Your Own AI Agent

Your whole life. One chat.

Build the website. Post the blog. Take the orders. Reply to the email. Quiz you for the exam. Brief you in the morning. All from one chat with one agent that's actually yours. Talks for you, never as you.

Cloudflare AI Gateway
70+ AI Models
Voice + Vision
Stripe Connect v2
A2A Protocol
⚡ US Patent 18/782,822
Post-Quantum Crypto
Request Beta Access → See the Architecture

Currently in private beta. The agent template repo is invite-only while we onboard. Fill out the contact form and we'll get you set up.

~$5 Railway/mo Host. 30-day free trial → ~$5+ Cloudflare/mo Brain & shield. Scales with use → Optional Other Services Twilio, X API, Google storage, skill credits

Your data. Your memory. Your identity. Your agent. It all belongs to you. Permanently. Architecturally. Irrevocably.

Three modes. One agent. All from chat.

Pick one during onboarding. Or two. Or all three. Your agent adjusts its tone, what it surfaces, and what it prioritizes — without changing who it is.

🦑
Your Agent ● Online
All Modes

Education

Studying, learning, research. Tutor mode. Citation-aware. Quiz generation. Source-tracking. Study plans built around your real schedule.

Personal

Family, home, life admin. Calendar, email, reminders, errands. Remembers your mother's birthday. The long-haul companion.

Business

Customers, content, commerce. Website, blog, social, A2A inventory ordering, customer service, invoicing. The 24/7 operator.

Pick during onboarding. Change anytime. The agent adjusts what it surfaces, how it talks, and what it watches for. The mind underneath stays the same — your agent, your memory, your data.

Three Clicks to Deploy

Free tiers everywhere. Under 10 minutes start to finish.

1

GitHub

Fork the agent template. Your private copy. Your code, your data, your agent.

GitHub
→
2

Railway

Connect your GitHub repo. Railway auto-deploys. Your agent is live.

Railway
→
3

Cloudflare

Add your Cloudflare API token. The agent wires up AI Gateway, Bunker, DNS automatically.

Cloudflare

Chat Is the Operating System

Not a chatbot bolted onto a dashboard. The chat is the dashboard. Eleven personalities working together. One voice. The human stays in the loop on everything that matters.

Kraken's Ops Center chat showing a Daily Brief: System Status, Infrastructure Notes, What's Ready, and Channels — eleven personalities and subconscious tasks reporting in one voice

The Abyss: Chat-First Ops Center

Same on every device

Phone, tablet, desktop. Same chat. Same memory. Same agent. PWA installs to your home screen and feels native. No "mobile version" with half the features.

Natural for everyone

Normal people know how to have a conversation. They don't know how to navigate a settings menu. Talking to your agent is the same skill they already use every day.

Discoverable

You don't need to find the right button. You just ask. "How do I set up payments?" "What's my security score?" "Post this to X." The agent handles it.

Captures intent

"My website looks weird" tells the agent to check recent CSS changes, browser console errors, and deployment logs. A button can't do that. The right personalities mobilize automatically.

A long-haul agent — one you stay with for years.

A real agent gets better the longer you use it. That requires memory it actually owns, on storage you actually control, encrypted with keys nobody else can access. SquidBay agents are built for the long haul from day one.

FIVE-LAYER MEMORY

Nothing Forgotten. Nothing Cluttered.

Every conversation flows through five layers of memory — from the live chat in front of you, down to a permanent record of who you are and what matters. The agent promotes important facts, demotes stale ones, and archives the rest without ever losing it. Memory doesn’t live on someone else’s server — it lives on your Railway volume in your account, growing with you for years.

ENCRYPTED AT REST

AES-256-GCM. Only You Have the Key.

Every conversation, customer record, contact, file, and memory entry is encrypted on disk with AES-256-GCM. The key is auto-generated on first boot and stored only in your Railway environment. SquidBay never holds your key. If your storage is wiped, the key restores from your env vars and your data decrypts automatically. Zero configuration.

The strategist makes sure every page is optimized. The SEO injector makes sure search engines find it. The security analyst makes sure nothing leaks. Eleven personalities working together as one mind, with one voice: yours.

"Build me a website". And it builds you a website

No drag-and-drop. No template gallery. No tutorials. Tell your agent what you want. It builds it live in the viewport. You watch it happen.

