Skip to main content
Use this page while creating or editing experts. It explains what each field controls, defaults, and how to choose values.

Basics and Behavior

FieldRequiredDefaultWhat It ControlsRecommended Value
NameYesNoneDisplay name in marketplace/cardsClear role-based name, e.g. Quarterly Planning Expert
DescriptionYesNoneOne-line summary in lists/details1-2 sentences focused on outcomes
Logo URLNoEmptyAvatar image shown in UIStable HTTPS image URL
About (Markdown)NoEmptyLong-form detail page contentScope, examples, limits, expected outputs
System PromptYesNoneCore behavior/rulesRole, boundaries, tools, output format
Custom Hiring MessageNoSystem messageFirst DM response after hireOne line with a suggested first question
SkillsNoEmptyReusable packaged behavior/filesAdd only for structured reusable workflows
CategoriesNoEmptyMarketplace discoverability tags2-5 specific categories
Primary ModelNoClaude Sonnet 4.6The AI model that powers this expert. Standard users can pick standard-tier models; admins can also pick premium models (e.g. Claude Opus 4.8 or GPT-5.5).Leave unset to use the platform default. Choose a lighter model for high-volume simple tasks; a heavier model for deep reasoning.
Reasoning effortNoModel-dependentPer-expert override of the reasoning token budget. Enum string: one of none, minimal, low, medium, high, xhigh, max. Only honored by reasoning-capable models (currently Claude Sonnet 4.6, the Claude Opus family, and the OpenAI GPT-5.4 / GPT-5.5 models); hidden for all other models (e.g. Claude Haiku 4.5 and the Gemini models).Start with medium. See Reasoning effort for the full value guide.

Configuration

Hives

Hives are Braintrust’s pre-built tool bundles. Always prefer hives over custom MCP servers when a hive exists for the service you need — they are tested, maintained, and handle per-user credential isolation automatically. See the Hives Reference for the full catalog.
FieldRequiredDefaultWhat It ControlsRecommended Value
HivesUsuallyEmptyBuilt-in tool bundles that connect the expert to external services (Google Drive, Sheets, Notion, Jira, etc.)Start minimal; add only what the expert actually needs
Hive Env OverridesNoEmptyShared default values for a hive that apply to all users of this expert (e.g. a default Jira project key, a default database name)Safe, non-sensitive shared defaults only — never credentials or API keys

MCP Servers

Use custom MCP servers only when no built-in hive covers the service you need. You are responsible for maintaining the endpoint and debugging connection issues.
FieldRequiredDefaultWhat It ControlsRecommended Value
MCP Server NameIf MCP usedNoneLabel for the server shown in the editorHuman-readable integration name
MCP Server URLIf MCP usedNoneThe custom tool endpoint the expert connects toValid https:// URL
MCP TransportIf MCP usedsseMCP protocol mode used to communicate with the serverKeep sse unless the server explicitly requires streamable_http
Verify MCP ServerRecommendedNot runRuns a connectivity and tool discovery check against the URLAlways run before saving — an unverified server fails silently at runtime

Automations (Advanced)

FieldRequiredDefaultWhat It ControlsRecommended Value
Automation TypeYescron when created as cronTrigger modecron for schedules, calendar for event-driven
Display NameYesEmptyUser-facing automation labelAction-focused and specific
DescriptionNoEmptyWhy/when automation is usedOne sentence with clear use case
Execution PromptYesEmptyTask instructions during trigger executionExplicit action + output format
Policy Prompt (calendar only)NoFalls back to agent name + descriptionTells the AI how to evaluate each calendar event — which to ignore, which to schedule, and how many minutes before the event to runWrite explicit criteria: which event types to act on, which to ignore, and your preferred lead time. See Triggers for examples.

Access and Visibility

FieldRequiredDefaultWhat It ControlsRecommended Value
Expert Email (local part)Yes (create)EmptyUnique expert identityLowercase, readable, stable (sales-ops-bot)
VisibilityYesPrivateWho can discover/hireStart private, publish after validation
Access List (Manage Access)DependsEmptyWho can edit/hireMinimum required users/domains only

Edit-Only Fields

FieldRequiredDefaultWhat It ControlsRecommended Value
ImagesNoEmptyVisual examples on expert page2-4 screenshots of real outputs
Available OnNoEmptyPlatform compatibility hintsOnly platforms you actively support
Google Drive FolderNoEmptySynced knowledge source folder (RAG chunk injection)See Knowledge Base for how retrieval works

Common Mistakes and Fixes

MistakeWhat HappensFix
Prompt is too genericLow-quality/unstable responsesAdd explicit role, scope, and output rules
Too many hives initiallyNoisy tool behavior and harder debuggingStart with minimal hives, expand later
Broken/incorrect Logo URLMissing avatar in UIUse a public HTTPS image URL
MCP URL saved without verifyRuntime tool failuresVerify each MCP server before saving
Automation prompt too vagueUnusable scheduled outputSpecify exact output shape and audience
Access too broadUnintended usage/edit riskRestrict access list and roles

Validation Rules (Current Product Behavior)

  • Name: minimum 3 characters; letters/spaces only.
  • Description: minimum 10 characters.
  • System Prompt: minimum 10 characters.
  • Email local part: minimum 3; lowercase letters/numbers/./_/-; no @.
  • MCP server: name and valid URL required.
  • Automation template: display name minimum 3; prompt minimum 10; names must be unique.

System Prompt Patterns

A well-structured system prompt is the single biggest factor in expert quality. Use these patterns as a starting point.

Role Definition

Start every prompt with a clear role statement: “You are [ROLE] for [CONTEXT].” This anchors the expert’s behavior.

Boundary Setting

Explicitly state what the expert should not do. This prevents scope creep and hallucinated capabilities.

Tool Usage Instructions

Tell the expert when to use which tool. Without guidance, experts may use tools unnecessarily or miss tools they should use.

Output Format Specification

Define the expected format, length, and structure of responses. Be specific — “a bullet list” is better than “a summary.”

Uncertainty Handling

Tell the expert what to do when it doesn’t know: ask the user, state uncertainty, or skip the question.

Template

You are [ROLE] for [CONTEXT].

**Your job:** [1-2 sentences on primary function]

**What you do NOT do:** [3-5 explicit exclusions]

**Tools available:**
- [Tool name]: use when [condition]
- [Tool name]: use when [condition]

**Output format:** [specify format, length, structure]

**When unsure:** [how to handle uncertainty - ask vs assume vs skip]

Q&A / Information Agent

A simple expert that answers questions using a knowledge base.
FieldRecommended Value
HivesMinimal — Google Drive if using a knowledge base
System PromptStrong, focused prompt with clear scope and refusal rules
SkillsDomain knowledge skill if available
TriggersNone — fully interactive
Drive FolderConnect a curated folder as a RAG knowledge source — see Knowledge Base for how retrieval works and when to pair it with the Google Drive hive

Data Analysis Agent

An expert that pulls data from various sources and produces analysis.
FieldRecommended Value
HivesGoogle Sheets, Supabase, NetSuite, or other data hives
System PromptInclude detailed output format instructions (tables, charts, summaries)
SkillsAnalysis templates or report format skills
TriggersOptional cron for recurring reports

Automated Reporting Agent

An expert that runs on a schedule and delivers reports to a Space or DM.
FieldRecommended Value
HivesData source hives (Sheets, NetSuite, Supabase)
System PromptFocus on report structure, audience, and delivery format
TriggersCron trigger with explicit execution prompt
Execution PromptSpecify exact data to pull, report format, and where to deliver
Delivery ModeSpace for team reports, DM for individual summaries