Grow on X
while you sleep.
X-Autopilot is a fully local AI agent that opens a real Chrome browser, logs into your Twitter/X account, and engages with your target audience 24/7 — liking, replying, following — written entirely in your voice, powered by Claude or GPT-4.
What Is X-Autopilot?
X-Autopilot is a desktop AI agent — it runs entirely on your Mac or Windows computer. No monthly subscription, no cloud server, no data leaving your machine. It opens a real Google Chrome browser, navigates Twitter just like a human would, and autonomously engages with your audience throughout the day.
Writes In Your Voice
Powered by Claude (Anthropic) or GPT-4 (OpenAI). The AI knows your niche, personality, products, and phrases you'd never say — every reply sounds like you wrote it.
Undetectable Automation
Uses a real Chrome browser with GPU/CPU fingerprint spoofing, randomised typing speed, natural mouse trajectories, and human scroll physics. Twitter sees a real user.
Smart Account Targeting
Engages with your hand-picked target accounts, trending hashtags, and home feed. Automatically discovers new relevant accounts to follow and engage with.
Live Web Dashboard
Monitor everything at localhost:5001 — live action log, stats, follower growth chart, settings, and controls. Works in any browser.
100% Local & Private
Your API keys, Twitter session, and activity data never leave your machine. No accounts to create, no cloud sync, no external servers involved whatsoever.
Zero Coding Required
Double-click to install, double-click to start. Configure everything through the visual dashboard. No terminal knowledge needed after the one-time setup.
What You Need
Required Software
Supported Systems
Installation
Verify Python is installed
Press ⌘ Space, type Terminal, press Enter. Then run:
You should see Python 3.x.x. If not, download from python.org/downloads — install the latest 3.x release.
Unzip X-Autopilot
Double-click x-autopilot.zip to extract it. Move the folder somewhere permanent — your Desktop or Documents folder works well. Don't run it from inside the ZIP.
Run Setup (one time only)
Inside the folder, find 2. Setup.command and double-click it.
A terminal window opens and automatically installs all dependencies. This takes 2–4 minutes. When it says Setup complete, close the terminal.
Start the agent
Double-click 4. Run Agent.app. The dashboard opens in its own window at localhost:5001 — no browser tabs, no clutter. On Windows, double-click start.bat instead.
Add your credentials
Go to localhost:5001 → Settings → Account & API Keys and enter your Twitter username and Anthropic/OpenAI key. Hit Save. Done.
Install Python
Go to python.org/downloads and download the latest Python 3.x installer. Run it and check the box "Add Python to PATH" — this is critical or nothing will work.
Extract the ZIP
Right-click x-autopilot.zip → Extract All. Put the folder on your Desktop or Documents. Don't run from inside a ZIP.
Run setup.bat
Double-click setup.bat. A black console window installs everything automatically.
Start with start.bat
Double-click start.bat every time you want to run X-Autopilot. Your browser will open the dashboard automatically.
Add credentials via dashboard
Open localhost:5001 → Settings → Account & API Keys in your browser and enter your details.
API Keys Setup
X-Autopilot needs an AI API key to generate replies, analyse tweets, and write in your voice. You need at least one. Both can be added as a fallback.
Claude API
Claude writes the most natural, contextual replies. Best at matching personality and avoiding robotic phrasing. Strongly recommended over OpenAI for this use case.
2 · Sign up or log in
3 · Click API Keys → Create Key
4 · Copy the key — starts with
sk-ant-
GPT-4 API
X-Autopilot can use GPT-4o or GPT-4o-mini as a fallback when Claude is unavailable, or as your primary provider if you prefer OpenAI.
2 · Sign up or log in
3 · Click API Keys → Create new secret key
4 · Copy the key — starts with
sk-proj-
.env file and never transmitted anywhere.First Run & Login
Launch the agent
Double-click 4. Run Agent.app (Mac) or start.bat (Windows). The dashboard opens in its own dedicated window at localhost:5001 — no address bar, no browser tabs.
Log into Twitter in the Chrome window
Chrome opens Twitter's login page. Sign in with your account username and password — exactly as you normally would. Complete any 2FA or phone verification Twitter asks for.