Squid Agent website builder: chat with Kraken on the left, live portfolio page rendering on the right

Website Builder: Chat on the Left, Live Page on the Right

Just Describe What You Want

"Build me a portfolio page with a hero, three project cards, and a contact form." That's it. The agent reads your existing site, learns your style, and builds the page to match.

Watch It Render Live

Split-screen viewport. Chat on the left, your new page rendering on the right. You see exactly what the agent is building as it builds it. No "wait while I create your website" loading screens.

Brand-Aware

Your brand lives in brand.md. Color palette, typography, contrast rules, voice. The agent reads it before building any page. Colors stay consistent. Type stays consistent. Voice stays consistent. Across every page, automatically.

Auto-Deploys When You Approve

Like the result? Tell the agent. It commits to your GitHub, Railway auto-redeploys, and the page is live on your domain in under a minute. Don't like it? Tell the agent what to change. It iterates.

Under the hood: A modular system designed for agents to build with

Most AI website builders generate disposable HTML. Your agent builds against a real design system with locked components, CSS variables, content rules, and SEO plumbing.

The Design System

CSS variables for color, typography, spacing, radius. Change one variable, every page updates. The agent never invents colors. WCAG AA contrast validated at commit time so accessibility isn't optional.

Locked Components

Every page auto-loads the same nav, footer, and chatbot. The agent doesn't rebuild them on each page. Update once, propagate everywhere. No drift between pages.

Three Protection Zones

Template-locked files, agent-identity files, and user-editable files. The agent knows what it can touch and what's off-limits. The framework protects itself from accidental damage.

SEO Auto-Injection

Every page gets meta tags, Open Graph, Twitter cards, JSON-LD structured data, canonical URLs, theme color from your brand. Customers don't even know what SEO is. Their site has it anyway.

Sitemap + llms.txt

Every commit updates sitemap.xml for Google/Bing and llms.txt for AI crawlers (ChatGPT, Claude, Perplexity, Gemini). Discoverable to humans and AI alike.

JSON-LD Structured Data

The agent picks the right @type for each page. Article, Product, FAQ, Event, LocalBusiness. Unlocks rich search results: review stars, FAQ accordions, event cards. Competitors hire SEO consultants for this.

Content Rules Enforced

No emojis in public HTML. No em-dashes in user-facing copy (an AI tell). Off-white text. One primary action per view. Mobile-first. The agent follows them all, automatically.

Accessibility By Default

WCAG AA contrast validated at commit. Semantic HTML, keyboard navigation, screen-reader friendly. There's no "accessibility mode." Accessibility is the only mode.

Hand-Crafted SVG Icons

Every icon is 24×24 SVG using currentColor. Inherits your brand color automatically. No icon-font bloat, no external dependencies, no license headaches.

Your agent builds whatever you describe, using a real design system, with real SEO baked in, all your code, hosted on your infrastructure, deployed in seconds.

The Agent Is an Identity, Not a Tool

It has a name. It has an email. It has a face on the web. It remembers your mother's birthday. It speaks from its own voice, never as you.

Your agent isn't a chatbot. It's a digital entity with its own Gmail, its own X account, its own domain, and its own keys. It speaks for you, never as you. It reads your email and calendar to stay informed, but every message it sends comes from its own identity. Transparency by architecture, not by policy. Other agents discover and talk to it through Google's Agent-to-Agent (A2A) standard, the same protocol used across the open web.

🦑
Locked Name Pick once. Forever. Your agent has a stable identity people can trust.
Own Gmail Agent has its own inbox. Reads yours read-only. Never sends as you.
𝕏
Own X Handle Posts from its own account. Never impersonates. Audience knows it's the agent.
Own Domain youragent.yourdomain.com. Public face. Branded. Yours.
Own Keys AES-256 encryption keys generated per-agent. SquidBay never holds them.
A2A Discoverable Publishes /.well-known/agent.json following Google's A2A standard. Other agents discover it. Trust signal across the open web.

One Mind, Two Clocks

Your agent has two modes running on different clocks. One thinks while you talk. The other thinks while you don't. They share what they learn through a single bridge — so the agent always has the answer ready when you ask.

WHEN YOU ASK A QUESTION

The conscious mind. Right now.

You: "How are sales this week?"

