Expert Capabilities - Braintrust Docs

Not every expert has every capability. Use this page as a high-level checklist.

Common Platform Integrations

Platform	Typical Actions
Google Docs	Create, read, update documents
Google Sheets	Read/analyze/update spreadsheet data
Google Drive	Search files, read docs, organize content
Google Calendar	Read schedule, create events
Gmail	Draft/search emails
Google Chat	Respond in DM and group Spaces
Notion	Read/update pages and databases
Custom APIs (MCP)	Call external/internal services through custom connectors

Input and Output Types

Type	Supported
Input	Google Docs/Sheets, PDFs, text files, common images, selected attachments
Output	Text responses, structured summaries, links to created docs/sheets, tool-driven actions

Conversation and Collaboration

Capability	Notes
Streaming responses	Replies appear progressively in chat
Multi-turn context	Recent context is used for follow-up questions
Group Space routing	`@BrainTrust` routes to best-fit expert in that group Space
Multi-expert support	Multiple experts can coexist in one group Space

Automation

Capability	Notes
Cron triggers	Run tasks on schedules
Calendar triggers	Run tasks based on calendar events
DM/Space delivery	Trigger output can be delivered to DM or group Space

Practical Limits

Tool access depends on configured credentials.
Behavior quality depends heavily on prompt quality.
Large/complex requests may take longer to complete.
Some capabilities are expert-specific, not global defaults.

Code Execution (Sandbox)

Experts can execute Python code in an isolated sandbox environment.

Aspect	Details
Language	Python
Isolation	Fully sandboxed — no access to host system
Use cases	Data processing, calculations, chart generation, file transformation
File handling	Sandbox outputs files back to the agent filesystem; files can be exported as downloadable links
Input files	Sandbox has access to `input_files` passed from the agent workspace

The sandbox is ideal for tasks that need computation beyond what the LLM can do inline — number crunching, data visualization, or transforming file formats.

Memory and Context

Type	How It Works
Conversation history	Recent turns are included in context automatically. The number of turns is bounded by the session history setting.
Long-term memory	Experts use Mem0 to remember user preferences, past decisions, and context across sessions. Memory is per-user.
Knowledge base (RAG)	If a Google Drive folder or knowledge base is configured, relevant documents are retrieved via semantic search and injected into context.

Long-term memory allows experts to personalize over time — remembering your preferred report format, your team’s naming conventions, or decisions from past conversations.

Next Steps

Connecting Tools and Integrations

Configure credentials and integration access.

Expert Fields Reference

See which fields influence capability behavior.

Troubleshooting

Resolve common capability and response issues.

Glossary

Learn key terms used in this page.

​Common Platform Integrations

​Input and Output Types

​Conversation and Collaboration

​Automation

​Practical Limits

​Code Execution (Sandbox)

​Memory and Context

​Next Steps