> ## Documentation Index
> Fetch the complete documentation index at: https://docs.ospribrain.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Brain Overview
> Your AI assistant that knows everything about your meetings, documents, and company context.
## What is the Brain?
The Brain is Ospri's AI assistant, powered by GPT-5.3. It has access to your entire organizational context — meeting transcripts, vault documents, CRM data, to-dos, assay catalogs, and external research. Unlike a generic chatbot, the Brain understands your company, your role, and the specific work you're doing.
Every conversation follows a structured approach: the Brain first creates a **plan** of research steps, then executes them (searching your vault, querying meetings, browsing the web, searching academic papers), and finally synthesizes a comprehensive response with citations.
## Two Chat Interfaces
Ospri Brain has two chat interfaces, each suited for different workflows:
### Simple Chat (Home Page)
The home page (`/`) provides a streamlined chat experience:
* Quick question-and-answer with the Brain
* File uploads via the paperclip or `/` shortcut
* Reasoning effort and search depth controls
* Auto-saves conversations to the sidebar
### Advanced Chat (`/chat`)
The dedicated chat page (`/chat`) adds:
* **Plan visualization** — See the Brain's research plan with step-by-step progress
* **Document panel** — AI-generated reports render in a side panel
* **Spreadsheet panel** — Generated Excel/CSV files display inline
* **Code execution** — The Brain can run Python code and generate charts
* **Computer Use** BETA — Browse live websites, take screenshots, extract data, and interact with web pages
* **File analysis** — Uploaded files are directly analyzed (attachments are sent to the API)
* **Voice mode** — Real-time voice conversations
* **Meeting context** — Persistent meeting context across turns
* **Project context** — Folder goal and background automatically injected
For quick questions, use the home page. For deep research, report generation, or file analysis, use `/chat` for the full experience.
## The Chat Interface
### Welcome Screen
When you start a new chat, you see a greeting and four suggestion chips:
| Chip | What It Does |
| ---------------------------------------------- | ------------------------------------------------------------- |
| "Summarize recent meeting insights" | Searches your recent meeting transcripts for key takeaways |
| "What are our top revenue risks this quarter?" | Analyzes CRM deals and meeting discussions for risk signals |
| "Analyze the latest external signals" | Pulls from Signals intelligence for market/regulatory updates |
| "Generate a compliance summary" | Searches vault documents and audit logs for compliance status |
Click any chip to instantly start a conversation with that prompt.
### Message Display
* **Your messages** — Right-aligned with a primary color background
* **Brain responses** — Left-aligned in a card with a border
* **Markdown** — Full support including headers, bold, italic, lists, tables, and code blocks with syntax highlighting (Prism, VS Code Dark+ theme)
* **Generated files** — Links to `/storage/v1/object/public/generated-files/` render as download buttons with file icons
* **Attachments** — Displayed as clickable chips with filenames; click to download
### Input Area
| Element | Description |
| -------------------- | --------------------------------------------------------------------- |
| **Text area** | "Ask anything..." placeholder. Enter to send, Shift+Enter for newline |
| **Paperclip** | Opens the file upload modal |
| **Reasoning effort** | Dropdown: Low / Med / High |
| **Search depth** | Dropdown: Basic / Deep |
| **Model badge** | Shows "Jules" (the Brain's model name) |
### Chat Header
Once a conversation is saved (after first exchange), a header appears:
| Button | Action |
| -------------- | ------------------------------------------------------------------------ |
| **Chat title** | Auto-generated (3-6 words, topic-specific) or manually renamed |
| **Share** | Marks conversation as shared; copies `/share/{chatId}` link to clipboard |
| **Delete** | Permanently removes the conversation and all messages |
## Reasoning Effort
Controls how deeply the AI reasons about your question:
| Level | Best For | Behavior |
| -------- | ----------------------------------- | ------------------------------------------------------- |
| **Low** | Quick factual lookups | Minimal reasoning, fast responses |
| **Med** | Most everyday questions (default) | Balanced reasoning and speed |
| **High** | Complex analysis, report generation | Deep multi-step reasoning, uses OpenAI reasoning tokens |
## Search Depth
Controls how extensively the Brain searches for information:
| Level | Behavior |
| --------- | ----------------------------------------------------------- |
| **Basic** | Searches your vault (Pinecone) and meeting transcripts only |
| **Deep** | Also searches the web via Tavily for current information |
For questions about your own company data, **Basic** search is faster and usually sufficient. Use **Deep** when you need external market data, competitor info, or regulatory updates.
## How the Brain Finds Answers
When you ask a question, the Brain follows this structured process:
The Brain calls `create_plan()` first, choosing a mode (**base\_analysis**, **competitive\_analysis**, or **writing**) and generating 2-9 research steps depending on complexity.
Each step uses a specific tool: knowledge base search, web search, meeting intel, paper search, etc. Progress is streamed in real-time.