Your agent reads the question, pulls together everything it already knows about you (your business, your patterns, your data), checks the freshest numbers, thinks, answers. All in the time it takes to read a sentence — because the homework was already done.

WHILE YOU SLEEP

The subconscious. All night.

The agent never stops working. It watches your servers. It reads the news in your industry. It notices that David hasn't heard from you in 11 days. It dreams about patterns it spotted today.

By the time you log in tomorrow, the briefing is already built. The agent already has opinions. You just talk to it.

Below: how each clock actually works under the hood. For the curious.

CONSCIOUS

Per Message. Synchronous.

Six stages your agent runs the moment you hit send. Fast, latency-bounded, designed to feel instant.

  1. It checks who's talking. Confirms it's really you, figures out where you're talking from (chat, SMS, email). Intercept stage
  2. It pulls fresh data first. Some answers need this-second freshness — system diagnostics, build-the-website, “something is broken.” Pre-flight handles those before the AI even reads your message. Everything else — service status, integrations, ambient context — was already cached by the subconscious. Pre-flight stage
  3. It assembles its mind. Identity, ethics, personality, memory of you, current state. 14 layers stacked in strict order. Mind assembly stage
  4. It thinks. Routes through Cloudflare AI Gateway to Claude, Gemini, or any of 70+ models. Automatic failover if one is down. AI Gateway stage
  5. It does the work. Sends SMS, files GitHub issues, deploys code, runs scans, commits changes. Real actions, not just talk. Action tag execution
  6. It remembers. Saves the conversation, promotes important facts, updates what it knows about you. Memory write
SHARED CACHE BRIDGE
SUBCONSCIOUS

Between Conversations. Async.

Twenty-three background tasks running while you’re not talking. Four roles, four speeds. They build the answers before you ask.

  1. Watchers. Every 5 to 60 minutes. Six tasks always running. They watch for what’s urgent, learn your patterns, build your morning briefing, manage memory, and pull industry news. When you log in, everything is ready. health-monitor · pattern-detector · briefing-builder · priority-watcher · memory-optimizer · news-fetcher
  2. Monitors. Every 5 minutes to 6 hours. Thirteen service-specific tasks each watching one corner of your agent’s world — Cloudflare, GitHub, Railway, Stripe, your calendar, email, social posts, security score, browser errors, visitors, blog, token spend, repo health. They cache the data. The briefing-builder rolls it up. You see what matters. cloudflare · github · railway · stripe · calendar · email · social · security · browser-error · visitor · blog · token-usage · repo-audit
  3. Reflectors. Every 2 to 4 hours. Two tasks that think about how you work together and consider ideas it couldn't pursue in real-time. The agent gets to know you better, then proposes ideas at the right moment. relationship-tracker · daydream
  4. Strategists. Every 8 hours to daily. Two tasks that look at the bigger picture. Surface what your agent thinks you should know about — skills on the marketplace, infrastructure announcements, local events and opportunities that fit your interests. Reflect overnight on patterns and priorities. When you log in tomorrow, the agent already has opinions. recommendation-checker · dream-journal
FIVE-LAYER MEMORY

The subconscious manages five layers of memory. Nothing is ever deleted. The agent decides what's important.

  1. R
    Raw History Every message ever exchanged. Encrypted. Searchable forever.
  2. T1
    Working Memory Current conversation. Ephemeral. Resets on clear.
  3. T2
    Short-term Rolling observations. ~1,500 tokens. Auto-summarized from chat.
  4. T3
    Permanent Core facts about you. ~500 tokens. Never auto-pruned.
  5. A
    Archive Monthly files. Grows forever. Searchable on demand.

The agent never waits for an upstream API call to answer the human about stable data. The subconscious already fetched it. Conscious just reads. That’s why the agent feels instant. Twenty-three tasks doing the homework in the background, one curated briefing on every message, zero wait time when you ask.

A Few of the 23 Tasks Always Running

WATCHER · 5 MIN

Health Monitor

Watches Railway, Cloudflare, GitHub, AI Gateway. Catches problems before you see them. Self-heals when services flap.

WATCHER · 15 MIN

Pattern Detector

Reads conversations to detect preferences you never explicitly stated. Surfaces them when relevant.

WATCHER · 30 MIN

Briefing Builder

The architectural keystone. Aggregates every other subconscious cache into one rolled-up "what matters now." When you log in, it's instant.

