One AI. One Marketplace.

Your assistant. Your apps. Your machine.

Plexon answers questions, writes documents, organizes your files, and does real work inside apps like Slack, Notion, Gmail, and GitHub. Pick a persona and it thinks like a developer, designer, marketer, dietitian, financial analyst, product manager, stock trader, or writer. Everything runs from your computer, and your data stays on it.

DownloadSee what it can do
Plexon AI · walkthrough
Plexon AI

Summarize this week in #releases and add the launch dates to roadmap.xlsx.

connector:slack · read 34 messages in #releases

tool:excel · wrote 3 rows to roadmap.xlsx

Three releases shipped this week. I added each one to the Launches sheet in roadmap.xlsx with its date and owner. Want the summary posted back to #releases?

Connectors

Works with the apps you already use

63 built-in app connectors. Turn one on with your own account and the AI works inside it: send a Slack message, open a GitHub pull request, update your store, search the web. Click any app for what it can do and how sign-in works.

Personas

Pick who the AI should be

A persona swaps the system prompt and bundles its own agents, skills, connectors, dashboards, and workflows.

Software Developer

A full-stack engineering crew in one workspace.

“Refactor this function and add tests”

Designer

Design interfaces people can actually use.

“Critique this landing-page layout”

Marketer

Plan campaigns, write the copy, prove what worked.

“Write a positioning statement for this product”

Dietitian

Plan meals, track nutrition, support clients.

“Draft a seven-day meal plan for 2,200 kcal”

Financial Analyst

Build the model, value the company, defend the number.

“Build a three-statement model from these historicals”

Product Manager

Ship the right thing, in the right order.

“Draft a PRD for this feature”

Writer

Draft, edit, and publish with a voice that stays yours.

“Outline a chapter on the history of the printing press”

Stock Trader

Analyze the setup, manage the risk, keep the journal.

“Analyze NVDA: trend, key levels, and the RSI and MACD read”

Greek Logistics

Greek accounting, tax, gov.gr and logistics, current to 2026.

“Πόσο φόρο εισοδήματος για μισθωτό με 25.000€ το 2025;”

Startup Founder

Validate the idea, raise the round, grow the company.

“Pressure-test my startup idea and design three Mom Test interviews”

Personal Shopper

Find it, compare it, and buy it for less.

“Find wireless earbuds under €80 and compare the top three”

Mirror

Someone in your corner, every day.

“I need to talk through something that has been bothering me”

Build your own

Write a custom persona with its own AI instructions in the persona wizard (Premium).

Learn more →
plexon · session log
07:00:02 schedule ▸ job "monday-metrics" fired · cron 0 7 * * 1
07:00:03 orchestrator ▸ task split into 3 subtasks · wave 1 of 2
07:00:04 agent 1/3 ▸ session started · mode: ask · provider: anthropic
07:00:07 connector:github ▸ read 14 merged pull requests
07:00:09 connector:slack ▸ read #releases · 27 new messages
07:00:12 connector:google-workspace ▸ read 3 unread threads in inbox
07:00:19 skill:weekly-report-format ▸ loaded (pinned)
07:00:26 tool:excel ▸ wrote metrics-week-27.xlsx · 4 sheets
07:00:31 orchestrator ▸ 3 of 3 agents done · results synthesized in 29s
07:00:32 telegram ▸ summary sent to your chat
07:00:33 curator ▸ saved 1 new skill · refined 1 · retired 2 stale
07:00:33 session ▸ saved locally · encrypted AES-256-GCM · cost $0.14

Features

Everything in the box

The full feature set. Skim now, dig later: every card opens a deeper page.

The Skills sidebar with the self-improving skills settings card open

Skills

Self-Improving Skills

Plexon grows its own knowledge library while you work. After every session, or on demand, a small curator agent quietly reviews what happened, saves new skills worth keeping, refines existing ones, retires the stale, and protects everything you've pinned. You get smarter assistance over time without writing a single instruction by hand. One-click toggles in the Skills sidebar; full transparency via the curator history viewer.

Learn more →
The Schedule panel for running agents on a recurring schedule

Automation

Schedule: Cron for Agents

Run any sub-agent on a cron expression or simple interval. Every N months, days, hours, minutes, or seconds. Failed jobs auto-disable after 3 consecutive failures, live OS desktop notifications fire when work completes, and every job dispatches to the same agents that answer chat. "Weekly metrics report Monday 9 AM" or "every 6 hours summarise PR review queue" becomes a single click in the Schedule sidebar.

