v0.0.16

[0.0.16] - 2026-10-10

Added

• Agent: a Memory toggle in each chat turns long-term memory off for that conversation.
• MCP: OAuth sign-in for clients that cannot take a fixed token (claude.ai, ChatGPT connectors), approved on a consent page with per-module permissions.
• Demo: Fundamentals opens with illustrative data on every page (macro, companies, ETFs, calendar, alternative data), clearly marked as made up.
• Mobile: responsive layout, phase 1: navigation, module pages, dashboard widgets and charts adapt to phone screens.
• Backtest: a small notice on the results when the run had no spread or no fees set.
• Security: social sign-in with Google, Microsoft, GitHub or any OpenID Connect provider, locked to one linked account, with recovery codes and the option to turn password sign-in off.
• Agent: per-chat limits on output tokens or dollars, with a default for new chats, a budget bar beside the message box, and a notice offering to raise the limit or start a new chat once it's reached.
• Voice: push-to-talk commands and dictation, chained voice commands confirmed before they run, and a choice of speech engine (browser, self-hosted Whisper, OpenAI, Groq).
• Settings: Clean in Manage data deletes a module's records older than a chosen number of days, weeks, months or years (mail, news, calendar, notifications, runs, chats and more).

Changed

• Backtest: new strategies start with no fees instead of 0.1%.
• Agent: a chat keeps the provider and model it started with, and picking a provider in a chat selects that provider's default model.
• Agent: the API key is optional for a provider, so local servers like Ollama work without one.
• Agent: the write confirmation shows how to skip it next time (approve writes automatically on the persona).

Fixed

• Security: actions that ask for your password again (new MCP token, network mode, credential export, vault, turning 2FA off) work again over HTTPS.