Core Experience
AI Chat& Interaction Modes
Five specialized modes for every workflow, from writing code to architecting systems. Switch modes instantly to get exactly the kind of help you need.
Clean, Focused Interface
Start every session with a distraction-free workspace. Suggested prompt cards let you jump into common tasks instantly, while the full keyboard shortcuts panel keeps power users productive.
- Suggested prompts tailored to your active persona
- Full keyboard shortcut reference built in
- Activity bar for instant panel switching
A Calm Reading Experience
Replies read like a well-set page, not a chat log. Text keeps a comfortable reading width while code blocks and tables use the full column. Headings, quotes, and links sit quietly in the conversation, and each reply carries a small action row with copy, read aloud, and the time it was sent.
- Comfortable reading width for long answers
- Restyled code blocks, tables, and quotes
- Copy, read aloud, and timestamp on every reply
- Animations follow your reduce-motion preference
Five Specialized Modes
Switch between Code, Plan, Ask, Auto, and Orchestrator modes. Each mode tailors the AI's behavior: Plan mode explores read-only and writes one implementation plan you approve, Code mode writes the implementation, Ask mode explains concepts.
- Code: Write and edit code directly
- Plan: Explore read-only in parallel, then approve an implementation plan
- Ask: Get explanations and learn concepts
- Auto: Let the AI pick the best approach
- Orchestrator: Delegate to specialized sub-agents
Transparent Thinking Process
Watch the AI reason through problems in real time. On a multi-step request, its narration, thinking, and tool activity gather in a single working line above the reply, ticking with the current step. Click it to read the full trail; the finished answer arrives as its own message, and anything it produced along the way (reports, file edits, agent cards) stays in the chat.
- One working line per request, with duration and step count when it finishes
- Expand the line to read the full trail: notes, thinking, and each tool step
- Tool Activity panel tracks every action
- Context usage gauge with exact tokens and cost on hover
Session History & Management
Every conversation is automatically saved with its mode, token usage, and full context. Each one names itself after the first exchange, so the sidebar reads like a list of topics instead of opening lines. Filter by mode, search across sessions, and resume any conversation instantly. Run more than one at a time: start a new session while another is still replying, and stop a background session from the sidebar. The ones you are working in sit in a tab strip above the chat, so switching is a single click and each tab keeps its own draft, mode, and scroll position.
- Sessions name themselves, and a name you set by hand always sticks
- Run multiple sessions in parallel, each with its own live status
- Tabs above the chat for the conversations you have open, with Ctrl+1 to 9 and Ctrl+Tab
- A tab shows when its conversation is working, finished, or waiting on you
- Closing a tab leaves the conversation running in the background
- Automatic session persistence with full context
- Filter by AI mode (Code, Plan, Ask, Auto)
- Token and cost tracking per session
- Export sessions as JSON for external analysis
Visual Replay & Live Replay
Right-click any session to access powerful replay features. Visual Replay plays back the conversation like a recording. Watch every message, tool call, and code change unfold at adjustable speed (0.5x to 8x). Live Replay re-executes the session against the AI, letting you see how a different model or updated codebase would handle the same conversation.
- Visual Replay: watch the entire session play back like a video
- Live Replay: re-run the session against the AI with current context
- Adjustable playback speed from 0.5x to 8x
- Rename sessions for easy organization
- Export full session data as JSON
- Per-session cost and token tracking
Behavior Settings & Context Control
Fine-tune the AI's behavior per-session from the input area menu. Toggle validation, retries, summaries, and follow-up questions. Set the maximum autonomous cycles for Auto mode and control reasoning depth. Changes apply immediately and persist across sessions.
- Skip Validation: bypass quality checks for faster responses
- Skip Retry: don't retry on errors. Useful for quick iterations
- Enable Summary: include structured summaries in responses
- Skip Custom Instructions: temporarily ignore .plexon/prompts/ files
- Disable Questions: prevent the AI from asking follow-up questions
- Max Cycles: set the limit for autonomous development loops (default 5)
- Reasoning Effort: set thinking depth (Low / Medium / High) or turn it Off on providers you bring a key for. On Plexon models the depth is set for you, per task, so Plan mode can think harder than everyday chat
Condense Context
Running out of context window? The Condense button next to the context gauge summarizes older messages to free up space, at any point in the conversation, letting you continue long conversations without starting over. The AI intelligently compresses the conversation history while preserving the key decisions, code changes, and context that matter. You can undo a condense at any time to restore the full history.