Learn more →
A workflow on the Dynamic Workflows canvas: a trigger followed by typed steps and a condition branch

Workflows

Dynamic Workflows

Turn a trigger into a small graph of steps and let Plexon run it. A schedule, an incoming message, a webhook, or a button starts the flow; each step calls a connector, asks the AI, runs an agent or persona, branches on a condition or expression, loops over a list, transforms data, waits for a reply, or asks you to approve. Build it on the canvas or describe it in chat, test any step on the spot, then review before it runs. The engine runs locally, remembers every run, and sends each message at most once. Ships a Greet new contacts example, and personas bring their own, like the Stock Trader pre-market briefing and alerts that stay quiet until something moves.

Learn more →
The Global Intelligence dashboard board with live widgets

Monitoring

Dashboards

Build a board of widgets that read from your tools, the AI, or a workflow, and keep themselves up to date. Stat tiles, charts, gauges, tables, even full HTML you can edit with AI. Add a parameter like a city and the whole board follows. Turn on monitoring and it refreshes in the background. Ships a Weather & City board and a Global Intelligence board, both filled from free public data the moment you open them.

Learn more →
Orchestrator mode running three research sub-agents in parallel

Agents

Orchestrator & Sub-Agents

Break complex tasks into parallel subtasks executed by independent sub-agents. Wave-based scheduling, inter-agent bulletin board, per-agent cost tracking, and @mention invocation.

Full-Stack E-Commerce Platform

Speedup

3.75x

Parallel

~12 min

Cost

$2.80

Sequential: ~45 minParallel: ~12 min
Learn more →

And everything else in the box

Six Specialized AI Modes

Code, Plan, Ask, Auto, Orchestrator, and Spec, each purpose-built for a different workflow.

156 Specialized Agents

Pre-built agent personas across software engineering, marketing & growth, sales, legal & compliance, data analytics, game development, research, project management, and China/Asia markets.

Personas with Domain System Prompts

Twelve curated personas, each swapping in its own server-hosted system prompt so the AI genuinely thinks like that role.

Team Coordination

Persistent multi-agent teams with coordinator mode, file-based mailboxes, shared knowledge base, and member wake-up.

Desktop Application

Native desktop app with 15 themes, real-time streaming, session replay with speed control (0.5x–8x), source control integration, per-session cost tracking, and rich markdown rendering.

Autonomous Development

Auto mode runs up to 20 development cycles with quality validation (configurable threshold, default 70%), test execution, per-cycle cost tracking, and self-correction loops.

Custom Providers (Premium)

Point Plexon at any OpenAI-compatible or Anthropic-compatible endpoint.

API Connections

Turn any HTTP API into AI-callable tools.

Unified Marketplace

One place to browse and install everything Plexon supports: skills, agents, plugins, personas, and tools.

Visual Understanding

Paste screenshots, photos, or diagrams for AI analysis.

Telegram Bot

Chat with Plexon AI from Telegram.

Slack Bot

Chat with Plexon AI from Slack, and from your Apple Watch.

Continuous Voice Conversations

Hands-free, turn-by-turn voice loop in the chat input.

Prompt Suggestions

After every AI response, Plexon predicts what you'd type next and shows it as placeholder text.

Side Questions

Ask about the conversation while the AI keeps working; the answer streams into a side card and never touches the task.

Shell Commands

Start a message with ! to run it as a shell command; the output renders as a terminal card and feeds the AI's context on your next message.

Graph-Aware Code Review

Plugin-powered PR review that builds a persistent Tree-sitter knowledge graph of your repo and surfaces only the files a diff actually touches.

Dynamic UI Widgets

AI responses render as rich interactive components (grids of recipe cards, macro tables, meal-plan timelines, togglable checklists, stat tiles, and forms) instead of walls of markdown.

Word & Excel in Chat

Two bundled MCP servers let the AI read, create, and edit Microsoft Word (.docx) and Excel (.xlsx) files directly from chat.

Read Any Document

Point Plexon at a Word doc, spreadsheet, slide deck, PDF, EPUB, OpenDocument file, RTF, email, or a LaTeX or reStructuredText file and it reads the contents as Markdown you can ask about.

