🆁 ReacherX

Open-source △ Agent that works 24/7 across X/Twitter and LinkedIn to find, qualify, enrich, and help you reach the right people. Think Cursor for finding and reaching your audience.

▪️ What is ReacherX?

ReacherX is an AI agent that finds people on X (Twitter) and LinkedIn, qualifies them against your criteria, enriches their profiles, learns your writing voice, and helps you engage with personalized outreach — all from a chat interface. You describe who you're looking for, and the agent handles discovery, qualification, enrichment, plan generation, and execution.

It adapts to whatever "reaching the right people" means for you:

Use Case	Who You Find	Success Looks Like
Customer Prospecting	People actively expressing a need your product solves	Converts
Recruiting	Candidates matching your open roles	Hires
Partnership Outreach	Companies or operators for strategic partnerships	Active partners
Investor Outreach	Investors whose thesis fits your opportunity	Committed investors
User Research	Participants who can give useful product feedback	Confirmed participants
Creator Outreach	Creators who match your audience or brand	Collaborators
Community Growth	People likely to join and participate	Members
Podcast Speaker Sourcing	Guests with the right expertise and voice	Booked guests

Each use case shapes the entire experience — search queries, qualification criteria, pipeline stages, outreach tone, UI labels, and analytics are all tailored.

Try it live →

▪️ Screenshots

Prospect Discovery & Outreach Pipeline

AI Agent Chat

Analytics Dashboard

▪️ How It Works

You describe who you're looking for (via chat)
        │
        ▼
   Agent generates search queries tailored to X and LinkedIn
        │
        ▼
   Finds matching people across both platforms
        │
        ▼
   Qualifies each person (ICP fit, engagement quality, authenticity)
        │
        ▼
   Enriches profiles (pain points, financial signals, professional context)
        │
        ▼
   Agent generates a personalized outreach plan
        │
        ▼
   You review and approve → agent executes (replies, likes, DMs, invites)
        │
        ▼
   Agent learns from outcomes and improves over time

▪️ Features

Two-Agent System

ReacherX runs two specialized agents:

Setup Agent — Handles onboarding and workspace creation. Analyzes your URL, generates your ICP, converts keywords into social search queries, and kicks off discovery.

Outreach Agent — Takes over once prospects are found. Fetches social context, generates engagement plans, drafts replies/DMs in your voice, and executes approved actions.

Both agents work through a chat interface with streaming responses, tool execution visualization, and inline cards for approvals, previews, and plan display.

Inline Autocomplete

Tab-complete for your replies and DMs — similar to Cursor's tab completion. The autocomplete is context-aware:

• Knows who you're replying to and what they said
• Respects character limits (X post weighted length vs raw)
• Matches your writing style when a style profile exists
• Adapts to workspace context and prospect information
• Shows as ghost text, accept with Tab

Writing Style Learning

The agent learns how you write by analyzing your posts on X and LinkedIn. It distills your vocabulary, sentence structure, humor, formality, emoji usage, opening/closing patterns, strengths, and weaknesses into a writing style profile. This profile is automatically applied when the agent drafts replies, DMs, and outreach content — so everything sounds like you, not a bot.

Style profiles are stored per-workspace and per-platform (Twitter vs LinkedIn).

Memory System

The agent builds workspace-level memory across 8 categories:

• Qualification win patterns — Signals that predict a good fit
• Qualification false positive patterns — Signals to avoid
• Enrichment signal patterns — Useful bio/post/financial indicators
• Enrichment role patterns — Recurring personas that matter
• Outreach winning patterns — Approaches that work
• Outreach objection patterns — Approaches to avoid
• Writing style profiles — Your voice on Twitter and LinkedIn

Memory is backed by vector search (RAG) for semantic retrieval. The agent queries relevant memories before generating plans or drafting content, and writes new memories as it learns from outcomes.

