Smart Stack

One workspace for your AI toolkit: track the tools you use, compare alternatives side by side, and start from complete agent build sheets.

My stack

The AI tools you run, with costs, coverage, and an honest audit.

Compare

Side-by-side pricing, features and compliance for anything in the directory.

Builds

Complete builds for running real AI agents. Start one and tick the parts off as you add them to your stack. Every part links to its record in the directory.

Dots Desk

OpenAI's Dots are always-on agents with their own cloud computer and browser. Nothing here is assembled: you pick the plan, the channel it reports in, the apps it may touch and the first job it owns, then correct it until it runs the job the way you would.

  • Agent: OpenAI dots — Runs on GPT-6 Astra with its own cloud computer and browser. The brain and the hosting are not swappable.
  • Plan: ChatGPT Pro, Business Premium, Enterprise (beta)
  • Where it reports: ChatGPT, Slack, Microsoft Teams
  • Calendar: Google Calendar — Connected through ChatGPT plugins, like the rest of the 4,000-plus apps.
  • CRM: HubSpot — Give the Dot the system of record for the job it owns.
  • Workspace: Notion — Where its write-ups land.
  • First recurring job: Watch customer feedback, Chase invoices, Keep launch materials current, Weekly status report — One job to start. It improves as you correct it; add a second Dot for a second lane.
  • Approval policy: Ask before anything leaves the building, Ask before spending or deleting, Autonomous inside its lane

Muse Errand Runner

Muse is Meta's personal agent: it acts inside Gmail, Calendar, Outlook, Docs, Plaid, OpenTable and WhatsApp from its own secure VM. There are no parts to pick — only the plan, the channel, the first errand and how much it may do before asking.

  • Agent: Muse — Runs on Muse Spark inside Muse Secure VM. Brain and hosting are Meta's.
  • Plan: Free, $20 a month, $100 a month
  • Channel: Muse app, WhatsApp
  • Calendar: Google Calendar — A native connection, alongside Gmail, Outlook and Docs.
  • First errand: Book travel, Negotiate a bill, Fill forms and applications, Shop and reorder
  • Approval policy: Confirm every payment, Confirm before sending anything, Run trusted errands end to end

Grok Bot Team

xAI's Grok Bots share a cloud computer, sign into the tools you already use and finish jobs end to end. The way xAI runs them internally is a small team: one Bot that assigns work, and a specialist per lane. This build is that team.

  • Agent: Grok Bot — Runs on Grok 4.6 on its own cloud computer. It signs into apps like a person, so apps with no API still work.
  • Plan: SuperGrok family, Cursor Pro, Pro+ or Ultra, Cursor Teams
  • First specialist: Sales outbound, Pipeline ops, Account follow-up, Demo readiness — The chief-of-staff Bot hands this lane its work and checks the result.
  • Second specialist: Inbox management, Expenses and invoices, Recruiting, Bug fixes
  • CRM: HubSpot — The Bot updates it from call notes and drafts the follow-ups.
  • Help desk: Zendesk — A native connection for support lanes.
  • Editor: Cursor — Bundled with Cursor plans; an engineering Bot reproduces the bug, files the ticket and hands the fix to a debugging Bot.
  • Approval policy: Approve every outbound send, Approve changes to shared records, Trust a routine once it has been shown

OpenClaw Household

OpenClaw is the open-source gateway: fifty-plus chat channels, ClawHub skills and a policy-as-code trust boundary. This build keeps it domestic — one WhatsApp number the household can message, a calendar, live web answers, and an approval policy you set before you trust it.

  • Agent shell: OpenClaw — Open-source gateway with 50-plus chat channels, ClawHub skills and a policy-as-code trust boundary.
  • Brain: Claude Opus 5.5 — Any provider; you bring the key.
  • Channel: WhatsApp, Telegram, iMessage, Signal
  • Calendar: Google Calendar —
  • Web search: Exa —
  • Approval policy: Approve every command, Approve anything that reaches the host, Sandboxed and trusted — OpenClaw ships with sandboxing off. Turn it on before you trust it with the house.
  • Runs on: Hetzner — OpenClaw has install guides for Hetzner, Hostinger and Daytona; any Linux server works.

OpenClaw Starter

The minimum viable OpenClaw setup: one agent that lives on a small server, answers you in your chat app, and can actually look things up on the live web instead of guessing.

  • Agent shell: OpenClaw — Open-source personal agent that runs 24/7 and talks to you through your messaging apps.
  • Channel: Telegram, WhatsApp, Slack, Discord — Where you talk to the agent. Both shells speak 50-plus channels; pick the one your phone already has.
  • Brain: Claude Opus 5.5 — Strong long-horizon reasoning — the difference between an agent and an autocomplete.
  • Web search: Exa — A search API built for agents: clean, structured results a model can actually read.
  • Calendar: Google Calendar — Lets the agent see and manage your schedule.
  • Runs on: Hetzner — A small cloud server is enough to keep it running around the clock.

OpenClaw Power Build

The full-strength OpenClaw rig. The agent can run code safely in a sandbox, trigger multi-step workflows, and speak — useful when you want it managing things, not just chatting.

