Replace your entire Go-To-Market SaaS stack with one self-hosted AI platform.
6sense + Clay + Outreach + Gong → $0/month, runs on your own infra.
Most B2B companies pay $80,000–$150,000/year for a stack of GTM SaaS tools that don't talk to each other, lock your data behind APIs, and charge per seat. AI GTM Engine replaces them with five autonomous AI agents that run on your infrastructure, use your LLM API keys, and put you back in control.
6sense → ICP Agent (lead scoring + ideal customer profiling)
Clay → Enrichment (Apollo.io + Hunter.io pipeline)
Outreach → Outbound Agent (3 personalized email variations per lead)
Gong → Deal Intel Agent (transcript analysis + deal risk scoring)
ChurnZero → Retention Agent (health scoring + churn prediction)
Everything is human-in-the-loop. Agents draft, humans approve. No email ever sends without a click.
- 5 AI Agents — ICP scoring, outbound personalization, content generation, deal intelligence, retention analysis
- Human Approval Queue — review AI drafts side-by-side before anything goes out
- CRM Sync — bidirectional HubSpot and Salesforce sync, every 4 hours
- Data Enrichment — Apollo.io + Hunter.io pipeline, auto-triggered on new leads
- Any LLM — OpenAI, Anthropic Claude, Ollama (local), or any LiteLLM-supported provider
- Real-time Dashboard — Next.js frontend with live agent status via SSE
- Self-hostable — one
docker-compose upcommand, no cloud lock-in - Multi-tenant — built-in org isolation, RBAC (admin / member / viewer)
- Observability — Prometheus metrics, Grafana dashboards, Langfuse LLM tracing
- Open source — MIT license, all data stays in your Postgres
┌─────────────────────────────────────────────────────────────┐
│ Browser │
└──────────────────────────┬──────────────────────────────────┘
│
┌─────▼─────┐
│ Nginx │ rate limiting, SSL, static files
└─────┬─────┘
┌────────────────┼────────────────┐
│ │ │
┌─────▼──────┐ ┌──────▼──────┐ │
│ Next.js │ │ FastAPI │ │
│ Frontend │ │ Backend │ │
└────────────┘ └──────┬──────┘ │
│ │
┌────────────┼───────────┐ │
│ │ │ │
┌─────▼────┐ ┌─────▼─────┐ ┌──▼───▼──┐
│ Postgres │ │ Redis │ │ Qdrant │
│+pgvector │ │ (cache + │ │(vector │
│ │ │ broker) │ │ search) │
└──────────┘ └─────┬─────┘ └─────────┘
│
┌──────▼──────┐
│ Celery │ async jobs, scheduled tasks
│ Workers │
└─────────────┘
Stack: Python 3.11 · FastAPI · LangGraph · LiteLLM · SQLAlchemy 2.0 · Alembic · Celery · Next.js 14 · TypeScript · Tailwind · shadcn/ui · PostgreSQL 16 + pgvector · Redis 7 · Qdrant
Prerequisites: Docker 24+, Docker Compose 2.23+, and at least one LLM API key.
# 1. Clone
git clone https://github.com/sohan-a11y/gtm-engine.git
cd gtm-engine
# 2. Copy the environment template
cp .env.example .env
# 3. Fill in required secrets
# JWT_SECRET → openssl rand -hex 32
# ENCRYPTION_KEY → python -c "from cryptography.fernet import Fernet; print(Fernet.generate_key().decode())"
# OPENAI_API_KEY → or ANTHROPIC_API_KEY
# 4. Start the full stack
docker compose up --build
# 5. Open the dashboard
open http://localhost:3000That's it. The full stack is up in under 2 minutes.
With observability (Prometheus + Grafana + Langfuse):
docker compose --profile observability up --build
See docs/getting-started.md for the first-run checklist and troubleshooting guide.
| Platform | Button |
|---|---|
| Railway |  |
| Render |  |
Both platforms read the railway.json / render.yaml in the repo root. You'll need to set LLM API keys and other secrets in the platform's environment variable dashboard after deploy. See docs/deployment.md for full instructions.
Trains on your best customers and scores every new lead against that profile using vector similarity + LLM reasoning. Returns a 0–1 score with fit signals and gap signals.
# Train on your best customers
POST /agents/icp_agent/score { "mode": "train", "profiles": [...] }
# Score a new lead
POST /agents/icp_agent/score { "contact_profile": { "title": "VP Sales", ... } }
# → { "score": 0.87, "explanation": "...", "fit_signals": [...], "gap_signals": [...] }Generates 3 email variations per lead, each with a different personalization hook (recent news, job posting, tech stack change, pain point). Word-count enforced under 150. Never mentions AI.
POST /agents/outbound_agent/generate
# → 3 variations go to approval queue, nothing sends without human approvalAnalyzes CRM stage data, email threads, and call transcripts. Returns risk score, risk factors with evidence, and recommended actions with owner and urgency. Fires Slack alerts when risk > 0.75.
Classifies accounts into 5 health states (Thriving → Critical), identifies the primary churn driver, and suggests the right playbook. Monitors contract renewal dates.