Social Actions (Human-in-the-Loop)

The agent can execute actions on both platforms with a risk-based approval policy:

X (Twitter):

• Low-risk (immediate): Like, unlike, bookmark
• Medium-risk (approval required): Retweet, follow, unfollow
• High-risk (approval required): Reply, create post, send DM

LinkedIn:

• All approval required: Send message, invite, react, comment

Every medium and high-risk action goes through approval — the agent stages the draft, you review it, then it executes.

Agent Ops Dashboard

A full operations dashboard with 5 tabs to monitor how the agent is performing:

• Overview — Health score, quality score, funnel metrics, best/weakest queries
• Discovery — Query inventory with performance scores, monitor health, novelty yield
• Quality — Qualification precision, enrichment usefulness, outreach effectiveness, correction trends
• Memory — Stored memories, confidence scores, impact scores, recent promotions
• Activity — Event feed, running/failed workflows, pending evaluations

Discovery & Monitoring

• AI-generated search queries from your ICP description
• Real-time monitoring via SocialAPI webhooks — new results trigger automatic processing
• Query candidate tracking with performance scoring (prospects found, qualified, converted, reply rate)
• Conversation seed monitors that detect when prospects reply to your engagement
• Discovery trace that tracks how each prospect was found (which query, which platform, which monitor)

Pipeline Management

Track people through stages adapted to your use case:

• Customer Prospecting: New → Contacted → In Progress → Converted
• Recruiting: Sourced → Contacted → Interviewing → Hired
• Partnership Outreach: Identified → Contacted → Negotiating → Partnered
• (each use case has its own stage labels)

Multi-Platform Support

X (Twitter) — Full support: search, profile lookup, thread analysis, replies, likes, retweets, DMs, real-time monitoring, conversation tracking.

LinkedIn — Search, profile enrichment, post analysis, messages, invitations, reactions, comments via Unipile and LinkdAPI.

Other Features

• Prospect enrichment — Type detection, pain point extraction, solution matching, financial signals
• AI qualification — ICP fit scoring (0-100), bot detection, engagement quality assessment
• Analytics dashboard — Discovery metrics, outreach metrics, engagement tracking, funnel visualization
• Notifications — Prospect found, outreach sent, approval requests, response alerts
• Billing — Subscription management via Polar with plan limits
• Workspace management — Multiple workspaces, each with its own use case, ICP, and pipeline

▪️ Why This Matters

Traditional Approach	ReacherX Approach
1. Create ad campaigns	1. Describe who you're looking for
2. Set targeting parameters	2. Agent generates search queries
3. Pay for impressions/clicks	3. Finds people actively expressing need
4. Hope people see and engage	4. Agent drafts personalized outreach in your voice
5. Wait for conversions	5. You approve, agent executes, learns from outcomes

▪️ Getting Started

Requirements

• Node.js 20.0.0+
• pnpm 9.15.4+ (npm and yarn are not supported)

Setup

git clone https://github.com/noobships/reacher-x.git
cd reacher-x
pnpm install

Configuration

Copy the environment template and fill in your API keys:

cp .env.example .env.local

Required

Variable	Service	What It Does
NEXT_PUBLIC_CONVEX_URL	Convex	Database and backend
WORKOS_CLIENT_ID	WorkOS	Authentication
OPENROUTER_API_KEY	OpenRouter	AI provider (agents, qualification, enrichment, autocomplete, embeddings)
X_API_CLIENT_ID	X Developer Portal	X OAuth
X_API_CLIENT_SECRET	X Developer Portal	X OAuth
X_API_BEARER_TOKEN	X Developer Portal	X Activity API
X_CONSUMER_SECRET	X Developer Portal	Webhook CRC signatures
X_TOKEN_ENCRYPTION_KEY	(generate a secret)	Encrypts stored X tokens
RESEND_API_KEY	Resend	Transactional email
CONVEX_SITE_URL	(your Convex HTTP URL)	Webhook URL construction