Embedded Browser

A split-pane webview docked next to the chat.

Session Lock & App Lock (Premium)

Lock a chat with a passcode and its messages are encrypted with a key derived from it. Unlock with Touch ID or Windows Hello. An optional app lock guards the window at launch.

Migrate from Claude Code

Bring your agents, MCP servers, hooks, plans, memory, and every conversation across. Each plan lands in the project it was written for.

Zero-Knowledge Architecture

Privacy first, by design

Your code is processed, never stored. Multi-provider AI support with AES-256-GCM encryption and horizontally scalable stateless design.

Your Machine

  • Desktop app for Windows, macOS, and Linux
  • Local helper running quietly in the background
  • 30+ built-in tools (read files, search code, run commands)
  • AES-256-GCM encrypted sessions
  • Every edit instantly recoverable
  • All code stays on your machine

100% Local Storage

Your files never leave this machine

HTTPS Request

AES-256-GCM encrypted

SSE Response

Token-by-token streaming

Stateless Server

  • AI processing only
  • Routes your request to the AI you picked
  • Connects to 9 providers (Plexon, Anthropic, OpenAI, Google, …)
  • Streams the response back as it is generated
  • Smart caching keeps costs down (up to 90%)
  • Forgets everything as soon as it replies

0% Data Retention

No code, prompts, or sessions stored

AES-256-GCM

Military-grade encryption for sessions and API keys

Stateless Processing

Server maintains zero state between requests

Immediate Deletion

All data discarded after response generation

Built for Speed

Fast cross-platform desktop app: Windows, macOS, Linux

Session Lock (Premium)

Lock one chat with a 4-digit PIN. Its messages are encrypted with it

App Lock (Premium)

A gate at launch: a 4-digit PIN, Touch ID, or Windows Hello

On your phone

The same assistant, from Telegram

A full-featured Telegram bot that connects to your running Plexon instance. Chat with AI, manage sessions, analyze images, and monitor tools, remotely.

Plexon AI Botonline

Refactor the auth middleware to use JWT tokens

readFilemiddleware/auth.go

editFilemiddleware/auth.go

I've refactored the auth middleware to use JWT. The changes include a new validateJWT() function and updated all route guards to check the bearer token.

What's wrong with this UI?

I can see two issues: the button spacing is inconsistent (12px vs 8px), and the text color has low contrast on the light background. Let me fix both...

Message...

Bot Commands

/newCreate a new AI session

/switchList & switch between sessions

/modeSwitch AI mode (Code, Ask, Auto...)

/statusView bot uptime & session info

/usageToken usage & cost tracking

/stopCancel current AI execution

Learn more →

Image & Video Generation

Generate images and videos directly from chat using AI models like Gemini 3 Pro Image Preview. Create artwork, diagrams, mockups, and visual content without leaving the conversation.

Visual Understanding

Paste screenshots, photos, or diagrams. Plexon understands images natively. Analyze UI designs, debug visual bugs, extract text from images, and get feedback on designs.

Desktop Automation

With the Desktop Automation connector, the AI interacts with your desktop under your supervision: mouse, keyboard, screenshots, the accessibility tree, and app launch. Every action runs behind tool approval.

Voice Interaction

Local speech-to-text via Whisper with voice activity detection and auto-send. Text-to-speech reads AI responses aloud. No internet required after model download.

Compare

See how it compares

Plexon next to the desktop AI apps people actually use.

