> This page is for version v8.3.0 (default). > For other versions, use one of these documentation indexes: > - v8.3.0 (default): https://docs.workflows.com/v-8-3-0/llms.txt > For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.workflows.com/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.workflows.com/_mcp/server. # Core Concepts ![Mind Reasoner MCP](/_fern-img/466e07efd0202b28f29ec8dda3be88918c85104c2652efceb8ff717127f861ca.webp) Mind Reasoner has three core concepts: **Minds**, **Snapshots**, and **Simulations**. Master these, and everything else becomes intuitive. > **Info** > > **The basics:** Create a Mind → Upload transcripts → Run Simulations --- ## Minds A **mind** is a container that represents a person. Think of it as a digital profile. #### Overview ### What it does * **Stores conversation transcripts** — Upload and manage training data * **Holds metadata** — Name, creation date, and configuration * **Creates snapshots** — Generate multiple AI models over time as data evolves Minds are the foundation of Mind Reasoner. Each mind represents one person and can have unlimited snapshots trained from different datasets or time periods. #### Example ### Real-world example Create a mind for **"Sarah - Customer Service Manager"** to capture how she communicates with customers. You can then: * Upload her support call transcripts * Train snapshots from different quarters to track style evolution * Run simulations to predict responses to new customer scenarios #### API Actions ### Available operations * `POST /minds` — Create a new mind * `GET /minds` — List all minds * `GET /minds/{id}` — Retrieve mind details * `PATCH /minds/{id}` — Update mind metadata * `DELETE /minds/{id}` — Delete a mind See the [API Reference](/api-reference) for full details. #### MCP Tools ### Using minds with Claude When using the MCP Server, Claude handles mind operations through natural language: **Creating minds:** ``` "Create a new mind for Alex, my sales manager" ``` **Finding existing minds:** ``` "Show me all minds I've created" "Find minds related to customer service" ``` **Getting details:** ``` "What's the status of the Alex mind?" ``` Behind the scenes, Claude uses: * `create_mind` — Create new minds * `get_all_minds` — List all minds * `search_minds` — Search by keywords See the [MCP Quick Start](/mcp-quickstart) to get started. --- ## Snapshots A **snapshot** is a trained AI model built from conversation transcripts. It captures how someone thinks and communicates at a specific point in time. #### How It Works ### Training process #### Upload Data Provide conversation transcripts (.vtt, .pdf, or .docx). Minimum: 10-20 conversations or 5,000+ words. #### AI Training Mind Reasoner analyzes across hundreds of dimensions of psychometrics. Takes 5-15 minutes. #### Ready to Use Once complete, run unlimited simulations without re-training. > **Tip** > > More data = better accuracy. Use recent transcripts (last 6 months) for best results. #### Requirements ### Data requirements **Minimum data:** * 10-20 conversations, OR * 5,000+ words total **Supported formats:** * `.vtt` — Video transcripts (Zoom, Teams, Meet) * `.pdf` — Documents with text * `.docx` — Word documents **Best practices:** * Use recent data (last 6 months) * Include diverse conversation types * Ensure transcripts are clean and accurate * More data improves accuracy #### API Actions ### Available operations * `POST /minds/{id}/snapshots` — Create and train a snapshot * `GET /minds/{id}/snapshots` — List all snapshots for a mind * `GET /snapshots/{id}` — Get snapshot details and status * `DELETE /snapshots/{id}` — Delete a snapshot Training status: `queued` → `processing` → `completed` or `failed` See the [API Reference](/api-reference) for full details. #### MCP Tools ### Training with Claude Claude handles the entire snapshot training workflow automatically through natural language: **Upload and train:** ``` "Upload the transcript at /Users/me/sales-calls.vtt to Alex's mind and start training" ``` **Monitor training:** ``` "Let me know when the snapshot is ready" ``` **Check status:** ``` "What's the training status for Alex's snapshot?" ``` Behind the scenes, Claude uses: * `get_upload_url` — Get secure upload URL * `upload_file` — Upload transcript to S3 * `get_snapshot_status` — Monitor training progress * `get_latest_snapshot` — Get the completed snapshot Claude monitors training automatically and notifies you when complete (typically 5-15 minutes). See the [MCP Quick Start](/mcp-quickstart) for a walkthrough. --- ## Simulations A **simulation** predicts how someone would respond to any scenario based on their trained snapshot. #### Writing Scenarios ### How to write effective scenarios #### Be Specific **Vague:** "How would they handle a complaint?" **Specific:** "A customer received a damaged product 3 days before their event. They need a replacement by tomorrow, but the product is out of stock. How would you respond?" #### Include Context * **Emotional state** — frustrated, confused, urgent * **Time constraints** — by tomorrow, within 24 hours * **Constraints** — out of stock, over budget * **Desired outcome** — resolution, information #### Add Relevant Details * Background information * Stakeholder concerns * Business constraints * Historical context #### Response Details ### What you get back Each simulation returns: * **Predicted response** — How the person would likely respond * **Confidence score** — Model's certainty in the prediction * **Reasoning** — Key factors influencing the response * **Tone analysis** — Emotional and stylistic characteristics Simulations are instant once a snapshot is trained. No re-training needed. #### API Actions ### Available operations * `POST /snapshots/{id}/simulations` — Run a new simulation * `GET /simulations/{id}` — Get simulation results * `GET /snapshots/{id}/simulations` — List all simulations for a snapshot **Parameters:** * `scenario` (required) — The situation to simulate * `context` (optional) — Additional background information * `include_reasoning` (optional) — Include detailed analysis See the [API Reference](/api-reference) for full details. #### MCP Tools ### Simulating with Claude Just ask Claude natural questions, and it handles simulations automatically: **Single scenario:** ``` "How would Alex handle a customer complaint about pricing?" ``` **Detailed scenario:** ``` "A prospect says our product is too expensive compared to competitors. They have a budget of $5k but our solution costs $8k. How would Alex respond?" ``` **Compare responses:** ``` "How would both Alex and Sarah respond to an angry customer whose order was delayed by 2 weeks?" ``` **Batch testing:** ``` "Test how the support team mind would handle these 5 scenarios: [list scenarios]" ``` Behind the scenes, Claude uses: * `run_simulation` — Execute simulations on trained snapshots * Automatically includes relevant context * Formats responses conversationally No manual API calls needed—just describe the scenario in natural language. See the [MCP Quick Start](/mcp-quickstart) for examples. --- ## Quick Start Choose how you want to get started: #### [MCP Server](/mcp-quickstart) Use Claude to create minds with natural language. No code required. #### [Direct API](/api-quick-start) Build custom applications with HTTP REST API. Full programmatic control. --- ## What's Next? #### [Explore Workflows](/workflows) See how organizations integrate Mind Reasoner into business processes #### [MCP Quick Start](/mcp-quickstart) Connect Claude in 5 minutes #### [API Guide](/api-quick-start) Build with the Direct API