-
Notifications
You must be signed in to change notification settings - Fork 0
Future Ideas
This page is a brainstorming collection of ideas that could extend Mosaic. These are not commitments — they're possibilities to inspire contributors and spark discussion. Ideas range from small quality-of-life tweaks to major architectural changes.
Feel free to open a GitHub issue if any of these excite you, or add your own ideas!
Side-by-side mode to compare two canvases or two branches visually, with synchronized scrolling and diff highlighting.
Save a node's system prompt, temperature, and model as a reusable template. "New node from template" option in the context menu.
Click on the mini-map to jump to that location. Drag the mini-map viewport to pan the canvas.
Custom background images or colors per canvas. Useful for visual organization (e.g., green background for "done", yellow for "in progress").
User-assignable colors and custom labels on nodes for visual categorization. Filter by color or label.
Small text labels on edges to describe the relationship (e.g., "refines", "critiques", "expands on").
One-click buttons to collapse or expand every branch on the canvas simultaneously.
Save snapshots of the canvas at regular intervals, with the ability to browse and restore previous versions.
A panel showing canvas size, node count, token usage, and warnings (e.g., "This canvas has 300 nodes — consider splitting").
Drag a node from one canvas tab to another to copy or move it between conversations.
Visual containers (boxes with titles) that group related nodes together. Collapse/expand the entire group.
Expand undo to cover text edits, bookmark toggles, collapse state changes — not just position changes.
Temporarily dim everything except the active branch, reducing visual noise.
Bundle a lightweight local LLM (like Llama 3.2 1B via WASM or Tauri sidecar) so Mosaic works fully offline without any API key.
The AI autonomously runs code, reads the output, and continues refining without user intervention — useful for data analysis tasks.
An AI-generated summary of the entire canvas or a specific branch, displayed as a sticky note overlay.
Let the AI suggest names for each branch based on its content, displayed as a small label on the branch's first node.
Export conversations as Markdown, HTML, or PDF with timestamps, model info, and confidence scores.
Let the AI call external tools (web search, calculator, code execution) as part of its response. Structured tool definitions.
Pre-built system prompts for common personas (tutor, code reviewer, creative writer, debate opponent) that users can browse and apply.
Expose advanced parameters: top-p, top-k, frequency penalty, presence penalty, stop sequences, response format (JSON mode).
Use {{variable}} syntax in system prompts that get replaced with canvas metadata (node count, total tokens, date, etc.).
Image input/output, audio transcription, file generation — leveraging provider vision and audio APIs.
Gradually decrease temperature across a branch so early responses are creative and later ones converge to a refined answer.
Auto-detect chart generation (matplotlib, plotly) and render it inline as an image, not just text output.
Let users create small code projects with multiple files. Run, test, and share them within a node.
Render simple UI widgets (sliders, buttons, text inputs) from code output that can modify variables and re-execute.
A virtual in-memory file system within the code sandbox so files written in one execution persist to the next.
Allow running any WebAssembly module, not just Pyodide — Rust, C, Go compiled to WASM.
A "share this code" button that creates a permalink to a gist or pastebin with the code and output.
Browse available packages within the sandbox, search for specific functions, and see documentation without leaving Mosaic.
Multiple users editing the same code block simultaneously (Operational Transform / CRDT).
Built-in SQLite in WASM for querying CSV/JSON data with SQL directly in chat.
A sandboxed shell (like WebContainers) for running shell commands, npm install, git operations — all in WASM.
Paste a URL and have Mosaic fetch, parse, and chunk the web page content for RAG — no manual save-and-upload needed.
Extract text from scanned PDFs and images using OCR (tesseract in WASM or via provider vision API).
Remember which chunks were used in previous turns and avoid repeating them. Track conversation context with document references.
When the AI uses RAG context, highlight which source document and chunk each sentence came from.
AI-generated tags for uploaded documents (e.g., "technical", "finance", "code") for better organization.
Build a graph of concepts extracted from documents, showing relationships between ideas across documents.
Show previous questions and answers for each document, so users can see what insights have already been extracted.
Combine keyword (BM25/TF-IDF) and semantic (embedding) scores with configurable weights.
Show how chunks overlap in the original document, with the ability to adjust chunk size and overlap interactively.
When a document is updated, only re-index the changed chunks instead of the entire document.
Multiple users editing the same canvas simultaneously with cursor presence, using CRDTs or WebSocket sync.