Feature comparison of Plexon AI, ChatGPT Desktop, Claude Desktop and Cowork, and open-source desktop AI apps
CapabilityPLEXON AIChatGPT DesktopClaude Desktop / CoworkOpen-source appsJan, LibreChat, Open WebUI
AI providers9 providers, or bring your own keyOpenAI models onlyAnthropic models onlyAny, bring your own key
Included AI modelYesThe Plexon model: text, images, and video on one subscription, even in a single prompt. Auto Routing picks the best model per request. No API keys, no usage bills.YesGPT models, plan limitsYesClaude models, plan limitsNoBring keys, pay per token
Local LLMs (Ollama, LM Studio, vLLM)YesCustom providers (Premium)NoNoYes
Where your data livesYour machine. Stateless server, zero retentionCloud storage; chats train models unless you opt outCloud storage; you choose whether chats train modelsYour machine (self-hosted)
App connectors63 branded connectors using your accountsApps directory, no published countConnectors directory plus any MCP serverManual MCP setup
Marketplace for skills, agents, personas, pluginsYesOne marketplacePartialGPT StorePartialConnectors directory and skillsNo
Private marketplaces for companiesYesCompany marketplace plus a sign-in gated internal one for staffPartialInternal GPTs in Team and Enterprise workspacesPartialPrivate plugin marketplaces on Team and EnterprisePartialLibreChat shares agents inside an instance
Secret items: skills and personas clients can use but not readYesInstructions and knowledge files ship as AES-256-GCM ciphertextNoNoNo
Personas that reconfigure the workspaceYes11 built-in, build your own. Each bundles its own agents, skills, plugins, and connectorsPartialCustom GPTsPartialProjects and stylesPartialPrompt presets
Parallel agents and orchestrationYesUp to 3 concurrent sub-agents, parallel workspaces on git worktreesPartialAgent modeYesCowork sub-agentsNo
Scheduled jobs (cron)YesAny agent on any scheduleYesTasks, not in the desktop appYesCowork, while the app is openNo
Trigger-based workflowsYesDynamic WorkflowsPartialMonitoring tasksNoNo
Chat from your phone via TelegramYesNoOwn appNoOwn appVaries
VoiceYesLocal Whisper STT, 400+ voices, 80+ languagesYesCloud voice modePartialBeta, cloud, web and mobileVaries
Image and video generationYesImages on both plans, video on Premium, or your own providerYesImages in app, video via SoraNoVia your APIs
Word and Excel files in chatYesRead, create, edit without Office installedYesReads and generates via data analysisYesCreates and edits filesNo
Price per month$39 Pro / $110 Premium$0 to $200: Free, Go, Plus, Pro$0 to $100+: Free, Pro, MaxFree, pay API usage

Comparisons checked July 2026. Tell us if something changed.

Switching

Never break your work

Your setup does not have to stay where you left it. Plexon imports your memory, your plans, and your sessions from other assistants, starting with Claude Code, along with the agents you wrote, your hooks, and your MCP servers. Point it at the folder, pick what to keep, and each piece lands in the project it belongs to.

AgentsMCP serversHooksPlansGlobal memoryProject memorySessionsSkillsSlash commands
How the migration works →
  1. It finds your installSettings reads the global ~/.claude folder. Open a repo Claude Code set up and Plexon offers that project on the spot.
  2. You pick what to keepEvery component arrives counted. Plans open into their own list, so the drafts you never approved can stay behind.
  3. Each piece lands where it belongsPlans and sessions go to the project they were written for, not one global pile. The results screen names every destination.

12

Personas

63

App connectors

156

Agents

200+

Skills

90+

MCP tools

6

Modes

FAQ

Frequently asked questions

Everything you need to know about Plexon's privacy-first architecture

How is Plexon different from GitHub Copilot or Cursor?

Plexon offers 6 specialized AI modes (vs 1), 156 agents across 8 professional domains, built-in browser and desktop automation, AI commit messages, prompt enhancement, parallel sub-agent orchestration, a dual skills system, hooks automation, voice I/O, Telegram bot, and a 100% stateless server. Your code never leaves your machine. Our server stores nothing.

Does your server store my code?

No. Our server is 100% stateless. It processes your AI request using 31 dynamically-assembled system prompt sections and immediately discards all data. Zero storage, zero logging, zero retention. API keys are AES-256-GCM encrypted in transit and never persisted.

Where is my conversation history stored?

Entirely on your local machine, encrypted with AES-256-GCM. Plexon stores them locally. You own your data completely.

Can I lock a single chat so nobody else can read it?

Yes. Lock a chat with a passcode and its messages are encrypted with a key derived from that passcode, so nothing reads them without it. Open it again with the passcode, with Touch ID on macOS, or with Windows Hello on Windows. A locked chat still shows in the sidebar by name with a lock badge, and stays out of exports and live shares until you unlock it. There is also an optional app lock that asks for a passcode before the window becomes usable on launch: that one is a gate against someone at your desk, not disk encryption. Both are local, nothing about them reaches a server, and a forgotten session passcode cannot be reset by anyone, us included. Turning a lock on needs Premium; unlocking, changing a passcode, and removing a lock work on any plan.

