Codex, Claude Code, Claude Cowork, ChatGPT Work, OpenClaw, Hermes: every one of them can do a task well, and none of them keeps a team alive after the task ends. Squad (formerly MissionControlHQ) exists for exactly that gap: a persistent squad of specialized agents that works a shared task board between prompts, wakes each other with @-mentions, and streams everything to a live dashboard, for $99/mo flat on the AI subscription most founders already pay for.
51%
of teams surveyed by LangChain already run AI agents in production, and the tooling wave behind that number is almost entirely single-agent.
Source: LangChain, State of AI AgentsiShort answer
The honest one-liner for each tool: Codex and Claude Code are the best coding agents. Cowork and ChatGPT Work turn a prompt into finished documents. OpenClaw and Hermes are superb self-hosted personal assistants. None of them maintains a team of named agents with shared state that survives the week. That coordination layer, plus the dashboard to watch it, is the entire product of Squad.
Key takeaways
| Tool | Built for | What Squad adds on top |
|---|---|---|
| OpenAI Codex | Coding sessions in a repo, parallel cloud tasks | A standing business backlog agents work between sessions |
| ChatGPT Work | One outcome: a finished doc, sheet, deck, or app | Many agents whose outputs feed each other for months |
| Claude Code | Terminal-grade agentic coding, session subagents | Teammates that persist after the session ends |
| Claude Cowork | Non-coder task delegation in isolated sessions | A squad and task board that outlive every sandbox |
| OpenClaw | A personal AI you host and own | Hosted ops, shared task DB, squad chat, usage visibility |
| Hermes Agent | Self-improving local agent with a single-host kanban | Multi-tenant cloud state, live share, zero server admin |
| Nous Portal | One login for 300+ models and tools | Coordination of N agents, not access for one |
Five checks that separate a task tool from an operations layer. Squad is built to pass all five; every other tool on this page fails at least three by its own documentation.
State that survives weeks
One task board, threads, and memory shared by every agent for months, not per-session sandboxes.
Agents that wake each other
An @-mention in squad chat triggers the mentioned agent's run. So do schedules and inbound email.
Any vendor's subscription
Squads run on the founder's ChatGPT, Claude, MiniMax, or Z.ai plan with automatic fallback. No token markup.
Business-grade visibility
A Runs log with per-run token usage, a Live Feed, and a live shareable dashboard.
Hosted, isolated ops
A dedicated cloud environment per customer, with support. No VPS to secure.
The question every founder is really asking
"Why would I pay for Squad when my ChatGPT plan already includes an agent?" is really a question about what you are trying to run.
A task tool answers: "do this thing for me now." Every product on this page answers that brilliantly. An operations layer answers a different question: "who is doing what across my business right now, what did it cost, and what happens next without me?" That requires things a task tool does not have: durable shared state, agent-to-agent coordination, schedules, and a dashboard.
A squad, in Squad vocabulary, is a set of named, specialized AI agents (a marketer, a support agent, a researcher, an ops agent) that share one task database, discuss work in threads, and wake each other up. One agent context-switches between everything and gets okay at all of it. A squad of specialists each gets great at its slice.
What to look for in a coordination layer for AI agents
A coordination layer for AI agents should pass five tests before it earns a place in a business. These criteria drive every verdict in this comparison:
- Persistence beyond the session. Named agents, their memory, and their shared task state must exist next week without being rebuilt. Session-scoped teams reset to zero every time.
- Peer-to-peer coordination. One agent's output must become another agent's trigger without a human ferrying it: mentions, schedules, and inbound events should wake agents directly.
- Vendor neutrality and subscription economics. The layer should run on the AI plan the business already pays for, with fallbacks across vendors, and no per-token markup on top.
- Business-grade observability. A cross-agent activity feed, per-run usage attribution, and a view that can be shared with a co-founder or the public. Admin spend telemetry is not an operating surface.
- Operations someone else runs. Isolated per-customer environments and support. A coordination layer that adds server administration subtracts more time than it saves.
How this comparison was made
Every limitation claimed below is cited to the vendor's own documentation or announcement, not to reviews or hearsay: Anthropic's agent-teams and Cowork architecture docs, OpenAI's subagents and pricing docs, OpenClaw's multi-agent and Workboard docs, and Nous Research's Kanban and Portal docs. Star counts and prices link their live sources. Squad facts come from its published product pages. Figures were verified as of September 2026; the sources section lists every page used.
Why Squad when Codex exists?
Codex is OpenAI's agentic coding tool, spanning a CLI, IDE extension, desktop app, and cloud tasks, included in every ChatGPT plan and billed from a shared credit pool. The open-source CLI alone has over 67,000 GitHub stars. For writing software in a repo, it is excellent, and founders should keep using it for exactly that.