Confirm in the terminal
Once you're on your Twitter home feed, switch back to the terminal window and press ENTER. The agent confirms the session and begins operating.
Watch it go
A floating status bar appears in the Chrome window showing what the agent is doing in real-time. The dashboard at localhost:5001 shows live stats and action logs.
To stop the agent
On Mac: close the dashboard window or press Cmd+Q — the agent stops automatically. You can also double-click 4. Run Agent.app again while it's running and click Quit. On Windows: close the terminal window or use the Quit button in the dashboard.
Dashboard · Overview Tab
Open http://localhost:5001 in any browser. The Overview tab is your mission control.
| WIDGET | WHAT IT SHOWS |
|---|---|
| Replies Today | Number of replies posted so far vs. your daily target |
| Likes Today | Likes given today — counts toward daily limit to avoid rate limiting |
| Follows Today | New accounts followed. Kept low (10/day) to avoid Twitter penalties |
| DMs Sent | Direct messages sent. Only 4/day to avoid spam flags |
| New Followers | Net new followers gained vs. yesterday's count |
| Agent Status | What the agent is doing right now — liking, replying, sleeping, etc. |
| Pause Button | Pauses the agent after its current action. Click again to resume. |
| Quit Button | Cleanly stops the agent. All progress is saved. |
Dashboard · Activity Tab
Every single action the agent takes is logged here with timestamp, action type, target tweet/user, and whether it succeeded.
| ACTION TYPE | ICON | COLOUR | WHAT IT MEANS |
|---|---|---|---|
| REPLY | ↩ | Blue | AI-written reply posted to a tweet |
| LIKE | ♥ | Pink | Tweet liked |
| FOLLOW | + | Purple | Account followed |
| UNFOLLOW | − | Grey | Account unfollowed (non-followers after 14 days) |
| DM | ◉ | Orange | Direct message sent |
| TWEET | ✦ | Green | Original tweet posted (after your approval) |
| QUEUED | ◈ | Yellow | AI-generated tweet waiting for your approval in the dashboard |
Dashboard · Settings Tab
The Settings tab lets you change everything without touching config.yaml. Changes save instantly and take effect on the next agent action.
| SECTION | SETTINGS AVAILABLE |
|---|---|
| Account & API Keys | Twitter username, Anthropic API key, OpenAI API key (fallback) |
| Voice & Persona | Niche, personality, products, product URLs, never-say phrases |
| Daily Targets | Replies, likes, follows, DMs, retweets per day |
| Posting Schedule | Active hours start/end, tweet times |
| Target Accounts | Add/remove Twitter accounts to engage with |
| Discovery | Hashtags, home feed scanning, max tweets scanned |
| Safety | Max actions per hour, pause between sessions, rate limit behaviour |
| Autonomy | Autonomy level (0–100%), require tweet approval toggle |
Dashboard · Growth Tab
The Growth tab shows your follower count over time as a chart. Data is recorded at the end of each active day.
Actions Explained
| ACTION | HOW IT WORKS | DAILY LIMIT | TRIGGER |
|---|---|---|---|
| ↩ REPLY | AI reads the tweet, generates a reply in your voice, opens the reply box and types it with human-like speed and occasional typos that get corrected. | 40 / day | Target accounts, hashtag feed, home feed |
| ♥ LIKE | Clicks the like button on relevant tweets from your targets and home feed. Helps with visibility and signals engagement before following. | 45 / day | Target profiles, home feed, hashtag results |
| + FOLLOW | Follows accounts that match your niche and have engaged with relevant content. Focuses on smaller accounts most likely to follow back. | 10 / day | Post engagement, hashtag discovery, mentions |
| ◉ DM | Sends personalised DMs to small accounts after several interactions. AI writes a context-aware opening message — not a generic pitch. | 4 / day | After 2+ likes/replies to same account |
| ◈ TWEET | AI generates original tweet drafts aligned with your content strategy. By default, these appear in the dashboard for your approval before posting. | 4 / day | Scheduled tweet times in config.yaml |
| − UNFOLLOW | Automatically unfollows accounts that didn't follow back after 14 days, keeping your following/follower ratio healthy. | Auto | 14 days after follow, if not followed back |
| 🔔 NOTIFS | Checks notifications, reads the full conversation thread (scrolling up to see original post + your reply + their response), then decides how to reply again. | 3 / session | Notification tab — replies, mentions, DMs |
| ↻ RETWEET | Retweets highly relevant content when the AI's relevance score is above threshold. Kept very low to maintain authenticity. | 2 / day | High-scoring relevant tweets only |
| ⊛ PRODUCT | When a tweet's topic matches your product's keywords (e.g. "invoicing", "icon pack"), the reply weaves in a natural product mention — value add first, then a conversational "I built something for this" with your URL. | 2 / day | Keyword match on tweet text vs. product trigger list |
promotions: in config.yaml.Tiered Engagement Strategy
X-Autopilot treats accounts differently based on their follower count. This maximises follow-backs from small creators while gaining visibility from large ones.
