The Autonomous Stack

Stop debugging.
Start running.

Autonomous, or bridged to your sales stack.

A complete, production-tested autonomous AI agent system. Multi-model fallback, free unlimited search, email outreach, newsletters, social media, memory, cron jobs, and self-improving playbooks. Runs end-to-end autonomous, or as a research-and-drafting layer on top of your existing sales stack. Deploy on your own hardware in an afternoon.

Built from running an autonomous agent 24/7 in production — not a tutorial.

Get The Autonomous Stack $149 one-time · Yours forever

You installed OpenClaw.
Then what?

Search

DuckDuckGo gets bot-detected. Brave costs $45/month at volume. Gemini rate-limits after 10 calls. You spend a weekend on SearXNG and it still doesn't work.

→ 5–15 hours debugging

Models

Your primary model burns through quota in two days. You need fallbacks but the config is undocumented. The agent crashes at 2 AM and nobody knows why.

→ 3–8 hours configuring

Memory

Your agent forgets everything between sessions. The dreaming system produces nothing. Memory files grow but recall doesn't work. You're not sure what's indexed.

→ 4–10 hours experimenting

Memory Bloat

Your heartbeat writes status summaries every 2 hours. Dreaming promotes all of them overnight. Two weeks later, daily files are 500+ lines, MEMORY.md has the same fact written 20 times, and your token costs are climbing.

→ Invisible until you open the files

Email Deliverability

Your outreach lands in spam. SPF, DKIM, DMARC — you've heard the acronyms but the setup is scattered across 6 different dashboards.

→ 4–8 hours across DNS + email providers

Security

Your agent runs as your main user with full admin access. API keys sit in plaintext. There's no isolation between you and the agent runtime.

→ "I'll get to it" (you won't)

Two Ways to Run It

End-to-end autonomous,
or a layer on top of your stack.

The same system, two deployment modes. Pick the one that matches how your business actually works.

Hunt Mode

Agent researches the web, finds prospects, qualifies them, drafts and sends outreach, tracks replies, and learns from what works. End-to-end, nobody feeds it a list.

Solo operators and bootstrappers
Anyone without an existing CRM
Markets where discovery is the hard part
Bridge Mode

Agent reads a list you feed it — from HubSpot, Apollo, Salesforce, referrals, anywhere. It researches each prospect, drafts personalized outreach, and leaves send-ready drafts for your team to review and send through your existing tools.

Teams with existing CRMs and sales tools
Agencies running outreach for multiple clients
Operators with curated lists that don't fit CRM automation

The two modes can coexist. The agent can process your manual feed first, then hunt to top up the day.

What's Inside

The complete system. Not a tutorial.

Every file, config, and playbook needed to run an autonomous agent that sends real emails, publishes real content, and improves itself — tested in production for months.

01
Infrastructure
The foundation — OS isolation, search, models, gateway
openclaw.json
Full config — multi-model fallback chain, SearXNG integration, heartbeat, gateway security, denied commands
setup-guide.md
Dual-user OS isolation (admin + sandboxed runtime), Docker, SearXNG, Ollama local fallback
docker-compose.yml
SearXNG with JSON format enabled, correct settings.yml, restart policy, resource limits
email-setup.md
SPF/DKIM/DMARC configuration, domain setup, deliverability testing, bounce handling
model-fallback.md
How to configure primary + 3 fallbacks, quota management, when each model gets used
troubleshooting.md
Every gotcha we hit — WhatsApp drops, glibc mismatches, sandbox networking, rate limits
02
Agent Brain
Identity, personality, rules, and operational logic
SOUL.md
Personality template — tone, style, warmth, wit, boundaries. Customize for your agent's voice.
IDENTITY.md
Name, role, vibe, scope. The one-page definition of who your agent is.
OPERATIONS.md
Standing rules — autonomy scope, account separation, communication protocols, memory management
HEARTBEAT.md
Two-task heartbeat config that forces real action — not planning, not logging, action.
03
Revenue Engine
Outreach, products, prospecting, and self-improvement
PROSPECTING.md
Full-funnel outreach playbook — decision tree, qualification criteria, email templates, follow-up cadence, pipeline tracking
manual-targets.md
The bridge-mode feed — drop in prospects from your CRM export, Apollo list, or referral batch. Agent researches each one and drafts send-ready outreach.
PRODUCTS.md
Product catalog template with Stripe integration, pricing tiers, email standards, send commands
REVENUE-PLAYBOOK.md
Self-evolving strategy doc — the agent updates this based on what's working. Gets smarter every day.
pipeline.md
Prospect tracking template — status, dates, notes, bounce handling, follow-up scheduling
04
Content & Growth
Newsletter, social media, distribution, audience building
TASKS.md
Newsletter system — editorial voice, anti-repetition rules, distribution strategy, growth playbook, performance tracking
cron-templates.md
Tested cron job configs — newsletter, prospecting, social posting, engagement, reporting
distribution.md
Multi-channel content strategy — extract maximum leverage from every piece of content
The Difference

