Function Settings Reference

Fields available under Resources → Functions. Labels are as they appear in the app. For concepts, see Functions.

Common fields

Field	Required	Description
Function Name	Yes	The tool name the model sees. For Local Python it must match the name of a function defined in the code. For skills: lowercase letters, digits and single hyphens, at most 64 characters.
Description	Yes	What the tool does and when to use it. The model reads this to decide whether to call the tool. Skills: at most 1,024 characters.
Function Type	Yes	Local (Python), MCP Stdio (Subprocess), MCP HTTP (SSE) or Skill (SKILL.md). Existing execution environment tools show their own type.
Approval Override	No	Inherit from risk appetite (default), Always require (pin) or Never require (pin). Team and agent pins take precedence over a function pin. A never-require pin can't skip approval for a critical-risk call unless the risk appetite bypasses all approvals.
Tags (comma-separated)	No	Labels for filtering and search. Shown as removable chips under Current Tags.

The function's folder in the library tree is its path. New functions are created at /; drag a function onto a folder to move it.

Authority

On the create form and the Definition tab.

Control	Description
Link Integration	Opens Link Integrations to Function to attach one or more integrations. Linked integrations appear as chips; click one to open it, or remove it with its delete icon.
Allowed Operations	Operations this function may call through the broker, picked from those the linked integrations declare. Empty means the function can't make brokered (credentialed) calls. A granted operation that has since been removed from the integration is shown with a warning icon; calls to it are refused.
Internal	The function authenticates to the FYAI platform itself as the caller (it calls FYAI's own API). The caller's identity is passed in jwt_claims, and internal MCP servers receive it as FYAI_API_TOKEN. Doesn't widen access to third-party integrations.

Local (Python)

Field	Required	Description
Code	Yes	Python source. Edited on the Definition tab in the Source editor, which shows the cursor line and column and whether there is an unsaved edit. Save saves; Save and run saves then runs the console.

The parameter schema is derived from the function signature. See Creating Custom Functions.

Contract chips under the editor: parameter count, secret-shaped parameters, egress (brokered, direct egress, MCP, no network, unreadable), env fallback, and declared or undeclared operations.

MCP Stdio (Subprocess)

Field	Required	Description
Command (e.g. npx)	Yes	Executable that starts the server, such as npx, uvx or python.
Arguments (JSON array e.g. ["-y", "pkg"])	No	Arguments as a JSON array. On the Definition tab they're edited one per line.
Server Placement	No	Workspace — needs the session filesystem (default) or Stateless — wraps a remote API. Labelled Placement on the Definition tab.
.npmrc Content	No	Custom .npmrc for npm registry routing. Supports ${ENV_VAR} substitution.
Environment	No	Definition tab only. Key/value environment variables for the subprocess. A value of integration:<id>:<secret name> is replaced with that secret from a linked integration.

MCP HTTP (SSE)

Field	Required	Description
URL	Yes	The server endpoint. Labelled Endpoint on the Definition tab.
Headers (JSON object)	No	HTTP headers. {{SECRET_NAME}} placeholders are filled from linked integration secrets. Edited as key/value rows on the Definition tab.

MCP tool controls

On the Definition tab of either MCP type, in Discovered tools:

Control	Description
On	Expose or hide the tool. Hidden tools are removed from the model's tool list.
Kind	read or write, guessed from the tool name.
Approval	Per-tool inherit, never require or always require.
Hide write tools	Hides every write-kind tool.
Rediscover	Starts the server as one of your agents and refreshes the tool list.

The Health card offers Retry start and Show full log.

Skill (SKILL.md)

Field	Required	Description
SKILL.md	Yes	The skill body in Markdown, injected when the model activates the skill.

Execution environment tool

Field	Required	Description
Execution Environment	Yes	The execution environment whose tools this function exposes.

Header controls

Control	Description
Killswitch (Active / Killed)	Killing the function prevents it being built or run anywhere until reactivated, and locks its configuration read-only.

Actions

Action	Description
Create	Save a new function. Back returns to the type chooser.
Update	Save changes. Only changed fields are sent.
Cancel	Discard unsaved edits.
Delete	Delete after confirmation.
Copy Function	Create a copy under a New Function Name, including linked integrations and allowed operations.
Create Data Source	Open Create Data Source from Function: Name, Description (optional), Attach integrations, Refresh mode (On demand or Scheduled (interval)), Interval (seconds) (minimum 60, default 900), Static arguments (JSON, optional). Create & Test creates it and runs a test fetch; Test again re-runs it.
Publish Template	Platform administrators only. Publish the current values as a platform template.
Import Skill	Tree header button. Upload a .zip, .tar.gz, .tgz or .tar archive, or give a public Git repository URL.

Platform-managed functions (installed from a platform template) are read-only for everyone except platform administrators.

Test console

Control	Description
Run as agent	The agent whose identity, integrations and execution environment the run uses.
Tool	Shown when the function exposes more than one tool.
Reload	Rebuild the function's tools.
I supply the arguments / Let the agent decide	Direct call with arguments, or a prompt to an agent that has only this function.
raw JSON	Enter arguments as a JSON object instead of the form.
Run	Run the test.
Result tabs	Output, Trace, Logs, Request.
Recent runs	Earlier runs from the console; click one to reload its arguments.

See Testing Functions.