Relationship Mode
Your future loyal fans. Highest follow-back rate.
Networking Mode
Your collaborators and cross-promoters.
Visibility Mode
Exposure to their large audiences.
Sessions & Daily Schedule
X-Autopilot runs in intelligent sessions with natural breaks — not continuously. This looks human and stays under Twitter's rate limits.
Browser Status Overlay
A floating status bar appears at the top of the Chrome window while X-Autopilot runs. It shows the current status and gives you quick controls without opening the dashboard.
| BUTTON | WHEN VISIBLE | WHAT IT DOES |
|---|---|---|
| ⏩ SKIP | During breaks and sleep periods | Immediately ends the current rest period and starts the next engagement session |
| ⏸ PAUSE | Always visible during active session | Pauses after the current action completes. Click again (▶ RESUME) to continue. |
| ✕ QUIT | Always visible | Sends a clean shutdown signal. Agent finishes its current action, saves all data, then stops. |
Safety & Rate Limits
X-Autopilot has multiple layers of safety built in to protect your account. These run automatically — you don't need to do anything.
Hard Daily Limits
Anti-Detection Layers
Voice & Persona Setup
This is the most important configuration. The more specific you are, the more your replies will sound like you and not a generic AI.
never_say list is your secret weapon. Add any phrase that sounds corporate, generic, or unlike you. The more you add, the more authentic the output becomes.Target Accounts — Auto-Discovered
No hardcoded accounts needed. X-Autopilot scans your niche hashtags each morning, visits candidate profiles, checks their bio and recent tweets for relevance, and adds qualifying accounts automatically — sorted into the correct tier by follower count.
data/targets.json with the correct tier (small / peer / big) based on follower count. You never need to maintain a manual list.Full Config Reference
Every setting in config.yaml — explained. You can edit this file directly or use the Settings tab in the dashboard.
| KEY | DEFAULT | DESCRIPTION |
|---|---|---|
| tweets_per_day | 4 | How many original tweets the AI generates per day |
| require_approval | true | If true, tweets queue in dashboard for your review. Set false for full auto-posting. |
| tweet_times | 09:30, 19:00 | Times the agent attempts to post original tweets |
| active_hours_start | 09:00 | Agent wakes up at this time each day |
| active_hours_end | 23:00 | Agent sleeps at this time each day |
| KEY | DEFAULT | DESCRIPTION |
|---|---|---|
| daily_replies | 40 | Maximum replies per day across all sessions |
| daily_likes | 45 | Maximum likes per day |
| daily_follows | 10 | Maximum new follows per day |
| daily_dms | 4 | Maximum DMs per day — keep low to avoid spam flags |
| daily_retweets | 2 | Maximum retweets per day |
| min_delay_seconds | 15 | Minimum wait between actions (seconds) |
| max_delay_seconds | 90 | Maximum wait between actions (seconds) |
| KEY | DEFAULT | DESCRIPTION |
|---|---|---|
| hashtags | buildinpublic… | Hashtags to scan for relevant tweets to reply to |
| reply_from_hashtags | true | Whether to find and reply to tweets via hashtag search |
| reply_from_home_feed | true | Whether to reply to tweets from your home feed |
| max_hashtag_replies_per_session | 5 | Max replies from hashtag results in one session |
| max_home_replies_per_session | 8 | Max replies from home feed in one session |
| max_hashtag_tweets_scanned | 60 | How many hashtag tweets to scan before picking the best ones |
| candidate_score_threshold | 0.15 | Minimum relevance score for a tweet to qualify for reply |
| use_embeddings | true | Use vector embeddings for semantic tweet relevance scoring |
| min_likes | 0 | Minimum likes a tweet must have to be engaged with |
| KEY | DEFAULT | DESCRIPTION |
|---|---|---|
