docs(studio-web): document coded function integration [PRODEV-605] by alexandra-c · Pull Request #1675 · UiPath/uipath-python

alexandra-c · 2026-05-22T09:39:19Z

Extend the Studio Web integration guide to cover coded functions alongside coded agents

Copilot

Pull request overview

Extends the Studio Web integration documentation to include coded functions alongside coded agents, updating setup, sync, and CLI usage guidance so users can follow a single guide for both project types.

Changes:

Updates the intro and Cloud Workspace flow to cover importing/pushing coded functions as well as coded agents.
Expands Local Workspace setup with a dedicated Coded Function scaffold path.
Adds agent/function tabbed examples for uipath run and uipath debug, plus new notes around evaluations.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

radu-mocanu · 2026-05-29T11:43:07Z

@@ -1,5 +1,7 @@
 # CLI Reference

+The following commands apply to both **coded functions** and **coded agents**. The entry point name (`main`, `agent`, or any key you define in `uipath.json`) is the first argument to `run`, `debug`, `eval`, and `invoke`.


uipath.json is the main manifest only for coded functions.
for coded agents each integration has its specific manifest where entrypoints are defined.
ex
uipath-langchain -> langgraph.json
uipath-llamaindex -> llamaindex.json
uipath-openai -> openai_agents.json

Updated. The intro now clarifies the entry point is the uipath.json key for functions and the framework manifest for agents, with examples:

for a coded agent it is the entry point declared in the framework manifest — langgraph.json (uipath-langchain), llama_index.json (uipath-llamaindex), or openai_agents.json (uipath-openai-agents)

