type	Type text into an element
selenium.element.clear	Empty an input
selenium.element.find	Find one or many elements
selenium.element.get_text	Read an element's text
selenium.element.get_attribute	Read an HTML attribute
selenium.element.get_property	Read a live DOM property
selenium.element.is_visible	Whether an element is visible
selenium.element.is_enabled	Whether an element is enabled
selenium.page.get_html	Get the rendered page source
selenium.page.get_title	Get the page title
selenium.page.get_url	Get the current URL
selenium.page.screenshot	Capture the page or one element
selenium.page.execute_script	Run JavaScript in the page
selenium.wait.element	Wait for an element condition
selenium.wait.time	Wait a fixed time

Authorization

Activity	Description
authz.list_resources	List resources that have policies
authz.get_resource_policy	Fetch a resource's policy
authz.check_privilege	Evaluate whether the caller may perform an action
authz.impersonate_user	Obtain an identity for another user

Deployment and administration

Seventy-five activities, documented by family on the Deployment & Admin page.

Family	Activities	Covers
Namespaces	builtin.deploy.namespace.* (5)	Create, read and delete namespaces
Workflowspecs	builtin.deploy.wfspec.* (4)	Register and list workflowspecs
Packages and files	builtin.deploy.package.* (9)	Versioned packages and their YAML files
Stages, deployments and targeting	builtin.deploy.stage.*, deployment.*, target.*, deploy_package, undeploy_package (11)	Rolling a package out to a stage and an audience
Deployment queries	builtin.deploy.query.* (6)	Which version a user resolves to, and why
Users	builtin.deploy.user.* (5)	User accounts
Groups	builtin.deploy.group.* (7)	Groups and nested membership
Roles and privileges	builtin.deploy.auth.* (21)	Stored RBAC: resources, privileges, roles, members
API keys	builtin.deploy.apikey.* (6)	The calling user's API keys
Audit log	builtin.deploy.audit.get_logs (1)	Querying deployment history