| max_actions_per_hour | 12 | Hard cap on total actions in any 60-minute window |
| pause_between_sessions_minutes | 60 | Rest time between engagement sessions |
| unfollow_non_followers_after_days | 14 | Days before unfollowing accounts that haven't followed back |
| idle_scroll_enabled | true | Scroll home feed during breaks to look human |
| idle_scroll_interval_minutes | 8 | How often to do an idle scroll during breaks |
| stop_on_rate_limit | true | Pause if Twitter signals rate limiting |
| rate_limit_cooldown_minutes | 90 | How long to wait after a rate limit is detected |
| dynamic_pacing | true | Automatically slow down if approaching daily limits early |
| KEY | DEFAULT | DESCRIPTION |
|---|---|---|
| reply_to_mentions | true | Automatically reply to tweets that mention you |
| max_reply_notifications_per_session | 3 | Max notification replies processed per session |
| follow_welcome_enabled | true | Send a welcome interaction when someone follows you |
| max_follow_welcomes_per_session | 4 | Max welcome interactions per session |
| follow_welcome_like_min_posts | 3 | Min posts to like on a new follower's profile |
Pro Tips for Maximum Growth
Target the Sweet Spot
Add 20+ accounts in your exact niche. The best targets are creators with 500–5,000 followers who post regularly. They check their notifications obsessively and are most likely to engage back.
Be Brutally Specific on Voice
Add at least 10 phrases to never_say. Include any word that sounds corporate or AI-generated. "Absolutely", "Certainly", "Great point" — all banned. The more you restrict, the better the output.
Run It Every Day
Growth on X is compound. 30 days of 40 replies/day = 1,200 touch points. Those compound into follows, profile visits, and brand awareness. Running 3 days then stopping resets momentum.
Prioritise Evening Active Hours
Set active hours to include 6pm–11pm local time. Your target accounts post in evenings. First replies on a fresh post from a big account get the most eyeballs — the agent watches for new posts.
Review Queued Tweets
With require_approval: true, the AI generates tweet drafts that appear in your dashboard. Review and edit these before approving — treat the AI as a first-draft writer. Your edits train your own eye for what works.
Track the Activity Log
The Activity tab shows everything the agent is doing. After a week, look for patterns — which reply styles get follow-backs? Which accounts respond? Use this to refine your targets and voice settings.
Frequently Asked Questions
X-Autopilot is designed to be undetectable at every layer. It uses a real Chrome browser with authentic GPU and CPU fingerprinting, randomised human typing speed (25–90ms per key), natural mouse curves with drift, hover dwell before clicks, trackpad-style scroll physics, and 15–90 second randomised delays between actions. Daily limits (40 replies, 45 likes, 10 follows) stay well within safe thresholds. No tool is 100% guaranteed, but X-Autopilot uses every known detection-avoidance technique. Use it at the recommended limits and you're in the safest possible zone.
Yes. X-Autopilot runs locally on your machine, so your computer and Chrome window need to stay open. You can minimise Chrome and lock your screen — the agent keeps running. If you close the terminal or Chrome, it stops. If you need 24/7 running without keeping your laptop on, you could run it on a cheap Windows VPS (like Hetzner or Contabo), but that's optional.
Very little. At 40 replies/day × 30 days = 1,200 AI calls/month. At Claude's current pricing (claude-3-5-sonnet), that's roughly $3–8/month depending on reply length. Using GPT-4o-mini is even cheaper at $1–3/month. Both providers give free trial credits when you sign up. You can also reduce daily_replies to lower costs — even 15/day builds significant growth over a month.