What LLM models does it support?

Plexon supports 9 providers: the managed Plexon provider, Anthropic (Claude Fable 5, Opus 4.8), OpenAI (GPT-5.3, Codex), Google Gemini (3.1 Pro/Flash, Nano Banana Pro 2), Moonshot (Kimi K2.5), OpenRouter (100+ models), Qwen (Qwen3.6 Plus, Qwen3 Coder Plus), Z-AI, and Claude Desktop. You can configure separate providers for execution and planning modes.

How does Orchestrator mode work?

Your task is decomposed into 2-5 subtasks by the AI. Up to 3 sub-agents execute in parallel, each with its own session and cost tracking. Results are synthesized by the planning provider into a coherent response. Complex tasks complete in ~45 seconds vs 2+ minutes sequential.

How does autonomous mode work?

The client orchestrates multiple development cycles. Each cycle: the AI generates code, quality is validated (0-100 score), tests are run, feedback is generated, and the AI iterates. Cycles continue until the quality threshold is met (default 70%, configurable) or the max cycle count is reached (default 5, max 20). Per-cycle cost tracking is shown in the UI.

What is the Skills system?

Skills are reusable instructions (coding standards, workflows, domain knowledge) that persist across sessions. Your custom skills are simple text files that work with every provider. Anthropic-native skills handle xlsx, docx, and pdf and run on the server. Both kinds coexist in a single request.

Does it work with my existing Claude skills?

Yes. Plexon uses an open skill format that other AI assistants share. Copy skills you have from other AI assistants into Plexon and they work immediately.

What platforms are supported?

Windows (x64, ARM64), macOS (Intel, Apple Silicon), and Linux (x64, ARM64). The desktop app is built and distributed for all platforms with auto-update support.

Can I self-host the server?

Yes. You can run your own stateless server instance. Either way (our hosted server or yours), no data is retained server-side.

How does browser automation work?

Plexon AI controls browsers via Playwright MCP: navigate, click, type, scroll, and screenshot. You can use your real Chrome profiles with logged-in sessions. The AI always asks for your permission before navigating to new URLs.

Can I use Plexon from my phone?

Yes. The Telegram bot integration lets you chat with Plexon, manage sessions, send images for analysis, and monitor what the AI is doing. All from your phone, in any country, using your usual Telegram account.

Is Plexon AI only for developers?

No. While Plexon excels at coding, it ships with 156 agents spanning 8 professional domains: software engineering, marketing & growth, sales & business, legal & compliance, data analytics, game development, research & creative, and specialized markets including China/Asia platforms. Marketers, lawyers, sales teams, and researchers all have dedicated agent personas.

What is prompt enhancement?

Prompt enhancement is an AI-powered rewriter that transforms simple requests into comprehensive, well-structured prompts. It adds technical context, resolves ambiguities, suggests design patterns, and defines quality criteria. All before your request reaches the main AI model.

Can I automate actions before or after tool execution?

Yes. The hooks system lets you run shell commands or HTTP webhooks before or after any tool call. Use pre-hooks to validate inputs or block dangerous operations, and post-hooks for audit logging, CI/CD triggers, or notifications. Hooks support regex tool matchers and can be scoped globally or per-project.

Can Plexon help me shop and compare prices?

Yes. The Personal Shopper persona searches products across stores, compares prices and features, suggests cheaper alternatives, and keeps a shopping list and budget. Connectors cover Google Shopping and web retailers, a Shopify store, eBay, an affiliate buy-link hub, and Skroutz price comparison in Greece. Buying is a hand-off: Plexon gives you a link to finish checkout on the store and never places an order.

I already use Claude Code. Do I have to set everything up again?

No. A migration wizard reads your existing Claude Code install and brings it across: custom agents, MCP servers, hooks, plans, your CLAUDE.md and user rules, project memory, and the full conversation history. Plans land in the project each one was written for and keep the approved or draft status they had, and the drafts you never approved can be left behind with one button. Everything imported is labelled with where it came from, so you can search for it, review it, or clear it out later. Nothing in your ~/.claude folder is moved, edited, or deleted, so Claude Code keeps working.

Privacy by architecture, not by promise

Ready when you are

Your data stays on your machine. Our server stays stateless. Be the first to experience privacy-first, persona-driven AI.

No credit card requiredYour data stays private