Semantic search across all Brain-accessible vaults using Pinecone with `text-embedding-3-large` (1024 dimensions). Searches up to 5 namespaces simultaneously, returns top 10 results.
Searches recent meeting transcripts for relevant discussions, decisions, and action items with verbatim quotes when relevant.
May search the web (Tavily), query academic papers (Semantic Scholar + PubMed), search your assay catalog, or create to-dos.
Combines all sources into a coherent answer with citations, links to original documents, and actionable recommendations.
### Plan Modes
| Mode | When Used | Typical Steps |
| ------------------------- | ------------------------------------------------ | ----------------------------------- |
| **base\_analysis** | Research, science, market questions, trends | 2-3 (simple) to 6-9 (deep research) |
| **competitive\_analysis** | Competitor questions, positioning, benchmarking | 4-6 steps |
| **writing** | Reports, emails, proposals, summaries, documents | 3-5 steps |
## Streaming Events
On the advanced chat page (`/chat`), you can see the Brain's work in real-time. These events appear as the Brain processes your request:
| Event | What You See |
| ------------------- | ------------------------------------------------------------------------------- |
| **Plan** | The research plan with numbered steps |
| **Step start/done** | Progress indicators as each step executes |
| **Sources** | Web search results with URLs |
| **Meetings** | Meeting intel results with titles and excerpts |
| **Papers** | Academic paper citations with titles, authors, abstracts |
| **Document** | A side panel opens with a streaming report/document |
| **Chart** | An inline chart or image generated by code execution |
| **Screenshot** | A page capture from Computer Use (browser automation) |
| **Spreadsheet** | An inline spreadsheet viewer for generated data |
| **Clarification** | The Brain asks you a clarifying question with optional quick-answer suggestions |
## Clarification Mode
If your question is ambiguous, the Brain may call `ask_clarification` instead of proceeding:
* A clarification card appears with the Brain's question
* 2-4 quick-answer suggestions are provided as clickable chips
* You can click a suggestion or type a custom response
* The Brain then continues with your clarification
The Brain only asks for clarification when the question is genuinely ambiguous. It won't ask unnecessary follow-ups.
## Inline Documents
When the Brain generates a report or document (via the `create_document` tool), it opens in a side panel:
1. The **document panel** opens alongside the chat
2. Content streams in real-time with Markdown formatting
3. The document is rendered with a title and full formatting
4. Generated files (CSV, XLSX, images) appear as download links
## Generated Files
The Brain can generate various file types during a conversation:
| File Type | How It Appears |
| --------------------------------- | ------------------------------------------------------ |
| **Charts/Images** (PNG, JPG, SVG) | Rendered inline as images |
| **CSV/TSV** | Downloadable with inline preview |
| **Excel** (XLSX) | Downloadable spreadsheet file |
| **Text files** | Inline content with download option |
| **Financial models** | `financial_model.xlsx` + `data.csv` generated together |
## Computer Use BETA
The Brain can browse live websites on your behalf. When you ask it to visit a URL, it launches a real headless browser, navigates the page, and interacts with it step by step — taking screenshots, clicking elements, filling forms, and extracting data.
**What you can do:**
* Take screenshots of any website and see them inline in the chat
* Extract structured data from web pages (pricing tables, contact info, product specs)
* Navigate multi-page workflows (click through menus, follow links)
* Monitor competitor websites or regulatory pages for updates
* Check that your own web properties look correct
**How it appears in the chat:**
The Brain shows a "Browser" card with:
* A blue monitor icon and the domain name (e.g., "Browser: example.com")
* Page screenshots displayed inline after each action
* A collapsible log of actions taken (navigate, click, extract, etc.)
* Extracted data formatted as tables or JSON
**Use cases:**
| Use Case | Example Prompt |
| --------------------- | --------------------------------------------------------------------- |
| Competitor monitoring | "Go to competitor.com/pricing and extract their pricing tiers" |
| Regulatory tracking | "Visit the MPI biosecurity page and screenshot any new notices" |
| Conference scouting | "Browse the ISVEE conference site and extract the speaker list" |
| Vendor research | "Go to this supplier's website and extract their PCR machine catalog" |
| Visual QA | "Take a screenshot of ospribio.com — does the homepage look right?" |
| Data extraction | "Visit this page and extract all the data into a CSV" |
Computer Use works with any publicly accessible website — no domain allowlisting or configuration needed. For sites that require login, the agent will ask for your confirmation before entering credentials.
## Context Injection
The Brain receives context from multiple sources for every conversation:
| Context Source | What It Provides |
| -------------------- | ------------------------------------------------------------------------- |
| **Personal Context** | Your role, industry, objectives, pain points, preferences (from Settings) |
| **Project Context** | Goal and background from the project folder (when chat is in a project) |
| **Meeting Context** | Full transcript and analysis from linked meetings |
| **Vault Documents** | Relevant document excerpts found via semantic search |
| **Today's Date** | Current date for time-aware responses |