Data access (get, set, input, request)

get() and set() are the two functions you use for almost everything in kdeps. input and request are property-style shorthands for the incoming HTTP request. All of them work the same way in string interpolation {{ }}, in before:/after: blocks, and in validations.check conditions. get()/set() are available in both modes; input/request are workflow mode only (they read the HTTP request).

get() - read any value

get('key') searches a priority chain and returns the first match. You rarely need to specify a source explicitly.

get('q') search order:
  1. items context    (loop/iteration current item)
  2. memory           (values set with set() this request)
  3. session          (values set with set(..., 'session'))
  4. resource outputs (values produced by prior resources)
  5. URL query params (?q=hello)
  6. request body     ({"q": "hello"})
  7. request headers
  8. system metadata
  9. uploaded files

To skip the chain and read from a specific source:

get('q', 'param')      # URL param or body field only
get('auth', 'header')  # request header only
get('user', 'session') # session only -- persists across requests
get('API_KEY', 'env')  # environment variable

Reading a resource output works the same way - get('llm') returns whatever the llm resource produced:

# resources/example.yaml
requires: [llm]
apiResponse:
  response:
    answer: get('llm')          # full output
    text: get('llm').answer     # field access when LLM returns JSON

set() - store a value

set() writes into memory (current request) by default. Pass 'session' to persist across requests.

# resources/example.yaml
after:
  - set('normalized', lower(trim(get('q'))))   # available to downstream resources
  - set('user_id', get('id'), 'session')        # survives to the next request

set() is like assigning to a variable. Downstream resources read it with get().

file() - read uploaded files

# resources/example.yaml
content: file('doc.pdf')    # file uploaded with the request
images: file('*.jpg')       # glob pattern -- returns first match

info() - request metadata

# resources/example.yaml
id: info('ID')           # unique ID for this request
ip: info('IP')           # caller IP address
path: info('path')       # URL path
ts: info('timestamp')    # current timestamp

Resource-specific accessors

get('resourceId') returns the main output of a resource. Use these accessors when you need lower-level details:

# resources/example.yaml
after:
  # Python and exec resources
  - set('ok',  exec.exitCode('build') == 0)
  - set('err', exec.stderr('build'))

  # HTTP resources
  - set('status', get('api').statusCode)
  - set('ct',     http.responseHeader('api', 'Content-Type'))

  # LLM resources
  - set('raw', llm.response('chat'))

The input object

input.field is a shorthand for a request body field - identical to get('field'), but with property syntax:

# resources/example.yaml
after:
  - set('query', input.q)
  - set('city', input.user.address.city)   # nested access
chat:
  prompt: "Hello {{ input.name }}, you asked about {{ input.topic }}"

input	equivalent get()
input.field	get('field')
input.user.name	get('user').name
input.items[0]	get('items')[0]

input reads body data only - no query params, no headers, no source hints. It returns nil for a missing field, so wrap it in default() when the field is optional. input.name is exactly request.body.name. Reach for get() when you need any other source or a fallback.

The request object

request gives direct access to the raw HTTP request when running in workflow mode.

Property	Type	Description
request.method	string	HTTP method (GET, POST, ...)
request.path	string	Request URL path
request.IP	string	Client IP address
request.ID	string	Unique request identifier
request.headers	object	All request headers
request.query	object	Query parameters
request.body	object	Request body (same as input)

File and lookup methods:

# resources/example.yaml
after:
  - set('doc',     request.file('document'))       # file content (text) or nil
  - set('path',    request.filepath('image'))      # temp path on disk
  - set('type',    request.filetype('upload'))     # MIME type
  - set('images',  request.filesByType('image/*')) # paths matching a MIME glob
  - set('count',   request.filecount())            # number of uploaded files
  - set('all',     request.files())                # all uploaded file paths
  - set('types',   request.filetypes())            # MIME type of every upload
  - set('auth',    request.header('Authorization'))
  - set('page',    request.params('page'))         # query parameter

Each request method has a get()/info() equivalent - use whichever reads clearer:

request	Unified API
request.params('key')	get('key', 'param')
request.header('Name')	get('Name', 'header')
request.file('name')	get('name', 'file')
request.filepath('name')	get('name', 'filepath')
request.filecount()	info('filecount')
request.method	info('method')

See also

• Expressions - expression syntax
• Expression helpers - Json(), Safe(), default(), and friends
• Expression functions reference - complete function list
• File upload example - request.file() end to end