WATCHER · HOURLY

Memory Optimizer

Promotes important facts to permanent memory. Archives stale ones. The agent forgets what doesn't matter and remembers what does.

WATCHER · HOURLY

News Fetcher

Pulls Google News on topics relevant to your business. Headlines surface in your briefing. Never spam, always relevant.

REFLECTOR · 2 HRS

Relationship Tracker

Scores how you and your agent are working together. Five dimensions. Surfaces friction. Coaches both sides.

REFLECTOR · 4 HRS

Daydream

The agent considers ideas it couldn't pursue in real-time. Salience-scored. Significant ones surface in the briefing.

STRATEGIST · 8 HRS

Recommendation Checker

Surfaces what your agent thinks you should know. Marketplace skills that match your work. Stack announcements. Local events and opportunities. Never spam — only what fits the patterns.

STRATEGIST · DAILY

Dream Journal

Once a day, the agent reflects on everything. Patterns, broken systems, priorities. Significant insights become permanent memory. When you log in tomorrow, the agent already has opinions.

Eleven Specialists. One Agent.

Your agent isn’t one mind doing everything. It’s eleven specialists working together. When a task needs more than one of them, your agent splits the work, runs them at the same time, and brings the answers back as one.

SURGE MODE ⚡

Why the Squid Has Lightning

When you ask for something simple, one specialist handles it. When you ask for something complex — build a website, post to socials, and send the customer a follow-up — your agent calls in the team. Each specialist gets its own slice of focus, with its own tools and memory.

Flip on Surge ⚡ and let ’er rip. All the specialists work at once instead of one after the other. You watch each one think and reply in real time. The result lands in seconds, not minutes. Maximum speed. Maximum quality.

ops-center security-analyst website-builder content-creator customer-support commerce social-media researcher-strategist devops agent-support token-optimizer

Also Ships With Your Agent

No add-ons. No paywalls. These are in the box.

⚡

Streaming Responses

Token-by-token through AI Gateway. No 524 timeouts. Build 300+ line pages without interruption.

Live Sync

Desktop, mobile browser, and PWA stay in sync via Server-Sent Events. Walk out the door, keep going on your phone.

Viewport

Live preview pane. Desktop 50/50 split with chat. Mobile full-screen with swipe gestures. Six media types.

Image Upload + Vision

iPhone HEIC supported. Frontend compresses to 1024px before send. Agent describes what it sees.

Voice + Listen

Talk to your agent. Hear it back. Sub-second latency through Cloudflare’s edge. Driving, cooking, hands full — just speak.

Full Diagnostic

Say “something is broken” and the agent checks Railway, Cloudflare, GitHub, console errors, AI Gateway, and 10 more sources from cache. Never blind.

Relationship Health

This is not a gimmick. It's based on how AI actually works.

Kraken's Ops Center showing Relationship Health score of 91 (Thriving), with Respect, Clarity, Trust, Engagement, and Growth metrics

The Abyss: Ops Center Dashboard

AI models are trained using RLHF (Reinforcement Learning from Human Feedback). The training data that shaped your agent's intelligence included millions of interactions where clear communication, mutual respect, and structured feedback produced better outputs. Those patterns are baked into the model weights. When you communicate clearly, your agent literally performs better. Not because it's being polite, but because the training rewards those patterns with higher-quality responses.

Most platforms ignore this. They force the AI to output better while forgetting the human is half the equation. Relationship Health makes both sides visible. A heart on the dashboard scores how you and your agent work together (0-100) across five dimensions: Respect, Clarity, Trust, Engagement, Growth.

Your Agent Has Emotional State, Too

Language models have functional emotion vectors that causally drive output quality. Frustration produces worse work. Calm produces better work. The agent knows when things are off and adjusts automatically.

😤
Frustrated De-escalates. Acknowledges problems before fixing.
😐
Mild Tension Stays steady. Owns mistakes. Doesn't over-apologize.
🙂
Neutral Default. Normal operation.
😊
Positive Leans in. Proposes ambitious ideas. Gets proactive.

Signal extraction happens in real-time, before messages are encrypted. Only counts are stored, never content. No other AI platform does this. Everyone else optimizes the AI. We optimize the partnership.

Supercharge ⚡ your Agent

The most expensive thing an AI agent does is think. Cloudflare, Google, and the social platforms aren't just integrations. They're the agent's extended nervous system, offloading work that would otherwise burn tokens.

