Skip to main content
The Agent Builder is where you configure every aspect of an agent. You open it by clicking an agent in the My Agents list or by clicking Create new agent (which goes to /agents/builder/new). Agent Builder for the Sales Assistant agent — all six tabs (Instructions, Tools, Resources, Channel, Monitor, Advanced) pinned on the left, Instructions tab active with the agent name and system prompt, and the right-hand Preview pane showing an active chat where the user asked 'Summarize this week's pipeline changes and highlight anything at risk' and the agent is replying The builder is a three-part layout:
  1. Top bar — back arrow, agent name (with quick actions), tab strip, Save and Release buttons.
  2. Tab panel (left) — the configuration surface. You can have up to six tabs pinned at once.
  3. Preview pane (right) — a live chat against your draft, or an Information view of the agent’s current config.

The top bar

From left to right:
  • ← Back arrow — returns you to /agents.
  • Agent name — click the chevron next to the name to open a small menu with Tag, Share, and Delete actions. Edit the name itself from the Instructions tab.
Agent name dropdown open showing Tag, Share, and Delete (in red) options
  • Tabs — a horizontal strip of the sections you’ve pinned. See below.
  • Save — writes the current form values as a new draft version. Disabled when nothing has changed and on brand-new agents until required fields are filled.
  • Release — opens the Publish dropdown. See Save & Release.
Some buttons fire “Please save the agent before [action]” toasts when you try to Tag, Share, or Delete a brand-new agent that hasn’t been saved yet.

The six tabs

Builder tabs strip with all six tabs pinned — Instructions, Tools, Resources, Channel, Monitor, Advanced — plus a trailing + button for un/re-pinning Every agent has up to six configuration tabs, listed here in their canonical order:

Instructions

Name, icon, description, system prompt, model, prompt bubbles. Always pinned — you can’t remove it.

Tools

MCP connections, embedded services, agents-as-tools, workflows-as-tools.

Resources

Datastores, filestores, knowledge bases with Read / Write / Admin permissions.

Channels

Slack, Microsoft Teams, External (public URL).

Monitor

Replay this agent’s chats from inside the builder.

Advanced

Output Schema — structured JSON responses for API callers.

Pinning tabs

Only Instructions and Tools are pinned by default. Click the + at the end of the tab strip to pin Resources, Channels, Monitor, or Advanced. Pinned tabs persist across sessions. If a tab has validation errors, its label turns red.
The Channels and Advanced tabs are hidden for Pinkfish’s built-in system agents.

The preview pane

The right side of the builder is a live sandbox against your current draft. It has two tabs:

Test

The default view — a chat interface connected to the draft you’re editing. Type a message, watch the agent respond with the current instructions, model, and tools. This is where you validate changes before releasing. Preview pane showing an active chat with the Sales Assistant agent — user prompt 'Summarize this week's pipeline changes and highlight anything at risk.' and the agent replying with a 'Working... 5s' indicator For brand-new agents you’ll see:
“Save the agent first, then start a conversation to test it here.”
Click + New Chat to clear the preview and start fresh — useful when you’ve changed instructions and want to see behavior from a clean starting point.

Information

The Information tab is a read-only summary of what the agent will actually do when invoked. Information pane showing Selected Model (Sonnet 4.5), Selected Tools (1) — Knowledge Base expanded showing four tools (Ask, Create/Rewrite, Search, Update/Delete), Auto-selected Tools (5) — Write Artifact, Read Artifact, List Artifacts, Write Artifact from URL, Code Execution — and Selected Resources (2) — two Blueberries Chat Collection entries with Read/Write/Admin summaries It shows:
  • Selected Model — the model label (e.g. Sonnet 4.5).
  • Selected Tools (N) — tools you picked, grouped by server. Shows All X tools when you’ve selected a server wholesale, or the exact tool names when you picked a subset. Expand rows with more than five tools via +N more.
  • Auto-selected Tools — built-in tools that every agent gets automatically (e.g. Write Artifact, Read Artifact, Run Code). These can’t be turned off.
  • Selected Resources (N) — attached datastores/filestores/KBs with a Read / Write / Admin summary.
The Information view is especially handy when debugging — if the agent isn’t using a tool you expected, this pane tells you whether it’s in scope.

The full workflow

1

Open or create

Click a row in My Agents, or click Create new agent to land on /agents/builder/new.
2

Fill the Instructions tab

Name, instructions, icon, description, model. These are the minimum to save.
3

Save

Click Save to persist a draft. You can test it in the Preview pane now.
4

Pin the other tabs as needed

Click + to add Resources, Channels, Monitor, or Advanced. Configure each.
5

Iterate in Preview

Chat with the draft. Check the Information tab to confirm model, tools, and resources look right.
6

Release

Open the Release dropdown and publish. The released version becomes available in Coworker and on enabled channels.