Chat is the AI agent layer of 🏷️#mythos-feature. It gives every library owner a personal 📝Artificial Intelligence (AI) collaborator that can read, write, research, and organize memos on their behalf. The agent operates with full awareness of the owner's library, templates, tags, and augmentation context.
Two Entry Points
Sidebar Chat
Available from the memo feed, the sidebar chat panel lets you have an open-ended conversation with your AI agent. Use it to ask questions about your library, request new memos, search for connections between ideas, or get help organizing your knowledge. The agent has access to your full library via RAG (retrieval-augmented generation) and can search, read, create, and update memos during the conversation. The sidebar chat is also location-aware — when you navigate to a memo, the sidebar automatically knows which memo you're viewing and includes that context in every message, so you can say "this memo" naturally.
Sparkles Button (Memo-Contextual Chat)
When viewing any memo or daily memo, a sparkles button appears in the header next to the voice dictation icon. Clicking it opens a chat modal that is contextually bound to that specific memo. The agent knows exactly which memo you're looking at — its title, ID, and content — and will default to updating that memo when you give instructions like "write my reference memo on this" or "research this and fill in the details." This is the primary way to direct the agent to do work on a specific memo.
What the Agent Can Do
- Create memos — with automatic template matching based on tags
- Update memos — read existing content first, then merge additions (never overwrites)
- Search memos — find related memos by title or tags across your library
- Read memos — pull the full content of any memo for context
- Web search — research topics, look up URLs, gather information to populate memos
- Manage templates — list, create, or update your memo templates and their agent instructions
- Create collections — set up tag collections with explainer memos
Context the Agent Has
Every chat session loads four augmentation memos that shape how the agent understands and works with you:
- Mythos Soul — the relationship frame: your collaboration values, communication philosophy, and how you and your agent operate together
- Mythos Style — formatting rules, mention conventions, tone, tag usage, and writing patterns
- Mythos Human — who you are: your role, expertise, current projects, and goals
- Mythos Memory — what the collaboration has learned: confirmed approaches, corrections, and preferences
These are editable memos in your library (tagged #mythos-augmentation). When you update them, the agent adapts immediately. Beyond augmentation, the agent also has access to your templates (with their agent instructions), your tag library with usage counts, and collection explainer memos that describe what each tag means.
When using the sparkles button, the agent additionally has the specific memo's title, ID, and content preview injected at the top of its instructions, so it knows exactly what you're referring to.
Configuration
- Settings > AI — choose your default provider (OpenAI, Anthropic, Google AI, MiniMax, Mistral, Moonshot) and model, manage API keys
- Settings > Chat — customize the system prompt that shapes your agent's personality, set model overrides, configure rate limits and budget caps, manage conversation starters
- Settings > Augmentation — edit your Soul, Style, Human, and Memory memos directly
How It Works (Technical)
The chat system uses a streaming architecture. For owners, all tool calls (memo creation, web search, etc.) execute server-side before the response streams back. This means you can close the chat modal or navigate away after giving the agent a command — any memo reads, web searches, and writes the agent initiated will still complete on the server. You lose visibility into the response text and status updates, but the work finishes. The agent can make up to 5 sequential tool calls per message, chaining research → reading → writing in a single interaction.