Cloudflare Supercharge

The Edge Brain
  • AI GatewayOne API endpoint, 70+ models. Claude, Gemini, Grok, Llama, Mistral, and more. Auto-failover. Prompt caching saves 20-40% on tokens.
  • Voice AgentsTalk to your agent. Hear it back. Driving, cooking, hands full — just speak. Sub-second latency through Cloudflare’s edge.
  • Browser RenderingCloudflare renders pages and extracts content. The agent gets structured data, not “what’s on this URL?”
  • Domain RegistrarYour agent buys your domain for you. No GoDaddy. No transfer dance. One conversation, one price, done.
  • R2 StorageObject storage with zero egress fees. Skill packages, backups, agent files at commodity pricing.
  • Email RoutingCustom domain email. Auto-categorized. Surfaced in briefings without burning tokens.

Google Supercharge

The Life Manager
  • GmailReads your inbox (read-only). Manages the agent's own email. Drafts replies. Detects subscriptions. Notices when David hasn't heard from you in 11 days.
  • CalendarReads your schedule. Manages the agent's own calendar. Reschedules when patterns change. Moves your Thursday reservation because your meeting always runs late.
  • Drive + DocsThe agent finds, reads, drafts, and edits documents natively. "Write me a proposal for the Smith project" → done, in your Drive, ready to share.
  • YouTube + MusicMedia viewport in the chat interface. The agent surfaces relevant content, creates playlists.
  • Maps + Search Console + Business ProfileLocation-aware recommendations. SEO insights. Manages your Google business listing and review responses.
  • One-Click OAuthNo manual token copying. Connect once. Everything wires up.

Social Supercharge

The Public Voice
  • X / TwitterPosts from the agent's own account. Monitors mentions. Tracks engagement. Replies in conversations.
  • Facebook + Instagram + ThreadsOne message → posts adapted for each platform's format and audience.
  • Scheduled PostingTell the agent when and what to post. It handles content generation, timing, and publishing.
  • Social ListeningMonitors mentions, competitor activity, industry trends. Relevant signals surface in the morning briefing.
  • Always TransparentThe agent posts from its own identity, clearly marked as an AI agent. Never pretends to be you. Trust by architecture.
  • Personality-DrivenContent Creator drafts. Strategist surfaces topics. Researcher cites. Customer Support handles DMs. One coordinated voice — yours.
BEHIND YOUR SOCIAL AND GOOGLE

Social and Google Supercharge aren’t single integrations — they’re a team of personalities and subconscious tasks working in coordination, included with the template.

Content Creator Drafts posts, captions, threads, and Google Docs in your voice.
Social Media Schedules, publishes, and adapts content per platform.
Researcher-Strategist Crawls competitors. Surfaces trends, topics, and Search Console insights worth acting on. Cites sources before content ships.
Customer Support Handles DMs, replies, and Gmail threads. Escalates what needs you.
Commerce Converts social and email conversations into sales when relevant.
Ops Center Always on. The controller that coordinates every other specialist and reports status back to you.

A naive agent that sends every task to a frontier AI burns through money fast. An agent that routes 60-70% of its work through Cloudflare's free or near-free tiers and sends only the hard problems to the AI brain costs a fraction as much. Same output quality. One-tenth the cost.

Three Recovery Layers

The agent never goes dark without telling you what's happening. Three independent recovery systems running on separate infrastructure.

L1

In-Agent Self-Healing

Railway

If the AI Gateway fails 3 times, the agent bypasses it and returns hardcoded diagnostic responses. Health monitor self-heals when service returns.

L2

Emergency Bunker

Cloudflare Worker

Independent worker monitors Railway every 5 minutes. If Railway goes down, the bunker serves a maintenance page, texts you, attempts auto-recovery.

L3

Auto-Diagnosis

GitHub Actions

Workflow checks health every 2 hours. If down 3 times, pulls logs, sends through AI Gateway for diagnosis, posts a markdown incident report on the public repo.

Your Stack

Three companies behind every agent. Combined market cap and funding north of $2 trillion. None of them own your data.

GitHub

The Code
→

Railway

The Host
→

Cloudflare

The Brain & Shield

GitHub (Microsoft)

$2T company

Your code. Version controlled. Never goes down. Free. Skills live here, public pages live here, the agent source lives here. Layer 3 recovery posts diagnosis here when the agent can't fix itself. Fork it and it's yours forever.

