The agent builder is the workspace for creating agent workflows. Use it when you need to define how an agent behaves, what tools it can use, and how it connects to other agents.

Main tasks

  • Create an agent workflow.
  • Edit agent configuration.
  • Connect nodes on the canvas.
  • Review workflow history.
  • Save changes before running the agent.

Workflow types

Pinter supports several workflow shapes: The builder canvas exposes the sequential, parallel, and loop shapes. The backend also supports two additional orchestration types configured through the agent API: conditional (routes between branches based on a condition) and human-in-the-loop (pauses for human confirmation). See Backend and ADK architecture for the full agent type table.

Write instructions with Rin

The instruction fields appear as expandable sections for identity, mission, methodology, boundaries, and examples. Open a section to edit it, or use Ask Rin in the agent header to get help writing all five fields.
1

Ask Rin

Select Ask Rin in the agent header at the top of the config panel. The Rin panel opens with tailored prompt suggestions for this agent (e.g. writing instructions, recommending tools, reviewing boundaries), or you can type a custom question.
2

Refine in chat

Answer Rin’s follow-up questions or ask for changes until the instructions fit the agent.
3

Review the updated fields

Rin updates the agent’s instruction fields through its tools. The builder picks up the change automatically; expand each section to review it before publishing.
Ask Rin only appears when the workspace Rin agent is configured (RIN_AGENT_UUID).

Version history

The history panel shows Published Versions alongside the Active Canvas Copy (Draft). Use it to:
  • Compare instruction changes between versions with the field-level diff view.
  • Restore an earlier version when a change regresses agent behavior.
Publish a version when the agent is ready for users; keep iterating on the draft without affecting the published version.

Deployment lifecycle (Stages)

Every agent in Pinter tracks an explicit deployment status to distinguish production-ready agents from prototypes:

Promoting deployment status

Once an agent is published, administrators (L1/L2 with agent:publish permission) can promote its deployment stage directly from the canvas header dropdown:
  1. Click the deployment status badge at the top of the canvas.
  2. Select the target stage: Experimental, Staging (UAT), or Production.
  3. Changing stage updates the active deployment tag immediately without incrementing the version number.

Auto-demotion on runtime changes

To protect users against running unverified modifications under a PROD or STAGING label, Pinter enforces auto-demotion:
  • If an agent in PROD or STAGING has any runtime-impacting field changed (instructions, model, tools, sub-agents, canvas connections, or asset bindings) and saved, its status is automatically reset to Experimental.
  • Cosmetic edits (agent name, description, avatar, canvas layout position) do not trigger demotion.

Attach assets and collections

In the agent configuration, Assets lets you attach rules, skills, Knowledge Hub documents, and collections. Open the picker, use the Collection tab (or the Collection button), and select the collections you want. Every document in an attached collection becomes retrieval knowledge for the agent and follows the collection as it changes. Only the collection creator or an L1/L2 admin can attach a collection. See Knowledge documents.

Share an agent

Use Share Agent in the canvas header to give other users access directly or through claim codes. See Sharing agents.

Builder route

Open the builder from the application navigation or go to:
Specific agents open under:

Good builder practice

Start with one agent and one tool. Add workflow structure only when the task needs it.
Keep each agent focused on one responsibility. Smaller agents are easier to test and reuse.