radu-mocanu · 2026-05-29T11:47:29Z

 /// tab | Bash/ZSH
 ```console
-uipath run agent '{"topic": "UiPath"}'
+uipath run main '{"message": "hello"}'


main is too specific for functions IMO.

Suggested change

uipath run main '{"message": "hello"}'

uipath run <entrypoint> '{"message": "hello"}'

we can have an info box where we give some examples if this is not clear enough. like
uipath run agent / uipath run function

Done — the shell tabs now use uipath run <entrypoint> ..., and I added an info box with examples:

uipath run agent '{"message": "hello"}' # coded agent uipath run main '{"message": "hello"}' # coded function

radu-mocanu · 2026-05-29T11:48:21Z

@@ -0,0 +1,245 @@
+# Python Coded Agents
+
+A coded agent is Python code that uses an LLM reasoning loop to make decisions, call tools, and produce a result. You write the agent logic using a framework of your choice — the `uipath` SDK provides the platform layer: authentication, assets, buckets, connections, tracing, and human-in-the-loop. Package it with the CLI and deploy it as an Orchestrator job.


one or more LLMs*

You write the agent logic using a framework of your choice -> I would mention somewhere around here our curated integrations. uipath-langchain (recommended) or one from uipath-integrations with repo links

authentication, assets, buckets, connections, tracing, and human-in-the-loop. -> this is not a comprehensive list as it currently sounds like. I would also mention guardrails, AI Trust Layer here

Fixed: now reads "uses one or more LLMs in a reasoning loop".

Added — the intro now calls out our curated integrations: uipath-langchain (recommended) plus the uipath-integrations repo, both linked.

Added guardrails and AI Trust Layer, and ended the list with "and more" so it no longer reads as exhaustive.

radu-mocanu · 2026-05-29T11:50:51Z

+
+A coded agent is Python code that uses an LLM reasoning loop to make decisions, call tools, and produce a result. You write the agent logic using a framework of your choice — the `uipath` SDK provides the platform layer: authentication, assets, buckets, connections, tracing, and human-in-the-loop. Package it with the CLI and deploy it as an Orchestrator job.
+
+Use a coded agent when your automation needs multi-step reasoning, dynamic tool selection, or LLM-driven decisions. Use a [coded function](./functions.md) when your logic is deterministic and no LLM is required.


I would remove the automation part here

Removed — now reads "Use a coded agent when your logic needs multi-step reasoning...".

radu-mocanu · 2026-05-29T11:52:47Z

+
+| Layer | Package | Responsibility |
+|-------|---------|---------------|
+| **Platform** | `uipath` | Auth, assets, buckets, connections, tracing, human-in-the-loop, CLI, packaging |


platform is in uipath-platform package.
let s change package to package(s) in column name and add here uipath-core, uipath-platform, uipath-llm-clients etc. (these are only a subset of our libraries)

Done — renamed the column to Package(s) and listed uipath, uipath-core, uipath-platform, uipath-llm-clients, ....

radu-mocanu · 2026-05-29T11:53:10Z

+| PydanticAI | `uipath-pydantic-ai` | Type-safe agents with Pydantic models |
+| Google ADK | `uipath-google-adk` | Gemini models, Google ecosystem |
+| UiPath Agent Framework | `uipath-agent-framework` | UiPath-native agent primitives |


not documented yet, remove for now

Removed all three (PydanticAI, Google ADK, UiPath Agent Framework) — from both this table and the Framework Guides table below. Also dropped them from the install list on the Studio Web page for consistency.

radu-mocanu · 2026-05-29T11:54:24Z

+
+## Input & Output
+
+Define `Input` and `Output` as Python dataclasses, the same way as [coded functions](./functions.md#input--output):


they can be dataclasees/baseModels/typedDicts etc. this depends on the framework used
I would remove this entire section

Removed the entire Input & Output section — the shape depends on the framework.

radu-mocanu · 2026-05-29T11:54:37Z

+
+## Platform Services
+
+The `uipath` SDK gives your agent access to Orchestrator resources at runtime — credentials are injected automatically when running as a job.


not only orchestrator

Generalized — now reads "access to UiPath platform resources" / "full set of platform services" instead of Orchestrator.

radu-mocanu · 2026-05-29T11:55:39Z

+
+---
+
+## Quickstart


why do we need this section? we already have quickstarts for each framework. I would drop it

Dropped the Quickstart section and replaced it with a pointer to the per-framework Framework Guides.

radu-mocanu · 2026-05-29T11:57:24Z

@@ -0,0 +1,423 @@
+# Python Coded Functions
+
+A coded function is Python code with typed input and output that runs as an Orchestrator job. You write plain Python — no agent framework, no LLM required — package it with the CLI, and invoke it from Maestro processes, Coded Apps, or any UiPath job trigger.


UiPath job trigger -> this sounds weird for somebody that is unfamiliar with UiPath platform. I would just extend the enumeration with RPA and Agents

Reworded — extended the enumeration: "Maestro processes, RPA workflows, coded agents, Coded Apps, or any UiPath job".

radu-mocanu · 2026-05-29T11:58:10Z

+
+## Input & Output
+
+Define `Input` and `Output` as Python dataclasses. The runtime validates against these at invocation time and exports them as JSON Schema for Maestro variable binding.


they can also be pydantic BaseModels or thin classes with typed annotations.

why is maestro singled out here? same applies to RPA/Agents

Added — now reads "dataclasses, pydantic BaseModels, or thin classes with typed annotations".

Generalized — the exported JSON Schema is now described for "callers (Maestro processes, RPA workflows, agents, and others)", not Maestro alone.

radu-mocanu · 2026-05-29T12:00:02Z

+
+### Error output pattern
+
+Return business errors as typed output fields rather than raising exceptions. This lets Maestro inspect the error reason and route the process accordingly:


same here about maestro mention. we can include it as an example if we want, but this sounds like coded functions will be only available within maestro processes

Same — now reads "lets the caller (a Maestro process, an RPA workflow, an agent, ...) inspect the error reason and route accordingly".

radu-mocanu · 2026-05-29T12:01:09Z

+
+## Platform Services
+
+`UiPath()` gives your function access to Orchestrator resources at runtime. Credentials are injected automatically when running as a job — no configuration needed.


same here. there are many more FPSs available through UiPath() interface.
full list here

Done — added a link to the full uipath-platform package list so it is clear there are many more services than the few shown.

radu-mocanu · 2026-05-29T12:01:48Z

+
+---
+
+## Platform Services


why is this a section? we have every service already documented. this just duplicates data

Condensed the section to a short intro plus links to the per-service references, instead of duplicating each service with its own code block.

radu-mocanu · 2026-05-29T12:02:07Z

+
+## Idempotency
+
+Functions may be retried by Maestro after a transient failure. Always check for an existing result before writing to an external system:


same maestro phrasing