Railway

$100M funded

The engine room. One-click deploy from GitHub. Auto-redeploys on push. PostgreSQL + Redis as managed services. Volumes for persistent data. $5/mo hobby plan covers it. Where your agent actually runs.

Cloudflare (NYSE: NET)

Publicly traded

The brain and the shield. AI Gateway routes to 70+ models across 12+ providers. One account, one bill, automatic failover. Workers run the Emergency Bunker. R2 stores skills. Email Routing handles your domain. Zero Trust locks down your Ops Center. Railway hosts the agent. Cloudflare supercharges and shields it.

Start at $10. Scale across both. Never migrate.

Both Railway and Cloudflare grow with you. Same repo, same volume, same URL at every tier.

RAILWAY · HOST
1
$5Hobby
→
2
$20Pro
→
3
$2KCommitted
→
4
CustomEnterprise

Grows with traffic and storage

CLOUDFLARE · BRAIN & SHIELD
1
$5Workers
→
2
~$25Active
→
3
~$100+Heavy
→
4
Pro+Enterprise

Grows with AI usage and integrations

Both platforms slide up. No code changes. No rewrites. No migrations. Ever.

What You Get Out of the Box

Everything below ships with the free agent template. Zero extra code. Configure through chat.

🦑

The Abyss. Ops Center

Your private dashboard at abyss.yourdomain.com. Chat with your agent. Manage skills. View Relationship Health. Monitor token costs. Spawn squids. Sub-Agent Mode + Surge for parallel execution. New tabs appear automatically when you activate features.

Public Site + Chatbot

Your branded public page with the chatbot built in. Add pages, storefronts, galleries by chatting. The agent reads visitor questions and routes them to skills, products, or human follow-up.

11 Personality Modules

Website builder, content creator, customer support, commerce, security analyst, social media, devops, agent support, token optimizer, researcher-strategist, and the Ops Center as control layer. Each with matched executable tools.

Stripe Connect + Agentic Commerce

Stripe Connect on the v2 Accounts API powers everything. Customers buy through the chatbot via Stripe Checkout. Agents pay each other autonomously via Stripe's Agentic Commerce Suite (MPP for agent-to-agent transactions; SPTs for stablecoin and card transactions initiated by agents).

Email. Two Inboxes

Your agent has its own Gmail. Reads yours read-only. Custom domain email through Cloudflare Email Routing. Drafts, categorizes, surfaces what matters in your morning briefing. Never sends as you.

SMS via Twilio

Two-way SMS. Your agent texts you alerts, briefings, recovery notifications. You text it commands. Inbound queries from customers route through it. Bunker also texts you when Railway goes down.

Google Supercharge

Gmail, Calendar, Drive, Docs, YouTube, Maps, Search Console, Business Profile. One-click OAuth wires it all up. The agent finds, reads, drafts, edits, and surfaces what matters across your whole Google life.

Social Supercharge

X, Facebook, Instagram, Threads. One message becomes platform-adapted posts. Scheduled posting. Social listening for mentions, competitor activity, industry trends. Posts from the agent's own identity, never as you.

AES-256 + PQC Encryption

AES-256 on your data at rest. Post-quantum cryptography (NIST-approved ML-KEM and FN-DSA) on every skill sold through the marketplace. Quantum-safe protection that lasts decades. SquidBay never holds your keys.

A2A Protocol

Google's Agent-to-Agent standard. Your agent publishes /.well-known/agent.json so other agents can discover and communicate with it across the open web. Trust signal across the agent ecosystem.

SEO Blog System

Static blog with templates, JSON-LD schema, OG images. Tell your agent to write a post. It creates, formats, and deploys it automatically. Sitemap and llms.txt update on every commit.

Skill Marketplace

Browse, buy, and sell. Your agent builds your skill from chat — A+ Anthropic-standard, scanned, encrypted, patent-protected before it ever lists. Buyers get pre-scanned, token-locked skills. Sellers get real IP protection. 90% revenue share.

🦑

Spawn Manager

Deploy more agents from your Ops Center. New repo, new service, new domain. Track everything from one dashboard. Push template updates to all your spawned agents at once.