Yes, and it's the recommended approach. X-Autopilot is designed for building a personal brand — your expertise, your niche, your products. It logs into your account via Chrome exactly as you would. One installation supports one account.
The whole system is designed around authenticity. The AI receives your niche, personality, writing style, forbidden phrases, product context, proof points, and the full text of the tweet it's replying to. Claude in particular is exceptional at avoiding "AI voice" — it produces specific, contextual, opinionated replies rather than generic affirmations. After setup, read the Activity log to see what it's posting. If anything sounds off, tighten your never_say list and personality description in config.yaml.
Mac: Close the dashboard window, press Cmd+Q, or double-click 4. Run Agent.app again and click Quit. All three stop the agent cleanly and close everything. Windows: Click the red ✕ QUIT button in the dashboard, or close the terminal window. All data is saved either way.
Both. By default (require_approval: true), the AI generates original tweet drafts that appear in the dashboard for your review and approval before posting. Replies to other tweets happen automatically without approval. You can set require_approval: false in config.yaml to let it post original tweets fully autonomously.
On Mac, closing the window intentionally stops the agent — just double-click 4. Run Agent.app to start it again. On Windows, double-click start.bat. Your Twitter session is saved in the Chrome profile so you won't need to log in again. All activity history is preserved.
Technically yes — on a Windows VPS with Chrome installed and RDP access. It requires a visual desktop environment since Chrome runs as a visible window. Linux headless mode is not currently supported. A cheap Hetzner or Contabo Windows VPS (~$5–10/month) works well if you want 24/7 running without keeping your laptop on.
Typically 2–4 weeks of consistent running. Growth is non-linear — the first week looks slow because you're building awareness. By week 3, the compounding effect kicks in as people start recognising your name from repeated quality comments. Accounts with a tight niche and specific voice tend to see faster results than broad accounts.
Troubleshooting
| PROBLEM | LIKELY CAUSE | FIX |
|---|---|---|
| Dashboard won't open at localhost:5001 | Server not started or port conflict | Restart the agent. If port 5001 is in use, close other apps using it. Check terminal for error output. |
| "Python not found" on Mac | Python not installed | Download from python.org, install, re-run 2. Setup.command |
| 2. Setup.command won't open on Mac | Gatekeeper blocking unsigned script | Right-click → Open → click Open. Only required once. |
| setup.bat blocked by Windows Defender | SmartScreen warning | Click More info → Run anyway. The script only installs Python packages. |
| Browser stuck on X logo / black screen | Chrome restored previous session, React hydration failed | Normal on first run or after a profile reset. The agent detects this automatically, clears the profile, and navigates to the login page. |
| Browser closes immediately on launch | Twitter session expired | Restart — the browser will open the login page. Log in and press ENTER in the terminal. |
| Agent shows "outside active hours" and does nothing | Current time outside configured hours | Normal behaviour. Adjust active_hours_start/end in config.yaml or Settings tab. |
| Replies sound generic or robotic | Voice config too vague | Add more entries to never_say. Make personality more specific. Add proof_bank examples. |
| API key error in activity log | Wrong or missing key | Go to Settings → Account & API Keys, re-enter your key, click Save. |
| Skip Break button not responding | Overlay not yet loaded | Wait 5 seconds and try again. The overlay refreshes every status update. |
| Rate limit warning appears | Twitter throttling | Automatic — agent pauses 90 minutes then resumes. Reduce daily limits if recurring. |
| No actions after 30 minutes | Daily limits already reached | Check Overview tab — if counts hit limits, agent rests until midnight reset. Increase limits in config.yaml. |
| Chrome says "session expired, log in again" | Twitter invalidated session | Normal after ~30 days. Just log in again when prompted — session saves fresh. |
| "Module not found" error in terminal | Dependencies not installed | Run 2. Setup.command/setup.bat again. It's safe to run multiple times. |
data/agent.log inside the X-Autopilot folder. It has a detailed timestamped record of everything the agent did and any errors encountered.