Generate a read-only or editable link to a canvas. Hosted via a lightweight sync server or peer-to-peer.
Add comments to specific nodes. Threaded discussions on individual messages.
Propose merging one branch into another, with a diff view and approval workflow.
A place to share and discover interesting canvases — templates, research trees, learning paths.
See who changed what and when on collaborative canvases.
Embed a read-only canvas view in a web page or blog post.
Tag other users in comments or nodes. Get notified when someone replies or mentions you.
A Progressive Web App version that runs in the browser without Tauri, with localStorage persistence and optional cloud sync.
Touch-optimized mobile interface for viewing canvases, reading conversations, and basic branching.
Open Mosaic canvases from within VS Code. Use selection context to automatically populate questions.
Import/export canvases to/from Obsidian. Link notes to conversation nodes.
Share nodes or entire branches to Slack or Discord channels.
Link commits, PRs, and issues to conversation nodes. AI-assisted code review with context from related canvases.
A command-line interface for exporting, searching, and managing canvases from the terminal.
Minimize to system tray with quick-access menu for recent canvases and new chat.
Search across all canvases simultaneously, not just the current one.
AI tags canvases based on content for automatic organization. Filter canvas list by tag.
Get notified when a streaming response completes, or when a long code execution finishes.
See total token usage, cost, and model distribution across all canvases over time.
AI-generated insights about your canvas: "You branch most frequently from Mistral Large responses", "Your average branch depth is 4.2 nodes".
Set a daily/weekly token budget and get warned when approaching it.
Export a visual graph (PNG/SVG) of the canvas, suitable for presentations or documentation.
Export canvas data as JSON, Markdown, HTML, CSV (flat node list), or OPML (outline format).
Track how canvas size, depth, and branch count change over time.
Log how long each provider takes to respond and surface performance trends.
Subtle audio cues for stream completion, errors, and new suggestions. Respects prefers-reduced-motion.
A clean, print-friendly view of a branch or entire canvas, stripping away UI chrome.
Automatically scroll to keep the currently streaming node in view.
Drag a node to change its position in the conversation order (not just spatial position).
Let power users inject custom CSS to personalize the appearance beyond the built-in themes.
An interactive walkthrough that highlights UI elements and explains features step by step.
Let users remap keyboard shortcuts to their preference.
Group canvases into workspaces (e.g., "Work", "Personal", "Research") with independent settings and themes.
Mark a canvas as "snoozed" — it disappears from the main view but can be unsnoozed later.
Pin frequently used canvases to the top of the tab bar.
Filter search by node type, date range, bookmarked status, or provider.
Save canvases as a declarative format (YAML/TOML) that can be version-controlled, diffed, and reconstructed.
When multiple models respond, let users vote on which answer is best. Aggregate votes across canvases to rank models by task type.
A slider that shows the canvas state at different points in time — scrub forward and backward through the conversation.
Predict which nodes the user is likely to click next based on their history. Pre-fetch the AI response for instant display.
Learn which follow-up suggestions users actually click and improve suggestion quality over time.
Visual overlay showing which branches were most explored, most collapsed, or most bookmarked.
Hide which AI model generated a response and let the user guess — a gamified way to compare providers.
Upload a meeting transcript and have Mosaic automatically build a conversation tree from the discussion flow.
Let the AI wander autonomously — generate chains of self-directed thought across the canvas, branching on its own curiosities.
Render the canvas as a 3D spiral or radial tree with WebGL, with smooth camera fly-throughs.
- Check the existing issues on GitHub Issues to avoid duplicates
-
Open a new issue with the
idealabel - Describe the idea: What problem does it solve? How might it work? What would it look like?
- Discuss and refine: Engage with the community to shape the idea
The best ideas come with use cases — describing a real problem you've encountered makes it much more compelling.
- Changelog and Roadmap — What's actually planned for upcoming releases
- Contributing Guide — How to help implement these ideas
- Home — Back to the wiki home
Mosaic — Branch, explore, and run code inline — an infinite canvas for AI conversations.
Built with Tauri, React, and Mistral AI.
GitHub Repository |
Report an Issue |
Releases
- Canvas and Node System
- LLM Provider Integration
- RAG System Guide
- Advanced AI Features
- Keyboard Shortcuts and UI Reference
- Tutorial - Branching and Parallel Exploration
- Tutorial - Using RAG with Documents
- Tutorial - Advanced Features in Practice
- Tutorial - Customizing Mosaic