A personal AI assistant that runs 24/7 on your machine. Talk to it through Telegram or a web browser. It remembers your conversations, handles voice, and runs tasks while you sleep. No extra costs beyond your Claude Code plan.
$47 one-time · Guide + full source code · Instant download
Popular frameworks have 3,400+ open issues, charge $50-200/mo in token costs, and have been flagged by Cisco, Kaspersky, and Palo Alto for critical security flaws. The blueprint takes a different approach.
Everything is built, tested, and documented. Run the setup wizard and your agent is live. Tweak the personality, add skills, or use it as-is.
If you have a Claude Code subscription, you already have everything you need. No extra API key to create. No separate billing account. No surprise charges. Your agent runs on what you're already paying for.
// No API key needed. // Uses your Claude Code CLI auth. import { Agent } from '@anthropic-ai/claude-agent-sdk' const agent = new Agent({ model: 'claude-sonnet-4-6', tools: ['bash', 'file', 'web'], // That's it. No API key field. })
Tell your agent something once and it sticks. "I'm allergic to peanuts." "My Wi-Fi password is..." "Never schedule meetings before 10am." Important facts persist. Casual chit-chat fades naturally over time, just like real memory.
remembers | "Allergic to peanuts" importance: high remembers | "Meeting with Sarah on Fri" importance: high remembers | "Prefers window seats" importance: medium fading | "Asked about the weather" importance: low (fading) fading | "Talked about lunch spots" importance: low (fading) -- Important facts stick around forever -- Small talk fades naturally over time
Text, voice messages, photos, and documents through Telegram. A full web dashboard on port 3141 with multiple conversations, image paste, and admin pages. Same memory pool, same agent, two interfaces.
Voice transcription runs locally via whisper-cpp. No cloud API, no costs, no rate limits. Scheduled tasks run via cron while you sleep.
Telegram | text, voice, photos, documents | persistent sessions across restarts | /schedule create "..." "0 9 * * *" Web Dashboard (:3141) | multi-conversation chat UI | shared memory pool | admin + monitoring pages Voice | local whisper-cpp transcription | $0 cost // no cloud API
The setup wizard asks you a few simple questions and does the rest. It checks your system, creates your config, connects your Telegram bot, and installs the agent as a background service. It starts when your computer starts and restarts itself if anything goes wrong.
$ npm run setup Checking environment... ✓ Node.js 22.1.0 ✓ Claude Code CLI authenticated ✓ whisper-cpp found Telegram bot token: ***pasted*** Your chat ID: ***detected*** Install as background service? yes ✓ com.agent.app installed ✓ Starts on login, restarts on crash Done. Run npm run dev to start.
Skills are simple text files that tell your agent how to handle specific tasks. Drop a file in the skills folder and your agent can manage your email, check your calendar, research topics, organize files, or anything else you describe.
5 ready-to-use skills are included. Creating your own takes minutes -- just describe what you want in plain English. No code to write, no packages to install, no marketplace to trust.
gmail/SKILL.md "Check inbox, read, reply, search" google-calendar/SKILL.md "View schedule, create events, find free time" web-research/SKILL.md "Research any topic, compare options, cite sources" daily-briefing/SKILL.md "Morning update: calendar, weather, news, tasks" file-organizer/SKILL.md "Sort downloads, find duplicates, clean up folders" + Add your own -- just describe it in a text file
Other agent frameworks have made headlines for all the wrong reasons -- remote code execution, exposed databases, malicious plugins. The blueprint was designed to keep things locked down from day one.
A step-by-step PDF guide that walks you through the whole system, plus the complete working codebase. Download, set up, done.
src/ index.ts Entry point bot.ts Telegram bot (grammY) agent.ts Claude Agent SDK wrapper dashboard.ts Web UI (Express) db.ts SQLite + FTS5 search memory.ts Dual-sector memory scheduler.ts Cron task runner voice.ts Local whisper-cpp media.ts Photo/document handling config.ts Environment config scripts/ setup.ts Interactive setup wizard status.ts Health check notify.sh CLI Telegram messenger workspace/ reddit-monitor/ Bonus: RSS keyword monitor
You need to be comfortable running a couple commands in your terminal. The setup wizard does the heavy lifting. If you can install an app and follow a guide, you can get this running.
No. It runs on your existing Claude Code subscription. No separate API key, no per-token charges, no surprise bills. You pay $47 once and that's it.
OpenClaw requires a separate API key ($50-200+/mo in token costs), has 3,400+ open issues, and needs ongoing maintenance after updates. This blueprint uses the official Agent SDK, costs nothing extra, and sets up in 5 minutes with zero configuration headaches.
The web dashboard works on its own -- just open a browser. Telegram is optional. You can also connect Discord, Slack, or other platforms.
Works on Mac and Linux. The setup wizard detects your system and configures everything automatically.
Everything about your agent -- name, personality, rules, what it can do -- lives in one file you can edit in plain English. No code changes needed. Make it professional, casual, funny, strict -- whatever fits you.
The four systems every robot has -- sensors, brain, memory, actuators -- and how they map 1:1 to an AI agent.
GuideNo separate API key. No per-token billing. How the Agent SDK uses your existing CLI auth.
ComparisonThe OAuth shutdown, the cost difference, and when each framework makes sense.
No extra monthly costs. No complicated setup. No Docker, no gateway, no debugging dependency errors. One download, one wizard, and your personal AI agent is running on your machine.
Instant delivery via Gumroad. PDF + repo access. No recurring fees.