Cold outreach that writes itself.
An AI browser agent fills out contact forms with personalized messages. A self-learning cold email agent writes, judges, and sends emails — then auto-creates CRM contacts from positive replies. Replaces an SDR, a cold email tool, and a form submission service.
Two outreach engines, one system
The platform has two complementary outreach systems that work together or independently:
The form outreach engine uses an AI browser agent to navigate websites, find contact pages, fill out forms, solve CAPTCHAs, and submit personalized messages — with a screenshot captured at every step.
The email outreach engine uses LLM personalization to write cold emails, an AI judge to score quality, human approval before sending, Brevo for delivery with open/click/reply tracking, and reply classification that auto-creates CRM contacts from positive responses.
AI browser agent for contact forms
Some businesses don't list an email — they only have a contact form. The AI browser agent handles this. It launches a real Chromium browser, navigates to the target website, finds the contact page, fills out every field with your sender details and a personalized message, solves CAPTCHAs (reCAPTCHA v2/v3, Turnstile, hCaptcha via CapSolver), and submits the form.
Every step is captured as a screenshot, so you can verify exactly what the agent did. If a form submission fails, you see why. If it succeeds, you see the confirmation page. No black box — full transparency.
- AI browser agent navigates websites and finds contact forms
- Fills out every field with sender details and personalized message
- CAPTCHA solving — reCAPTCHA v2/v3, Turnstile, hCaptcha
- Screenshot capture at every step — full transparency
- Concurrent browser processing for scale
- Rotating residential proxies for stealth
Self-learning cold email agent
The email outreach engine is a self-learning cold email system. It pulls leads from your lead generation campaigns, generates personalized emails using LLM, scores each email with an AI judge, routes through human approval, sends via Brevo, tracks opens/clicks/replies, and classifies replies as positive or negative.
The self-learning part: when you approve an email, it's added to the example library. When you reject one, it's added as a negative example. The system uses these examples as few-shot prompts for future emails — so the more you use it, the better it gets at matching your voice and style.
- LLM-generated personalized emails using lead context and business summaries
- Few-shot learning from approved and rejected examples — improves over time
- AI judge scores on personalization, relevance, clarity, conciseness, and call-to-action
- Human approval queue — review, edit, or reject before sending
- Brevo delivery with open, click, reply, bounce, and unsubscribe tracking
- Reply classification — positive replies auto-create CRM contacts and deals
AI judge: quality control before anything sends
Every generated message — whether for email or form submission — is scored by an AI judge on five dimensions: personalization, relevance, clarity, call-to-action, and appropriateness. The judge gives a score and a recommendation: SEND, REVISE, or REJECT.
High-scoring messages can be auto-approved (if you enable self-learning). Low-scoring messages are auto-rejected. Everything in between goes to the human approval queue. You can override any decision — the judge is a filter, not a final say.
Email warmup and CAN-SPAM compliance
New sending domains get flagged as spam if you send too many emails too fast. The system includes configurable email warmup — start with a small daily count, increment gradually over a ramp period, and cap at your daily limit. This builds sender reputation naturally.
Every email includes a physical address in the footer (CAN-SPAM requirement), a unique unsubscribe link, and suppression list management for bounces, spam complaints, and unsubscribes. You stay compliant automatically.
- Email warmup — configurable start daily, increment, and ramp period
- Physical address in email footer (CAN-SPAM)
- Unique unsubscribe link per email
- Suppression list — bounces, spam complaints, unsubscribes
- Send window controls — hours, timezone, days of week
Auto outreach: the complete pipeline on autopilot
Turn on auto outreach and the system runs the entire pipeline: discover leads from Google Maps → generate personalized messages → judge quality → route through human approval → send via email and/or form submission → track replies → auto-create CRM contacts from positive responses.
Daily limits per channel (emails, forms, leads) keep things controlled. Proxy health monitoring pauses lead generation if proxies run low. One campaign manages the entire lead-to-reply lifecycle — set it up once and monitor the results.
Auto outreach combines lead generation + email + forms. Requires both add-ons. See Lead Generation
Replies flow into your CRM automatically
When someone replies to a cold email, the AI classifies the reply as positive, negative, or neutral. Positive replies automatically create a CRM contact and a deal in your pipeline — with the original email and reply threaded together.
You don't need to check a separate inbox. Replies show up in your CRM, in your outreach analytics, and in your AI assistant's context. The loop is closed: lead → outreach → reply → CRM → follow-up.
Analytics and self-improvement
Track everything: emails generated, approved, rejected, sent, opened, clicked, replied. Form submissions attempted, succeeded, failed. Reply rates by campaign, by niche, by city. Judge scores over time — see if the AI is getting better.
The system learns from every approval and rejection. Your approved emails become examples for future generation. Your rejections become negative examples. Over time, the AI converges on your voice, your style, and what works for your market.
Explore More Features
Your website should work while you do.
Custom-built, AI-powered, and loaded with everything your business needs — content, CRM, voice agent, automations, and more. Live in seven days.
Or try the live demo — no signup needed