uip agent

uip agent is the Agent tool: it scaffolds, configures, validates, packages, and operates low-code agents — the agent.json-driven projects authored in Agent Builder / Studio Web. Install it with uip tools install agent, or invoke uip agent … and let auto-install run. See Tools (plugins) for the install model and uip tools for the command that manages tool installation.

This tool is distinct from the coded agent tool (uip codedagent), which targets Python agents built with LangGraph / LlamaIndex / OpenAI Agents. uip agent only operates on low-code agent projects: a directory containing agent.json, entry-points.json, project.uiproj, an evals/ tree, and features/ / resources/ folders.

What a UiPath Agent is

A low-code agent is described declaratively by a single agent.json file — model, prompts, input/output schemas, tools, contexts, escalations, and settings. Supporting files describe the entry points that Studio Web uses to invoke the agent, the default evaluation set, and any RAG / HITL resources.

The typical lifecycle covered by uip agent is:

1. Scaffold a project on disk with uip agent init.
2. Author: update agent.json directly or via the resource-management verbs (tool, context, escalation, input, output, eval).
3. Validate the project (uip agent validate) — runs static checks and the schema-migration pipeline.
4. Push the project into Studio Web so it appears in the Agent Builder UI (uip agent push), or keep editing locally and pack it into a .uis archive (uip agent pack).
5. Publish the packed agent to Orchestrator as a solution package version (uip agent publish).
6. Deploy the published package to a folder (uip agent deploy) — this creates the folder, installs the process, provisions a serverless machine and robot user, and activates the deployment.
7. Run the deployed agent as a job (uip agent run start) and monitor it (uip agent run status, uip agent run list).
8. Evaluate the agent against an evaluation set (uip agent eval run start), review results (uip agent eval run results), and compare against a baseline (uip agent eval run compare).

Operations that round-trip a project to and from Studio Web use the push / pull pair and the list / share sharing verbs. File-level operations inside a Studio Web project are exposed through uip agent file.

Verbs

命令	用途
init	Scaffold a new agent project on disk (standalone, or inline inside a flow project).
config	Read or update keys in agent.json (model, name, systemPrompt, etc.).
validate	Static-check the project and run the schema-migration pipeline; write migrated files back.
tool	Add, discover, connect, list, or remove agent tools (Integration Service / process / API workflow / etc.).
context	Add, list, or remove RAG context resources bound to index sources.
escalation	Add, list, or remove HITL / escalation resources.
input / output	Manage the agent's input and output schema parameters.
eval	Manage evaluators, evaluation sets, test cases, and evaluation runs.
pack	Bundle the project into a .uis archive.
publish	Pack and push the agent to Orchestrator as a solution package version.
deploy	Install and activate a published package version in an Orchestrator folder.
run	Start, inspect, and list agent jobs on Orchestrator.
list	List solutions visible to the current user in Studio Web.
pull	Download a solution from Studio Web as a .uis archive.
push	Import a local project into Studio Web.
文件	List, download, and upload files inside a Studio Web project.
share	Grant, list, or revoke user/group access to a Studio Web solution.

Synopsis

uip agent <verb> [subverb] [arguments] [options]

Always uip agent, never uip agents or uip a. agent is the tool's commandPrefix (see uip tools list).

All uip agent verbs honor the four global options (--output, --output-filter, --log-level, --log-file). Exit codes follow the shared contract.

端到端流程

The minimal expected flow from zero to a running deployed agent:

# 1. Scaffold
uip agent init ./my-agent --model gpt-4o-2024-11-20

# 2. Validate (run after any edit to agent.json)
uip agent validate ./my-agent

# 3a. Iterate in Studio Web (optional)
uip agent push ./my-agent
# …edit in Studio Web, then pull back:
uip agent pull <solutionId> -d ./downloads --extract

# 3b. Or pack locally for inspection
uip agent pack ./my-agent -d ./dist

# 4. Publish to Orchestrator
uip agent publish ./my-agent --package-version 1.0.0
# → captures PackageVersionKey

# 5. Deploy to a folder (auto-creates folder + provisions machine and user)
uip agent deploy <PackageVersionKey> --name invoice-agent

# 6. Find the release, then run
uip agent run list --folder-id <FolderId>
uip agent run start <ReleaseKey> -i '{"input":"hello"}'
uip agent run status <JobId>

# 7. Evaluate (push must have happened first — eval runs against the cloud project)
uip agent eval run start --set default --path ./my-agent --wait

身份验证

Every verb that talks to Studio Web or Orchestrator (push, pull, list, share, publish, deploy, run *, eval run *, file *) requires a valid CLI session. Run uip login first. See Authentication for the session model and --login-validity behaviour.

Local-only verbs (init, config, validate, pack, tool, context, escalation, input, output, eval add / eval set / eval evaluator) operate purely on files and do not require a session.

另请参阅

• Concepts: tools (plugins) — the plugin model and version pinning.
• Concepts: skills — how agents surface as skills and how skills relate to tool resources.
• Authentication — sessions, tenants, and folders.
• Orchestrator: jobs — the underlying job APIs that uip agent run delegates to.