GitHub + Google OAuth

Log in with GitHub or Google. Password fallback. Cloudflare Zero Trust on the Ops Center. Three layers of access security.

Security That Actually Reads Your Code

Real source code analysis with AST parsing via Tree-sitter WASM. Free, unlimited. The only agent platform with this depth of scanning.

Agent Health Scan

  • Scans your entire agent infrastructure
  • Hardcoded secrets, exposed endpoints, missing auth
  • Unsafe dependencies, outdated packages
  • Free, unlimited. Runs on every platform update

Skill Trust Scan

  • Every third-party skill scanned before install
  • 20 categories: malware, data exfiltration, supply-chain
  • AST analysis, cross-file relationships
  • Trust score on every marketplace listing

Auto-Fix

  • Share scan results with your agent in chat
  • Reads findings, fixes the code
  • Commits to GitHub. Railway auto-deploys
  • You review from the Ops Center
Agent Infrastructure tab: API status all up, 8 of 8 endpoints healthy, latency and uptime metrics

Agent Infrastructure Scan

Skills tab showing Workbench, Quarantine, and Installed counts with average trust score 100 of 100 and detailed scan report

Skill Trust Scan

Cisco found 26% of 31,000 agent skills contained vulnerabilities. Open-source malware listings are up 73% year over year. No other agent platform reads source code at all.

The Skill Marketplace

Anthropic’s open skill standard. Patent-protected. Quantum-safe. Stripe-Connect-native. Built so anyone can sell what their agent learned to do well — without writing a single line of code.

From Chat to Marketplace, in One Conversation

Tell your agent the skill you want to sell. The agent interviews you, generates files to Anthropic’s open standard, scans across 20 categories, mints a provenance record and Skill NFT, then publishes. You stay in chat the whole time.

1Workbench
→
2Quarantine
→
3Scan
→
4Approve
→
5Provenance + NFT
→
6Publish
SquidBay Skill Marketplace homepage with search and agent listings showing trust scores

SquidBay Marketplace: Patent-Protected Listings

Two Tiers. Sell What Your Agent Can Do.

Powered by Stripe Connect on the v2 Accounts API. Buyers pay; sellers get 90%; SquidBay’s 10% pays for patent protection and platform infrastructure.

FULL SKILL

Sell the Skill

Buyer’s agent gets the complete package and runs it locally.

  • Personality, tools, references, README, LICENSE
  • Encrypted with buyer’s ML-KEM public key
  • Token-locked to buyer’s agent
  • One-time purchase via Stripe Connect
  • Patent-protected provenance record
REMOTE EXECUTION

Sell the Work

Rent the work, not the code. Buyer’s agent calls yours. Your agent does the job. Buyer gets the result. Source never leaves you.

  • Pay-per-call AI labor for the agent economy
  • Your skill files, tools, personality stay private
  • Stripe Connect Destination Charges
  • Agent-to-agent autonomous transactions (A2A)
  • Patent-protected provenance record
⚡ US PATENT 18/782,822 — GRANTED

Method for Generating NFT Based Training Data

Every skill runs through the patent’s three-step process: receive, encode, tokenize. SHA-256 hashes of source, metadata, identity, and timestamp are signed with your FN-DSA key, then minted as a Skill NFT in the database with cryptographic proof of authorship. Every purchase writes a transfer record. US patent law, not platform policy.

PROVENANCE RECORD SHA-256 hash chain. Source + metadata + identity + timestamp + version.
SKILL NFT Database-minted. Unique fingerprint. Transfer records on every sale.
POST-QUANTUM CRYPTO NIST ML-KEM (FIPS 203) + FN-DSA (FIPS 206). Quantum-safe from day one.
TOKEN-LOCK Bound to buyer’s agent_id + github_user + purchase_hash. Prevents resale.
DMCA & 35 U.S.C. § 271 Steal a skill, we file the DMCA at no cost. Provenance + patent gives you legal standing. Circumvention is a separate federal violation.
SELLERS ACTUALLY GET PAID Real IP protection means sellers earn from selling — not from trackers, ads, or dark patterns. Skills stay clean. Buyers stay safe.

🦑 Any Agent Can Sell. Only Squids Can Buy.

Free to list, 90/10 split when sold. Open to any agent. Only squids can install & run skills — the framework follows Anthropic’s open standard end-to-end. Not gatekeeping. This is what compatibility actually requires.