- One-click context compression, available at any usage level
- AI-powered summarization preserves key decisions and code context
- Shows before/after token counts (e.g. 180k → 45k tokens)
- Undo condense to restore full conversation history
- Configurable auto-condense threshold in Settings → Tools
More Chat Capabilities
Autonomous Development
Auto mode runs up to 20 development cycles with quality validation (configurable threshold, default 70%), test execution, and self-correction loops. Set the goal and walk away.
Visual Understanding
Paste screenshots, photos, or diagrams for AI analysis. Image and video generation available through providers that support it.
Prompt Enhancement
AI-powered prompt rewriter that transforms simple requests into comprehensive, well-structured prompts. Adds context, clarity, architectural guidance, and edge case coverage automatically.
Zero-Knowledge Architecture
Your code never leaves your machine. The stateless server generates responses and forgets everything immediately. AES-256-GCM encryption for all local session data.
Connects to
Nothing here works alone. See the whole map for how the pieces fit together.
This one is taught in step 2 of 8: have a real conversation.
What feeds it
App Connectors feeds this.
Turn one on with your own account and the assistant works inside that app.
API Connections feeds this.
Import an OpenAPI spec or a Postman collection and every endpoint becomes a tool the assistant can call.
Data Sources feeds this.
A built-in tool reads a source from any prompt, and you can @-mention a source or exact rows.
Document Reader feeds this.
A Word doc, spreadsheet, deck, PDF, EPUB or Outlook email arrives as Markdown you can ask about.
Embedded Browser feeds this.
Point at any element on your dev server and the selector, HTML, computed styles and a cropped shot land in the prompt.
Visual & Voice feeds this.
Speak instead of typing; local Whisper transcription becomes the message.
Shell Commands feeds this.
A ! command's output renders as a terminal card, joins the session, and rides into context on your next message.
Plan Mode & Saved Plans feeds this.
Approve a plan and the same conversation switches to Code mode with its checklist loaded.
Spec-Driven Mode extends this.
A sixth mode, driven by living markdown contracts a verifier checks at every step.
Quality Harness extends this.
Turn it on and the model writes a spec before multi-step work and files a verdict after.
AutoDream Memory Consolidation feeds this.
Rules distilled from your thumbs-down feedback are followed afterwards.
Knowledge Graph feeds this.
Ask about a client and every fact comes back with the note and the line it was read from.
Side Questions extends this.
Ask @btw about the conversation while the assistant keeps working; the answer never enters the session history.
Office Integration feeds this.
Read and edit a real .docx or .xlsx from the conversation, with tracked changes while Word is open.
Plexon Provider feeds this.
One subscription covers text, images, video, voice and music, even inside a single prompt.
Custom Providers feeds this.
Point Plexon at any OpenAI- or Anthropic-compatible endpoint, including a local model.
Fireworks AI feeds this.
Open-weight models over the Anthropic Messages protocol, so cached prompts and tool calls arrive intact.
Session Lock & App Lock (Premium) extends this.
The same conversation, encrypted under a passcode only you hold.
Where it goes
This renders into Dynamic UI Widgets.
Structured answers render as cards, tables, timelines and forms instead of a wall of markdown.
This renders into Telegram Bot.
Every session is reachable from your phone.
This renders into Slack Bot.
DM the bot or @mention it in a channel.
This triggers Dynamic Workflows.
A finished reply or a tool call is an event a flow can watch for, matched on a phrase you choose.
This authors Dynamic Workflows.
Describe a flow in chat and the AI drafts it; it lands switched off, with a review link.
This authors Dashboards.
Draft a board in chat, then keep editing it by describing the change.
This authors Skills & Tools.
Author a skill from the conversation, through the plexon_skill tool.
This authors 156 AI Agents.
Author a sub-agent from the conversation, through the plexon_agent tool.
This authors Personas.
The persona wizard drafts the description and welcome prompts for you.
This authors Customization.
Describe a setting in your own words and it changes, behind a confirmation.