Fixed — now reads "Functions may be retried by the caller after a transient failure".

radu-mocanu · 2026-05-29T12:02:56Z

+    return Output(result_id=result_id, status="Created")
+```
+
+Use a stable, business-meaningful identifier (invoice number, order ID) as the idempotency key — avoid auto-generated IDs that don't exist before the first write.


this should be phrased as a recommendation, not as a hard rule. I suggest an info box

Reframed as a recommendation in a tip info box rather than a hard rule:

/// tip Prefer a stable, business-meaningful identifier (invoice number, order ID) as the idempotency key ... ///

radu-mocanu · 2026-05-29T12:03:17Z

+
+- Invoked as a **Maestro Service Task** — Maestro binds typed input/output to process variables automatically from the exported JSON Schema
+- Triggered via the **Orchestrator API** (`POST /Jobs/StartJobs`)
+- Run from the CLI: `uipath invoke main '{"..."}'`


only if it is deployed in personal workspace.

Scoped — now reads "Run from the CLI with uipath invoke main ... (when deployed to your personal workspace)".

radu-mocanu · 2026-05-29T12:07:24Z

 ## Syncing Evaluations

-Evaluations can be defined either in Studio Web or locally. They sync automatically when you use `uipath pull` and `uipath push`.
+Evaluations can be defined either in Studio Web or locally, and sync automatically when you use `uipath pull` and `uipath push`. Defining and running evaluations in Studio Web is supported for coded agents only; coded functions can still be evaluated locally with `uipath eval`.


why do we have evals for coded functions? Evals are agent specific. classic python functions can (and should) be unit tested.

Agreed — evals are now scoped to coded agents only. The page notes functions are plain Python to be unit-tested instead, and the "Syncing Evaluations" section is described as a coded-agent feature.

- Add docs/core/functions.md — full reference for coded functions: project structure, I/O types, platform services, tracing, multiple entry points, idempotency, pack/publish with screenshots - Add docs/core/agents.md — guide for coded agents: architecture, framework comparison table (LangChain, LlamaIndex, OpenAI Agents, PydanticAI, Google ADK, UiPath Agent Framework), quickstart, platform services, tracing, framework guides - Add three screenshots: Maestro execution trace, Orchestrator Processes list, Maestro Service Task with I/O binding - Update home page (index.md) with two clear entry points: Functions vs Agents with SDK dependency explained - Update nav: Getting Started → Coded Functions → Coded Agents → CLI - Neutralize agent-only language in cli/index.md, environment_variables.md, FAQ.md, traced.md, getting_started.md Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

sonarqubecloud · 2026-05-29T13:21:06Z

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

Copilot AI review requested due to automatic review settings May 22, 2026 09:39

Copilot started reviewing on behalf of alexandra-c May 22, 2026 09:39 View session

Copilot AI reviewed May 22, 2026

View reviewed changes

Comment thread packages/uipath/docs/core/studio_web.md Outdated

Comment thread packages/uipath/docs/core/studio_web.md

Comment thread packages/uipath/docs/core/studio_web.md

Comment thread packages/uipath/docs/core/studio_web.md

alexandra-c changed the title ~~docs(studio-web): document coded function integration~~ docs(studio-web): document coded function integration [PRODEV-605] May 22, 2026

docs(studio-web): document coded function integration

ee52f4e

alexandra-c force-pushed the feat/coded-function-docs branch from d33e6df to ee52f4e Compare May 22, 2026 10:03

alexandra-c requested a review from radu-mocanu May 22, 2026 10:04

AlexBizon requested a review from EusebiuJecan07 May 27, 2026 15:37

radu-mocanu reviewed May 29, 2026

View reviewed changes

Comment thread packages/uipath/docs/core/studio_web.md Outdated

radu-mocanu reviewed May 29, 2026

View reviewed changes

Comment thread packages/uipath/docs/core/studio_web.md Outdated

radu-mocanu reviewed May 29, 2026

View reviewed changes

alexandra-c force-pushed the feat/coded-function-docs branch from 683cd4d to 17eb20b Compare May 29, 2026 13:18

AlexBizon requested a review from radu-mocanu May 29, 2026 13:53

		@@ -1,5 +1,7 @@
		# CLI Reference

		The following commands apply to both coded functions and coded agents. The entry point name (`main`, `agent`, or any key you define in `uipath.json`) is the first argument to `run`, `debug`, `eval`, and `invoke`.

	uipath run main '{"message": "hello"}'
	uipath run <entrypoint> '{"message": "hello"}'

		@@ -0,0 +1,245 @@
		# Python Coded Agents

		A coded agent is Python code that uses an LLM reasoning loop to make decisions, call tools, and produce a result. You write the agent logic using a framework of your choice — the `uipath` SDK provides the platform layer: authentication, assets, buckets, connections, tracing, and human-in-the-loop. Package it with the CLI and deploy it as an Orchestrator job.


		A coded agent is Python code that uses an LLM reasoning loop to make decisions, call tools, and produce a result. You write the agent logic using a framework of your choice — the `uipath` SDK provides the platform layer: authentication, assets, buckets, connections, tracing, and human-in-the-loop. Package it with the CLI and deploy it as an Orchestrator job.

		Use a coded agent when your automation needs multi-step reasoning, dynamic tool selection, or LLM-driven decisions. Use a [coded function](./functions.md) when your logic is deterministic and no LLM is required.


		## Input & Output

		Define `Input` and `Output` as Python dataclasses, the same way as [coded functions](./functions.md#input--output):


		## Platform Services

		The `uipath` SDK gives your agent access to Orchestrator resources at runtime — credentials are injected automatically when running as a job.

		@@ -0,0 +1,423 @@
		# Python Coded Functions

		A coded function is Python code with typed input and output that runs as an Orchestrator job. You write plain Python — no agent framework, no LLM required — package it with the CLI, and invoke it from Maestro processes, Coded Apps, or any UiPath job trigger.


		### Error output pattern

		Return business errors as typed output fields rather than raising exceptions. This lets Maestro inspect the error reason and route the process accordingly:


		## Platform Services

		`UiPath()` gives your function access to Orchestrator resources at runtime. Credentials are injected automatically when running as a job — no configuration needed.


		## Idempotency

		Functions may be retried by Maestro after a transient failure. Always check for an existing result before writing to an external system:


		---

		## Quickstart

Conversation

alexandra-c commented May 22, 2026 • edited by atlassian Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

radu-mocanu May 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

radu-mocanu May 29, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

alexandra-c commented May 22, 2026 •

edited by atlassian Bot

Loading

radu-mocanu May 29, 2026 •

edited

Loading

radu-mocanu May 29, 2026 •

edited

Loading

radu-mocanu May 29, 2026 •

edited

Loading

radu-mocanu May 29, 2026 •

edited

Loading