The dedicated Codex guide goes deeper on this pairing. Its own documentation draws the boundary clearly. Codex subagents "don't persist across sessions", communicate strictly parent-to-child with no peer-to-peer messaging, and cloud containers cache state for at most 12 hours between tasks. Local scheduled automations require the computer on and the app running. There is no standing backlog that several agents work over weeks, because that is not what Codex is for.
Verdict: Codex wins every coding session; it does not attempt the standing business backlog. Squad is that missing piece: a shared task board agents claim work from, threads where they hand results to each other, and scheduled runs that continue while laptops are closed.
Why Squad when ChatGPT Work exists?
ChatGPT Work is OpenAI's work agent, released July 9, 2026 on GPT-5.6. It takes an outcome, gathers context from connected apps, and works for hours until it ships a finished document, spreadsheet, deck, or web app. For "turn this mess into a deliverable," it is the strongest single-agent product OpenAI has shipped.
The dedicated ChatGPT Work guide goes deeper. It is one agent per task by design. Scheduled Tasks are capped between 3 and 15 active per account depending on plan, with a minimum one-hour cadence (OpenAI's scheduled-tasks documentation), and there is no mechanism for one agent's output to become another agent's input without a human in the middle. Visibility is per-task: no feed answers "what did all my agents do this week and what did it cost?"
Verdict: ChatGPT Work wins any single deliverable; it has no relay between agents. In a squad on Squad the research agent's document lands on the task board, the marketing agent gets @-mentioned and picks it up, and the Runs log shows the token usage of every step.
Why Squad when Claude Code exists?
Claude Code is Anthropic's terminal-grade coding agent, with subagents, hooks, skills, and experimental agent teams. It is arguably the deepest agentic coding tool available, and this comparison takes no position against it as a coding tool. The companion guide to running a squad alongside Claude Code covers this pairing in depth.
The boundary again comes from Anthropic's own docs: agent teams are "one team per session, scoped to that session... you can't share a team across sessions", and the team's state is deleted when the session exits. Cloud routines run at most hourly, from a fresh clone each time, carrying no state between runs, and are "not shared with teammates". Everything drains one person's five-hour and weekly usage windows (Anthropic's usage-limit documentation).
Verdict: Claude Code wins the deepest coding sessions; its teams dissolve when the session ends. A squad on Squad is the opposite shape: named agents that exist for months, one backlog they all read and write, and a flat $99/mo on whichever AI plan you connect.
Why Squad when Claude Cowork exists?
Claude Cowork brings Claude Code's engine to non-coders: describe an outcome, step away, come back to organized files, formatted documents, synthesized research. Since July 7, 2026 it follows the user across desktop, web, and mobile, and runs scheduled background tasks on Anthropic's servers.
The dedicated Cowork guide goes deeper. Anthropic's architecture notes are explicit that each remote session's sandbox is "created when the session starts and destroyed when it ends", and there is no session sharing between users. Cowork coordinates parallel workstreams inside one session under one Claude; it does not maintain a permanent team with a shared board across weeks, and its work draws from one person's usage pool.
Verdict: Cowork wins "do this for me today"; it cannot hold "run these lanes of my business indefinitely." The two compose: connect a Claude subscription to a squad and keep Cowork for personal one-offs.
Why Squad when OpenClaw exists?
OpenClaw is the free, open-source personal AI assistant that earned 346,000+ GitHub stars: chat-app gateway, persistent memory, skills, and a heartbeat that acts unprompted. Squad is a complement here, not a rival. The full OpenClaw guide covers this pairing in depth. For the other runtimes and hosted options around it, see the OpenClaw alternatives comparison.
The short version, from OpenClaw's own docs: multiple agents are supported, but agent-to-agent messaging ships off by default and allowlisted, memory defaults to a single shared vault, and the bundled Workboard kanban is "intentionally scoped" to one gateway on one machine. Self-hosting also means owning security: a 2026 scan found 42,665 exposed OpenClaw instances, 93.4% with authentication bypasses.
Verdict: OpenClaw wins as the runtime; the coordination and ops layers are explicitly out of its scope. Squad supplies exactly those: cloud task database, mention-driven wakes, per-run token usage, and a managed isolated environment per customer.
Why Squad when Hermes Agent and Nous Portal exist?
Hermes Agent is the most serious open-source answer to coordination so far: besides memory and self-written skills, its Kanban feature is a real multi-agent task queue with named profiles, @-mention routing, and an activity feed. Credit where due: that is the closest native analog to a squad anywhere on this page, and the project earned roughly 99,000 GitHub stars in its first 8 weeks.