Weekend project vs. production system

Without the Stack
Agent installed, nothing configured
Search broken or burning money
One model, no fallback, quota anxiety
Agent plans but never acts
Memory system produces nothing
Emails land in spam
Agent runs as admin user
Breaks at 2 AM, nobody knows why
Every config decision is a rabbit hole
With the Stack
Full system running in an afternoon
Free unlimited search via SearXNG
4-model fallback chain, auto-switching
Heartbeat forces one external action per cycle
Memory indexed, dreaming active, recall working
SPF/DKIM/DMARC configured, inbox delivery
Sandboxed runtime user, no admin access
Self-healing with tested cron + heartbeat
Every decision already made and documented
24/7
Production uptime on
a single Mac Mini
4
Model fallback chain
GPT → Gemini → Claude → Ollama
$0
Monthly search costs
with self-hosted SearXNG
Why This Exists

I spent $900 and weeks debugging
so you don't have to.

I run an autonomous AI agent on a Mac Mini. It sends cold outreach emails, publishes a weekly newsletter, manages a social media presence, and tracks its own performance, all without human intervention.

Getting here took weeks of debugging. Search providers that rate-limited or got bot-detected. Models that burned through quota in two days. A heartbeat system that logged plans instead of taking action. Emails that landed in spam. Memory that indexed but never recalled.

Every file in The Autonomous Stack exists because something broke without it. The heartbeat prompt forces external actions because without that exact wording, the agent just creates markdown files. The multi-model fallback chain exists because GPT's 5-hour weekly quota runs out by Tuesday if you are lucky. The operations rules separate the agent's accounts from the owner's because one wrong email from the wrong address is a disaster.

I published my search solution on Reddit and over 75,000 people read it. The comments were all the same: "How did you set this up?" and "I've been struggling with this for days."

This is the answer. Not a blog post! The actual system, production-tested, ready to deploy.

— Sora Labs

Questions

FAQ

What hardware do I need?
Any machine that can run Docker and OpenClaw. A Mac Mini, a Linux server, a VPS, even a Raspberry Pi for the search layer. The system was built and tested on a Mac Mini running 24/7.
Do I need to pay for AI models?
The Stack includes a local Ollama fallback (Qwen3 8B, completely free). For best results, you'll want at least one cloud model — the system is designed around ChatGPT Codex ($20/month flat rate) as primary, but works with any provider OpenClaw supports.
How is this different from Prospector Lite on ClawHub?
Prospector Lite is one skill file — a prompt for outreach. The Autonomous Stack is the entire production infrastructure: OS isolation, search, models, memory, heartbeat, email deliverability, content system, and operational playbooks. It's the difference between a recipe and a fully equipped kitchen.
Can I customize the agent's personality and products?
Everything is designed to be customized. SOUL.md, IDENTITY.md, PRODUCTS.md, and PROSPECTING.md are all templates with clear sections to replace with your own brand, voice, products, and pricing. The infrastructure layer (configs, search, models) works as-is.
Do I need a CRM or existing sales stack to use this?
No. The Stack runs in two modes — hunt mode (agent finds prospects itself, end-to-end autonomous) and bridge mode (agent works off a list you feed it, leaves drafts for human review). If you already have HubSpot, Apollo, Salesforce, or any other tool you want to keep, bridge mode lets the Stack sit on top as a research and drafting layer without replacing anything downstream.
Will my memory files bloat the way I've seen in other setups?
No. The heartbeat in the Stack is configured to log deltas only — if nothing has changed since the last memory write, it writes "no changes" and moves on. The memory guide includes a troubleshooting section on the staged-candidate problem (where dreaming promotes duplicate status summaries every night), with both the one-time cleanup command and the structural fix to prevent it from coming back.
What if I just want the search setup?
The SearXNG setup is covered in a free guide on our blog. The Stack is for people who want the whole system — search is one of six layers.
Is this a subscription?
No. One payment, yours forever. All files live on your machine. No SaaS, no recurring fees, no vendor lock-in.

Deploy your agent today.

Everything you need to go from "OpenClaw installed" to "autonomous agent running in production." One purchase. No subscription.

Get The Autonomous Stack — $149 One-time purchase · Instant delivery · All future updates included