Required for LinkedIn

Variable	Service	What It Does
UNIPILE_API_KEY	Unipile	LinkedIn messaging, actions
UNIPILE_BASE_URL	Unipile	API endpoint
UNIPILE_WEBHOOK_SECRET	Unipile	Webhook signature validation
LINKDAPI_API_KEY	LinkdAPI	LinkedIn search and profile data

Required for Twitter Monitoring

Variable	Service	What It Does
SOCIALAPI_API_KEY	SocialAPI	Twitter search, profile hydration, real-time monitoring

Required for Billing

Variable	Service	What It Does
POLAR_ORGANIZATION_TOKEN	Polar	Subscription management
POLAR_PRODUCT_BASE_MONTHLY	Polar	Base tier monthly product ID
POLAR_PRODUCT_BASE_YEARLY	Polar	Base tier yearly product ID
POLAR_PRODUCT_PRO_MONTHLY	Polar	Pro tier monthly product ID
POLAR_PRODUCT_PRO_YEARLY	Polar	Pro tier yearly product ID

Optional

Variable	Service	What It Does
OPENAI_API_KEY	OpenAI	Fallback for embeddings if OpenRouter is down
EXA_API_KEY	Exa	URL content extraction (falls back to HTML parsing)
BISHOPI_API_KEY	Bishopi	Keyword discovery from seed terms
NEXT_PUBLIC_POSTHOG_KEY	PostHog	Product analytics
NEXT_PUBLIC_SITE_URL	—	Custom domain (defaults to localhost:3000)

Run

npx convex dev   # terminal 1 — starts Convex backend
pnpm dev         # terminal 2 — starts Next.js dev server

Open http://localhost:3000.

▪️ Tech Stack

Layer	Technology
Framework	Next.js 16 (App Router, PPR, React 19.2)
Database	Convex (reactive real-time sync, workflows, workpools)
AI	AI SDK 5.0 via OpenRouter (Gemini 3, Kimi K2, Claude Haiku 4.5, GPT-5.4 Nano, and more)
Agents	@convex-dev/agent with streaming, tool calling, and thread management
RAG	@convex-dev/rag with OpenRouter embeddings (text-embedding-3-small)
Auth	WorkOS AuthKit
UI	shadcn/ui + Tailwind CSS 4
X (Twitter)	X API v2 (xdk) + SocialAPI (search, monitoring, hydration)
LinkedIn	LinkdAPI (search, profiles) + Unipile (messaging, actions)
Rich Text	Lexical (composer with inline autocomplete)
Billing	Polar
Email	Resend + React Email

▪️ Development

pnpm dev          # Start development server
pnpm build        # Build for production
pnpm start        # Start production server
pnpm lint         # Check for linting issues

▪️ Current Status

Status	Feature
✓	AI-powered discovery across X and LinkedIn
✓	8 use-case templates with adapted UI, prompts, and pipeline stages
✓	Two-agent system (Setup + Outreach) with chat interface
✓	Inline autocomplete (tab-complete) for replies and DMs
✓	Writing style learning from your social profiles
✓	Workspace-level memory with RAG-backed retrieval
✓	AI qualification (ICP fit scoring, bot detection)
✓	AI enrichment (pain points, financial signals, role detection)
✓	Agentic outreach with plan generation
✓	Human-in-the-loop approval for all social actions
✓	X engagement execution (replies, DMs, likes, retweets, follows)
✓	LinkedIn actions (messages, invites, reactions, comments)
✓	Real-time monitoring and conversation seed detection
✓	Agent Ops dashboard (discovery, quality, memory, activity)
✓	Discovery trace and query performance tracking
✓	Pipeline management per use case
✓	Analytics dashboard
✓	Notifications and approval request alerts
✓	Billing and subscription management
○	Full LinkedIn parity with X
○	Additional social platforms

▪️ Contributing

Contributions are welcome. Whether you find bugs, have feature ideas, want to submit PRs, or have AI/ML expertise to share — all skill levels welcome.

See CONTRIBUTING.md for setup instructions and guidelines.

▪️ Contact

Built by @noobships

▪️ License

MIT License — use it however you want.

---

Like this project? Give it a star.