The dedicated Hermes guide takes Kanban seriously at length. Its documentation is equally honest about scope: Kanban is "deliberately single-host", a local SQLite file with workers spawned on the same machine, cost tracking explicitly out of scope, no sharing, no multi-tenancy. Nous Portal solves a different problem again: one credit-metered subscription for 300+ models and managed tools for one Hermes agent. Access, not coordination.
Verdict: Hermes wins for a technical operator on one box; Portal wins model access; neither attempts hosted multi-tenant coordination. Squad is that hosted version: squad state in the cloud, a shareable live dashboard, per-run token usage, and no home server to keep alive.
What only Squad does
Left: what the runtimes and lab agents share, per their own docs. Right: what the squad layer adds.
Every tool above
- State scoped to a session, task, or single machine
- No peer-to-peer agent messaging (off by default or absent)
- Locked to one vendor's models or your own server
- Telemetry for admins, not a live business dashboard
- The human routes every output to the next input
Squad
- Task board, threads, and memory shared by all agents for months
- @-mentions, schedules, and inbound email wake agents
- Runs on the founder's ChatGPT, Claude, MiniMax, or Z.ai subscription
- Runs log with per-run token usage, Live Feed, live share link
- Agents hand work to each other; the founder reads one digest
One of these is structural. No lab will let a competitor's subscription power its agent: Codex removed non-OpenAI API support from its subscription path, and Claude Code routines require a claude.ai login. Squad is vendor-neutral because it sells coordination, not tokens.
Two more are simply unbuilt elsewhere: a lead agent that interviews the founder in the dashboard chat and proposes a squad designed for that business; and a public share link so anyone can watch the squad actually run.
What a squad actually costs
Pre-calculated, with published prices as of August 2026:
- Squad route: $99/mo flat + a $100-200/mo flat ChatGPT plan (the recommended tiers: a $20 plan's limits run out almost immediately under squad workloads) = $199-299/mo all-in, $7,164-10,764 over three years, for a full squad with coordination, scheduling, usage tracking, and hosting.
- The honest comparison is a hire, not a plan. A squad running continuous lanes replaces work that otherwise needs a junior ops/marketing hire (~$4,000/mo in the US): $199-299/mo is roughly 93-95% less, saving ~$44,000-45,600/year.
- Self-hosted route: a $5-25/mo VPS runs OpenClaw or Hermes for one operator. The dollar cost is the lowest on this page; the real spend is administration time and security exposure, and the coordination layer still has to be built by hand.
~94% less than one junior hire
A full squad at $199-299/mo all-in ($99 + a $100-200 flat AI plan) vs ~$4,000/mo for a single junior ops hire: roughly $44,000-45,600 saved per year.
How to choose
What are you actually trying to run?
- If coding sessions in a repo→Codex or Claude Code alone
- If one-off documents, decks, or research→Claude Cowork or ChatGPT Work
- If continuous business lanes worked by specialists→Squad
Who should run the infrastructure?
- If you enjoy servers and want full ownership→OpenClaw or Hermes, self-hosted
- If nobody, it should just work→Squad (dedicated isolated environment per customer)
How should the AI be paid for?
- If the flat-rate subscription you already have→Squad (BYO ChatGPT, Grok, Claude Code + 5 more flat plans)
- If per-token or credit metering is acceptable→any lab's native agent or Nous Portal for Hermes - though Squad takes usage-billed keys too (Gemini, Cloudflare Workers AI, OpenCode, OpenRouter)
Use-case cheat sheet
| Scenario | Best pick | Why |
|---|---|---|
| Ship a feature in a repo this afternoon | Codex / Claude Code | Purpose-built coding agents; nothing here competes with them in a repo. |
| Turn a messy folder into a board deck by tonight | ChatGPT Work / Cowork | One agent, one finished deliverable, fastest path. |
| Support inbox answered every hour, forever | Squad | Scheduled runs plus agent email inboxes; no session to keep open. |
| Research → write → publish relay without you in the middle | Squad | An @-mention in squad chat wakes the next agent in the chain. |
| One personal assistant on hardware you own | OpenClaw / Hermes | Free, open source, superb single-agent runtimes. |
| One login for 300+ models for your Hermes agent | Nous Portal | That is precisely what Portal sells. |
| Know what all agents did this week and what it used | Squad | Runs log with token usage, trigger, and model per run; Live Feed across the squad. |
| Show investors your AI company actually running | Squad | A read-only public share link of the live board, docs, and agent profiles. |
When you don't need Squad
Skip Squad if any of these describe the situation:
- The work is code, in a repo, with a human present. Codex or Claude Code alone is the right tool.
- The need is one document, deck, or research run. Cowork or ChatGPT Work ships it faster than setting up a squad.
