Intelligence
Intelligence chat and AI Live voice share durable history, bounded recall, saved memories, journal context, product and market tools, live-web verification, insights, and feedback.
5 min read
Chats
The app stores typed and voice turns in the same durable chats, allows renaming and deleting them, and supports both normal chat and streaming progress updates. Ask “Give me a summary of what happened in this chat” to recap the full chat, including older turns outside the normal recent-message window.
Pin an answer
Choose Pin answer below a saved reply. Open Library → Pinned to find it and jump back to the original reply. Unpin from the chat or Library. Deleting the source answer also removes its pin. Pinning is available in saved, regular chats.
Projects
Open Projects to group conversations and give them standing instructions. Start a chat from a project or choose Move to project in a chat’s history menu. Typed chat and Live voice use those instructions for filed conversations. Removing a conversation from a project, or deleting the project, returns the conversation to your history.
Composer controls
Choose Quick for a focused answer or Deep research for more room to analyze. Customize the assistant to set a response style or switch memory recall. Type / to open shortcuts for memory settings, usage, a fresh chat, and help. Starting a fresh chat leaves the previous conversation in history.
Model and reasoning
Open More model options from Customize the assistant to set reasoning effort for new typed chat turns. Default follows the chat speed, while Low, Medium and High set the effort. GPT-6 Luna is the only selectable personal model. Connect your OpenAI account to use it for typed chat; OpenAI bills that account directly. Server default follows your plan’s managed routing. Older personal model selections cannot start a new chat until you choose Luna or Server default. You can remove an older encrypted connection from Model settings. Your Intelligence data sharing choice still applies.
Custom instructions
Open Settings → Intelligence to write a standing instruction for chat and Live voice. Save to apply it. Clear and save to remove it. If another device saves while you are editing, compare its saved text with your draft before choosing what to keep.
Data sharing
Review what AI providers receive and choose Allow and continue from your first chat, dictation or Live voice request. Journal writing suggestions can ask for the same permission. You can turn sharing off in Settings → Intelligence at any time. Turning it off stops new requests through those features and ends active voice calls. Saved conversations stay available.
Connected clients and activity
Open Settings → Intelligence to review connected apps, find the Developer API address, and view recorded requests from your API keys and connected apps. Filter by outcome or load older requests. Developer settings handles creating, revoking and disconnecting credentials.
Connect with MCP
Copy the MCP server address from Connected clients. Use the client’s published identity or choose automatic registration if prompted, then choose its read and action permissions on the TheProfitPath consent page. Clients that require a pre-registered ID can use a Developer app with their callback URL and scopes. For manual read-only setup, replace YOUR_KEY in the Claude Code command with a scoped Developer API key. Tool calls and their outcomes appear in Activity. Disconnect OAuth access or revoke a key in Developer settings.
Review MCP actions
A connected client with action permission can propose appending a daily journal review or moving a trade to recoverable Trash. Open Action approvals to review the exact change. Approve a journal append once, allow future appends from that connection, or decline. Every move to Trash requires approval. Remove a standing permission from Allowed without asking. Activity shows which requests asked first.
Attach charts and documents
Attach up to four images, CSVs or PDFs to a message. CSVs and PDFs can be up to 1 MB each. Intelligence reads a bounded excerpt and reports which rows or pages were included. Scanned PDFs need a text layer or page screenshots. Download a saved document from its chat or from Library. Temporary chat documents stay out of Library.
Evidence tools in chat and voice
Typed chat and Live voice can use the same guarded read tools for trades, positions, tracking, analytics, insights, notes, emotions, goals, accounts, playbooks, backtests, reports, achievements, Community, Mentor Mode, notifications, workspace metadata, Economic Calendar, Market News, Market Monitor analysis, product documentation, and current web facts. The assistant must complete required lookups before making user-specific or time-sensitive claims.
Memory controls
Open Settings → Intelligence to turn automatic saving or proactive recall on or off, search memories, correct their wording, pin important context, forget one item, or clear everything. You can also say “Remember that…” or “Forget…” in either chat or voice. Security credentials, payment secrets, and incidental chatter are excluded from automatic memory.
Deletion and provenance
Saved memories retain a link to their source conversation. Forgetting a memory leaves its transcript intact; deleting a chat removes its transcript, chat summary, searchable episodes, queued indexing work, and any memory supported only by that chat.
Learning status
Intelligence learns from your trades and the feedback you give its answers, so its coaching adapts to you over time.
Insights
Insights surface recurring themes and can be marked read individually or in bulk.
Feedback loop
Message feedback is tracked so the system can measure what worked and what failed in the AI responses.
AI sections
Intelligence and the learning surfaces behind it.
Did this article answer your question?