Three modes: SEO_BLOG_POST, EMAIL_SEQUENCE, LINKEDIN_POST. Respects your brand voice config. Auto-flags quality issues: too_salesy, missing_hook, over_length, weak_cta.
| Category | Supported |
|---|---|
| LLM | OpenAI GPT-4o, Anthropic Claude, Ollama (local), any LiteLLM provider |
| CRM | HubSpot, Salesforce |
| Gmail (OAuth), Outlook / Microsoft 365 | |
| Enrichment | Apollo.io, Hunter.io |
| Notifications | Slack webhooks |
The Next.js dashboard includes:
- Pipeline view — deal kanban with risk indicators
- Leads list — ICP score badges (green/amber/red), enrichment status
- Approval queue — side-by-side email variation review with edit + approve/reject
- Analytics — 4 tabs: Pipeline, Outbound, Retention, AI Usage & Cost
- Agent cards — status, last run, trigger button, run history
- Settings — LLM config, CRM connections, brand voice, team management
All configuration is via .env. See .env.example for every available option with inline documentation, or read the full configuration reference.
Minimum required:
JWT_SECRET=<openssl rand -hex 32>
ENCRYPTION_KEY=<python -c "from cryptography.fernet import Fernet; print(Fernet.generate_key().decode())">
OPENAI_API_KEY=sk-... # or ANTHROPIC_API_KEYOptional integrations (agents fall back gracefully without these):
HUBSPOT_CLIENT_ID=...
APOLLO_API_KEY=...
HUNTER_API_KEY=...
GMAIL_CLIENT_ID=...
SLACK_WEBHOOK_URL=...Full PostgreSQL schema managed by Alembic:
organizations → users → contacts → email_sequences
→ companies → deals
campaigns → email_sequences
agent_audit_log (append-only, immutable)
agent_configurations
integrations
job_runs
Materialized views refreshed every 5 min: mv_pipeline_summary, mv_outbound_performance.
# Run migrations
make migrate# All tests
make test
# Backend only
cd backend && pytest
# Frontend only
cd frontend && npm testTests use an in-process SQLite database — no running Postgres needed for unit tests. Integration tests spin up real Postgres and Redis via GitHub Actions services.
gtm-engine/
├── backend/
│ ├── agents/ # 5 AI agents (ICP, Outbound, Content, Deal Intel, Retention)
│ ├── api/ # FastAPI routers + Pydantic schemas
│ ├── core/ # Auth, encryption, LLM router, rate limiting, circuit breaker
│ ├── db/ # SQLAlchemy models, repositories, Alembic migrations
│ ├── enrichment/ # Apollo + Hunter enrichment pipeline
│ ├── integrations/ # CRM, email, notification adapters
│ ├── prompts/ # Agent prompt templates (.txt files)
│ ├── services/ # Business logic layer
│ └── workers/ # Celery tasks + beat schedule
├── frontend/
│ ├── app/ # Next.js App Router pages
│ ├── components/ # React components (agents, leads, approvals, charts, ...)
│ └── lib/ # API client, Zustand store, TanStack Query hooks, types
├── infrastructure/
│ ├── nginx/ # Nginx config (rate limiting, SSE support, security headers)
│ ├── prometheus/ # Scrape config
│ └── grafana/ # Dashboards + datasource provisioning
├── docker-compose.yml # Full stack (9 services)
├── Makefile # Dev workflow shortcuts
└── .env.example # Every config option documented
| Dashboard | Pipeline Kanban |
|---|---|
 |
 |
| Approval Queue | Analytics |
|---|---|
 |
 |
See docs/screenshots.md for the full gallery.
| Guide | Description |
|---|---|
| Getting Started | Prerequisites, 5-step setup, first-run checklist, troubleshooting |
| Configuration | Every environment variable documented with defaults |
| Agents | How each of the 5 agents works, inputs, outputs, training |
| Integrations | HubSpot, Salesforce, Gmail, Outlook, Apollo, Hunter, Slack setup |
| Deployment | Docker Compose, Railway, Render, scaling, backups |
| API Reference | Key REST endpoints grouped by domain |
- JWT access tokens (15 min) + refresh tokens (7 days, httpOnly cookie)
- AES-256-GCM field-level encryption for all stored credentials
- RBAC: admin / member / viewer roles with explicit permission sets
- Org-scoped multi-tenancy — every query is filtered by
org_idfrom the JWT - HMAC-SHA256 webhook verification
- Append-only audit log for all agent actions
- Rate limiting at Nginx (IP) and application (user/org) levels
- CORS strict origin whitelist
- LinkedIn enrichment via Sales Navigator API
- Pipedrive CRM integration
- Zoom / Gong transcript sync (automatic ingestion for Deal Intelligence Agent)
- A/B testing for email variations with statistical significance tracking
- Mobile app (React Native) for approval queue on the go
PRs are welcome. Please open an issue first for large changes.
# Setup dev environment
make setup
make dev
# Lint before submitting
make lintSee CONTRIBUTING.md for code style and commit conventions.
MIT — do whatever you want, including running it in production for your company.
Built by Sai Sohan Merugu · Artifex · Hyderabad
If this saves you money on your GTM stack, please ⭐ the repo — it helps more people find it.