Skip to main content

Agent designer

The Agent Designer is the advanced UI for building and editing agents. URL: /app/agent-designer-advanced.

For a guided alternative aimed at non-technical creators, use the Agent Wizard.

Layout

┌─────────────────────────────────────────────────────────────────┐
│ Identity Behaviour Tools Knowledge Sharing Advanced │
├─────────────────────────────────────────────────────────────────┤
│ │
│ <fields for the active tab> [Test in chat] ▶ │
│ │
└─────────────────────────────────────────────────────────────────┘

The right-side Test in chat pane is always live — every change you save reflects there immediately so you can iterate.

Tab 1 — Identity

FieldNotes
NameRequired. Visible in the picker.
DescriptionOne-liner.
AvatarImage upload, emoji, or generated.
CategoryPicker (Marketing / Engineering / etc.) — affects discoverability in the marketplace.

Tab 2 — Behaviour

FieldNotes
ModelSpecific model ID (e.g. claude-opus-4-7) or null = inherit org default.
EffortLow / Medium / High — controls reasoning depth where supported.
System promptMarkdown, can include {{user_name}}, {{org_name}}, {{today}} template variables.
Starter promptsUp to 4 quick-click prompts shown at the top of every empty conversation.
Threads enabledAllow branching.
Workspace paneAuto-open results in the side pane.
Copy buttonShow "copy reply" buttons next to each message.

Tab 3 — Tools

SettingNotes
Tool selectionCheck / uncheck individual tools the agent may call. Defaults to "all tools the user has access to".
Tool deny-listBlock specific tools even if the user has them.
Per-tool promptsOverride the default tool description with your own — useful for guiding the model.

The list of tools is computed live from the three-layer access matrix, so the picker only shows tools that would actually be available at runtime.

Tab 4 — Knowledge

FieldNotes
Attached KBsOne or more. Searched automatically when the agent answers.
Source referencesShow inline citations (off) or as a dedicated card list (on).
Auto-searchSearch KB on every turn (default), or only when the agent decides to call kb_search.

Tab 5 — Sharing

FieldNotes
VisibilityJust me / Specific groups / Whole org.
Shared groupsMulti-select NC / Bee Flow groups.
Embed enabledAllow embedding in an iframe outside Bee Flow.
Embed origin allow-listComma-separated list of permitted parent origins.
PublishableAllow this agent to be published to the public marketplace.

Tab 6 — Advanced

FieldNotes
Privacy Shield levelOverride the org default for this agent.
ModerationToggle Azure Content Safety moderation on responses.
Web-search guardApply PII filter to search results.
Memory accessPull user's memory into context. Default: on.
Conversation retentionDefault / 30 days / 7 days / no retention.
Stop sequencesCustom strings that end generation.
Temperature0–2.
Max output tokensDefault model max, or override.

Test in chat

The right-hand pane is a real conversation — except it's scoped to the current draft and isn't saved to the user's history. Switch agents on the left and the test resets.

Saving + publishing

ActionWhat it does
SaveStores the draft. Visible in your library, not yet to others.
Publish to orgVisible to whoever you specified in Sharing.
Publish to marketplaceSubmit for the public marketplace (admin approval if required).

Forking

Click Duplicate on any agent (yours or marketplace-installed) to create a copy you can edit freely. Useful for starting from a template without polluting the original.

Memory of agents

Each agent has its own memory store keyed by (agentId, userId). The Memory Extractor system agent populates it with facts the agent has learnt across conversations ("Alice prefers concise replies", "Alice's project is called Helix"). The user manages it under Settings → Account → Memory.