  • Agent shell: OpenClaw — Same shell as the starter — this build deepens what it can do.
  • Channel: Telegram, WhatsApp, Slack, Discord — Where you talk to the agent. Both shells speak 50-plus channels; pick the one your phone already has.
  • Brain: Claude Opus 5.5 — Handles multi-step plans without losing the thread.
  • Code sandbox: E2B — Isolated cloud sandboxes so the agent can run code without touching your machine.
  • Workflows: n8n — Self-hostable automation — the agent triggers flows across hundreds of apps.
  • Voice: ElevenLabs — Natural text-to-speech for spoken replies and briefings.
  • Runs on: IDCloudHost — Affordable Southeast-Asia hosting with full root access.

Hermes Desk with Hands

Every part here is yours to choose and to pay for separately: the shell, the model behind it, the channel you message it from, the search provider, the hands, the memory and the server. That is the trade a kit makes against a bundle — more to set up, nothing locked.

  • Agent shell: Hermes Agent — MIT-licensed personal-agent platform: CLI, messaging gateway, plugins, any OpenAI-compatible model.
  • Brain: Claude Sonnet 5 — Any provider; you bring the API key. Sonnet 5 is the price-to-quality middle.
  • Channel: Telegram, WhatsApp, Slack, Discord, Email (IMAP and SMTP) — Where you talk to the agent. Both shells speak 50-plus channels; pick the one your phone already has.
  • Web search: Exa — Hermes auto-detects the provider from the API key you set. Firecrawl is its default; Exa is a first-party plugin too.
  • Hands: Composio — Composio's MCP server gives the agent managed-auth access to 1,500-plus apps; free up to 100k tool calls a month.
  • Memory: Mem0 — Hermes has a memory store of its own; mem0 adds one shared across agents.
  • Runs on: Hetzner — A small VPS is enough to keep the gateway up around the clock.

Hermes Research Desk

Hermes handles long research runs: it searches the live web, pulls sources into a structured index, and writes findings into your workspace where you actually work.

  • Agent shell: Hermes Agent — Autonomous research agent built for long unattended runs.
  • Channel: Telegram, WhatsApp, Slack, Discord — Where you talk to the agent. Both shells speak 50-plus channels; pick the one your phone already has.
  • Brain: DeepSeek V4 — Open-weights frontier model — strong reasoning at a fraction of the token price.
  • Web search: Exa — Semantic search tuned for research queries, not ad-ranked links.
  • Source index: LlamaIndex — Turns collected documents into a queryable knowledge base.
  • Workspace: Notion — Findings land as structured pages your team can read.
  • Hands: Composio — One MCP endpoint for 1,500-plus apps, so findings can be filed into the tools you already run.

Claude Code Ship Crew

Claude Code does the heavy lifting in the terminal, Cursor gives you the same brain inside the editor, and a review layer catches what either of you missed before it merges.

Browser Operator

For work that only exists behind a browser: form-filling, bookings, account chores, scraping into spreadsheets. The agent operates pages directly, and structured scraping feeds it clean data.

Outbound Sales Agency

A pipeline where the agent finds prospects, writes the first line itself from real signals, and keeps the CRM honest — while a human stays on every send button that matters.

Support Desk Agent

A front-line support agent grounded in your own docs. It answers what it actually knows, hands off cleanly when it doesn't, and every conversation lands in your help desk.

Voice Front Desk

Callers get a natural voice that understands them, checks the calendar, and books or routes — around the clock. Built from specialist layers rather than one black box.

Content Studio

A production line for marketing content: long-form drafts grounded in your brand voice, visuals and video generated to match, and scheduling handled at the end.

Meeting Ops

Every call gets recorded, summarized, and turned into action items in your workspace; the calendar reshuffles itself around the work. Quietly one of the highest-ROI stacks.

Automation Backbone

Before agents can manage real work they need rails: triggers, queues, retries, and connections into the apps a business actually runs on. This is that layer.

Knowledge & RAG Base

A retrieval layer your whole agent fleet shares: documents in, grounded answers out, with model routing so each query runs on the cheapest brain that can handle it.

App Builder Line

The stack non-engineers ship real software with: describe the product, iterate on the generated app, drop to an AI editor when you outgrow the builder, and host it all in one place.

Short-Video Factory

Generate footage, cut it to platform-native shorts, voice it, and score it — a complete short-form video line driven from a single script document.

OpenClaw with Claude Code Hands

The hybrid shape: a self-hosted shell owns the channels, the memory and the trust boundary, and a vendor's coding agent does the engineering as one of its hands. OpenClaw runs Claude Code CLI and the Codex app-server as native runtimes.

Solo Creator Starter

Everything one person needs to write, design, and publish content — record a video, polish the audio, turn the transcript into posts.

Small Marketing Team

A lean marketing stack: long-form drafting, on-brand copy at scale, scroll-stopping visuals, scheduled socials, and automation glue between them.

Developer Starter

Ship faster with an agentic coding CLI, an AI-native editor, instant UI prototyping, and a research engine that cites its sources.

Customer Support Team

Answer customers around the clock: a helpdesk with AI agents, a trainable site chatbot, and meeting recaps so handovers never drop context.

Everyday Productivity

The all-purpose starter: a do-everything assistant, your own documents turned into a research buddy, meeting recaps, instant decks, and automations.

Not sure where to start? Run Smart Match and we rank the listings that fit what you are doing, or browse the full AI directory.