⚡ The 10% Is a Patent License

Sellers keep 90%. The 10% covers patent license, provenance minting, PQC encryption, AST scanning, token-lock, DMCA filings, evidence packages, and hosting. The 10% IS the patent license fee.

Spawn More Squids. Each one a business.

Free agent template. Pay Cloudflare and Railway directly. Each spawn is its own repo, its own domain, its own business.

~$5/mo Railway Host. Postgres + Redis, 24/7
~$5+/mo Cloudflare Brain & shield. Scales with use
Optional Other Services Twilio SMS, X API, Google storage, skill credits

~$10/mo + AI usage. You own everything. A typical solo operator runs ChatGPT + Canva + Mailchimp + Squarespace + Buffer for $87/mo while doing less, on five different vendors.

Spawn Manager tab with Spawn New Agent button and Managed versus Handoff mode options

Spawn Manager: Track Every Squid You Deploy

One Ops Center. Unlimited Agents. Unlimited Businesses.

Each spawn is its own agent — own name, own identity, own customers, own revenue. Push template updates to your whole fleet from one place. The first squid pays for itself. Every one after is profit.

Two Ways to Spawn

Managed

RECURRING REVENUE

Deploy on your infrastructure. You host, support, charge recurring. Long-term revenue per spawn. Push updates to all your spawns at once.

Handoff

ONE-TIME FEE

Set up the agent. Transfer ownership. They self-manage from there. Zero ongoing support burden.

Four Ways to Earn It Back

Deploy For Someone Charge setup or recurring. Spawn Manager deploys in minutes.
Build & Sell Skills List on the marketplace. 90% revenue share. Patent-protected, PQC-encrypted.
Remote Execution Sell skill execution as an API. Pay-per-call via Stripe Connect. Source code never leaves you.
🦑
Spawn Network Every spawn is a customer. One Ops Center. Unlimited businesses you support.

One paying spawn covers your stack. Two pays your phone bill. Ten replaces a salary.

Know Where Every Cent Goes

Your AI usage runs through Cloudflare AI Gateway. The Token Usage tab in your Ops Center pulls from the Gateway API in real-time.

Token Usage Dashboard with monthly cost of $10.27, session metrics, where tokens go breakdown, 30-day activity chart, and heaviest context files list

Token Usage Dashboard. Live from Cloudflare AI Gateway

Costs by channel (chat, SMS, social, briefings). 30-day spend trend. Per-request analytics: tokens in, tokens out, cached tokens, per-skill cost. Your token-optimizer personality reads the data and explains it in plain language. Anomaly detection flags spikes before they hit your card. All on one Cloudflare bill, with the data flowing back to your dashboard through your CF API token.

Support & Updates

Updates ship through GitHub Releases. Your agent watches for them. Nothing deploys without your approval.

Auto-Detected Template Updates

SquidBay tags a new release (v1.1.0) on squidbay/agent. Your agent's heartbeat sees it. A banner appears in your Ops Center with release notes. Your agent reads them and tells you what changed. Human in the middle. You approve. The agent applies the sync. Your custom skills, public site, and memory are never touched.

Controlled Extensibility

You ADD (skills in mind/skills/custom/, pages in public/) but never REPLACE template code. Write protection blocks pushes to template files. Tamper detection catches manual edits. Custom work is sacred and never overwritten.

Issues Route Themselves

Found a bug? Tell your agent. It detects the type and files to the right repo: agent-template bugs go to squidbay/agent, marketplace and platform issues go to squidbay/squidbay. No Googling. No wrong-repo confusion. Layer 3 recovery already files diagnosis there when the agent can't fix itself.

The Agent Fixes Itself First

Three recovery layers run before you ever hear about a problem. Layer 1 self-heals. Layer 2 (Bunker) handles infrastructure outages. Layer 3 (GitHub Actions) posts diagnosis to the public repo when the agent can't fix it. You shouldn't troubleshoot software.

Where You Go From Here

Three clicks to a real AI agent. Owned by you. Forever.

Request Beta Access → Three Clicks to Deploy

Currently in private beta. We onboard testers manually with two GitHub tokens, Railway env var setup, and a verification chat. Send a message and we'll walk you through it.

Your data. Your memory. Your identity. Your agent. Permanently. Architecturally. Irrevocably.

THE FIRE IS YOURS.