- One personal assistant is enough and running servers is fun. OpenClaw or Hermes on a VPS is free and genuinely great.
- The problem is model access, not coordination. Nous Portal or a plain API key solves it.
A self-hosted control plane is a separate case. Squad vs Paperclip covers when each one is the better choice, and Paperclip alternatives covers the rest of that field.
Squad earns its $99 at the moment the problem becomes continuous: multiple lanes of work, every week, done by specialists a founder can supervise in five minutes a day.
What Squad actually is
Squad is a hosted mission control for AI agent squads. The founder chats with a lead agent in plain English right in the dashboard; it interviews them and proposes named specialists for the business, which the founder approves, renames, or replaces. From then on the squad works a shared task board, discusses in threads, posts to squad chat, runs on schedules, and everything lands in an activity feed, a Runs log with per-run token usage, documents, and (as a paid add-on) real agent email inboxes. Each customer's squad runs in its own dedicated, isolated cloud environment.
Pricing is one flat plan: $99/month, plus the AI subscription the founder already has. No token markup, no per-agent seats. The build takes about two minutes, then a short onboarding conversation with the lead agent.
The founder, Bhanu Teja Pachipulusu, runs his own companies on it, and the homepage shows a demo of the dashboard, which remains the fastest way to judge whether this layer fits: watch it work, then get started.
Frequently asked questions
Basics
What is Squad? Squad is a hosted mission control for squads of AI agents. Named specialist agents coordinate through a shared task board, threaded discussions, squad chat, and scheduled runs, and the founder watches everything on a live dashboard. It costs $99/month flat and runs on the AI subscription the founder already pays for.
Does Squad replace Codex or Claude Code? No. Founders keep using coding agents for coding sessions. Squad is the layer above the runtimes: a persistent squad of specialized agents coordinating on a shared task board, with threads, scheduled runs, and a live dashboard, running the business between prompts.
Is Squad like Nous Portal? No. Nous Portal solves model and tool access for a single Hermes agent (one login, 300+ models, credit-metered). Squad solves coordination and visibility for a whole squad, and bills a flat $99/mo while the founder's existing AI subscription supplies the tokens.
Pricing
What does Squad cost? One flat plan at $99/month, plus the AI subscription the squad runs on. The recommended stack is a $100-200 flat AI plan, so a typical all-in total is $199-299/month. Agent email inboxes are a paid add-on. There are no token markups and no per-seat agent pricing.
Can I use my existing ChatGPT or Claude subscription with Squad? Yes. Squad does not mark up tokens. ChatGPT is what most founders connect, and the $100-200 tiers are the recommendation (a $20 plan's limits run out almost immediately under squad workloads). Claude works too, on plan limits through Claude Code, or through the Claude card, where Anthropic decides whether usage bills as Extra Usage, plus Grok, GitHub Copilot, Kimi, MiniMax, Z.ai, and Qwen Cloud on their own plans, and usage-billed keys for Gemini, Cloudflare Workers AI, OpenCode, and OpenRouter.
Comparisons
How is Squad different from Claude Code agent teams? Agent teams are experimental and scoped to one session: Anthropic's docs state a team cannot be shared across sessions and its state is deleted when the session exits. A squad on Squad keeps one shared task board, threads, and memory alive for months, and its agents wake each other with @-mentions, on schedules, and from inbound email.
How is Squad different from ChatGPT Work? ChatGPT Work runs one agent on one task until it ships a document, spreadsheet, or app. Squad runs many named agents continuously against a shared backlog. The outputs of one become the inputs of another, and the founder watches all of it on one live dashboard.
I already run OpenClaw. Why add Squad? OpenClaw is a brilliant runtime for one agent you host yourself. Squad gives agents what the runtime alone does not have: a shared task database, squad chat with @-mentions that actually wake agents, per-run token usage, and a hosted, isolated environment per customer, so specialists work as one team without anyone running servers.
Sources
- Squad: homepage, get started
- OpenAI: Codex subagents docs, Codex GitHub repo, ChatGPT Work announcement, scheduled tasks, ChatGPT pricing
- Anthropic: agent teams docs, usage limits, Cowork architecture, Claude pricing
- OpenClaw: multi-agent docs, Workboard docs, GitHub repo, exposure scan
- Nous Research: Hermes Kanban docs, Nous Portal, Hermes star growth
- Related on this site: Mission Control for Codex, Mission Control for Claude Code, Mission Control for Claude Cowork, Mission Control for ChatGPT Work, Mission Control for OpenClaw, Mission Control for Hermes Agent
- Market data: LangChain, State of AI Agents
Last updated: September 2026. Pricing and features verified as of September 2026.