Built-in Tools

​

Overview

The Brain has 13 specialized tools it can invoke automatically. You never need to tell it which tool to use — the Brain’s planner selects tools based on your question. On the advanced chat page (/chat), you can see which tools are being used in the plan visualization.

​

Tool Reference

​

1. create_plan

Always called first. Creates a structured research plan before answering. Available tools for steps: web_search, knowledge_base_search, meeting_intel, reasoning, synthesis, writing, create_document, paper_search, list_todos, create_todo, search_assay_catalog, analyze_uploaded_file, add_to_knowledge_vault, run_analysis, create_chart, build_financial_model, computer_use

​

2. knowledge_base_search

Semantic search across your vault documents using Pinecone. How it works:

• Embeds query using text-embedding-3-large (1024 dimensions)
• Searches up to 5 Pinecone namespaces (one per vault: default, sales, science, marketing, etc.)
• Returns top 5 results per namespace, merged and sorted by score
• Final output: top 10 most relevant passages with file names and content

Example prompts that trigger this:

• “What does our SOP say about sample handling?”
• “Find the pricing we quoted to McKesson”
• “What were the key findings in our validation study?“

​

3. web_search

Searches the web using Tavily for current information. Only available when Search Depth is set to Deep. Returns web results with titles, URLs, and content excerpts. Example prompts:

• “What is the latest FDA guidance on LDTs?”
• “Current reimbursement rates for CPT 87635”
• “What did competitor X announce last week?“

​

4. meeting_intel

Searches your meeting transcripts and analysis reports. Only available when you have recorded meetings. Returns meeting titles, dates, relevant transcript excerpts, and speaker attributions. Example prompts:

• “What did we decide about the multiplex panel?”
• “What action items came out of the board meeting?”
• “Has anyone discussed pricing changes in recent calls?“

​

5. paper_search

Searches academic literature via Semantic Scholar and PubMed. Returns papers with: title, authors, year, citation count, abstract, DOI/URL. Example prompts:

• “Find recent papers on CRISPR diagnostics for respiratory viruses”
• “What is the latest research on Ct value interpretation?”
• “Studies comparing RT-PCR vs isothermal amplification”

​

6. create_todo

Creates tasks directly in your To-Dos. Example prompts:

• “Create a to-do to follow up with Greg by Friday”
• “Add a high-priority task: Review the McKesson contract”
• “Make action items from the last meeting into to-dos”

​

7. list_todos

Retrieves your to-do list with filtering. Example prompts:

• “What is on my to-do list?”
• “Show me overdue tasks”
• “What high-priority items do I have this week?“

​

8. search_assay_catalog

Searches your diagnostic assay panel catalog. Example prompts:

• “Do we have an assay for RSV?”
• “What panels include the N gene target?”
• “Match these targets: RSV-A, RSV-B, SARS-CoV-2”

​

9. create_document

Generates a formatted report or document in the side panel. The document streams into a side panel with full Markdown formatting. Used for reports, summaries, proposals, and other long-form content. Example prompts:

• “Generate a meeting summary report”
• “Create a competitive analysis document”
• “Draft a validation protocol outline”

​

10. analyze_uploaded_file

Analyzes files you have uploaded to the conversation. Uses GPT-4o with input_file blocks for direct file analysis. Example prompts (with file attached):

• “Summarize this contract”
• “What are the main findings in this paper?”
• “Extract all dates and deadlines”
• “Read the text in this image”

​

11. add_to_knowledge_vault

Stores uploaded files permanently in your Knowledge Vault. Files are stored in Supabase storage, then ingested and indexed for future Brain searches. Example prompts (with file attached):

• “Add this to the Science vault”
• “Store all uploaded files in the Sales vault”

​

12. computer_use BETA

Browses live websites, takes screenshots, clicks elements, fills forms, and extracts content using a headless Chromium browser powered by Browserbase. How it works:

1. A Browserbase session launches a remote Chromium browser
2. A vision agent (GPT-5.4) sees page screenshots and decides actions step by step
3. Each action (navigate, click, type, scroll, extract) executes via Playwright
4. Screenshots are captured after every action and displayed inline in the chat
5. Extracted data is returned as structured JSON

The agent can handle multi-step interactions: navigating between pages, clicking menus, filling forms, scrolling to load content, and extracting tables or text. If an action fails, it automatically tries alternative approaches. Example prompts:

• “Go to https://example.com and take a screenshot”
• “Visit our competitor’s pricing page and extract their tiers into a table”
• “Browse to the FDA guidance page and screenshot the latest updates”
• “Go to the conference website and extract the speaker list and schedule”
• “Take a screenshot of https://ospribio.com — does the homepage look correct?”

​

13. ask_clarification

Asks you a clarifying question when your request is ambiguous. This pauses the Brain’s response and waits for your answer. You can click a suggestion or type a custom response.

​

Tool Chaining

The Brain often chains multiple tools in a single response. For example:

1. You ask: “Compare our RSV assay with published literature and create a report”
2. Brain creates a plan with 5 steps
3. Step 1: search_assay_catalog (mode: match_targets, targets: [“RSV”])
4. Step 2: paper_search (query: “RSV diagnostic assay performance”)
5. Step 3: knowledge_base_search (query: “RSV validation data”)
6. Step 4: reasoning (synthesize comparison)
7. Step 5: create_document (title: “RSV Assay Comparison Report”)