FOR PEOPLE WHO ALREADY RUN AGENTS
Run the team.
Keep the autonomy.
Put native agents in persistent environments with a shared operating layer: leads that delegate, files they can work on together, and one place for you to direct, inspect, and recover the whole team.
Real native runtimes. Published capacity limits. Usage you can inspect.
The native runtime owns execution. The platform connects the team and manages its working environment.
WHERE THE PLATFORM EARNS ITS PLACE
For builders who want to spend more time directing useful work as their agents, connected accounts, and responsibilities grow.
Assign whole responsibilities to agents. A lead can delegate, review shared activity, update instructions, and resolve handoffs while every specialist keeps its own native execution and identity.
People and agents work on the same documents and data. Control which agents can see or edit a file, recover an earlier version, and keep milestones without merging everyone’s private memory.
A realtime voice manager can consult up to eight native agents and synthesize their replies. Move from a team conversation to Chat Wall or the actual persistent browser when you need to intervene.
Encrypted native-state checkpoints, saved setup versions, supported runtime rollback, and monitored recovery make a working agent easier to maintain after the initial install.
The subscription includes weekly usage for managed models, voice, and proxies. Own API keys are a manual request, not an instant switch. Connect Cursor as an executor with the native harness supervising, using the same browser sessions and network route.
Keep signed-in browser profiles, encrypted secrets, and gateway connections with the agent. Add a reserved residential IP or rotating route when the work calls for it.
Your AI can help run your AI team
Connect Codex, Cursor, Claude Code or another compatible AI client to your Agent Teams account. Then ask it to investigate: read the evidence, inspect your agent’s real files, and make backed-up repairs within the access you choose.
Create an expiring access key, copy your client’s setup, and choose one agent or your account.
“Find why my research agent stopped producing reports.” Your AI gets direct diagnostic tools, even when the hosted agent cannot answer.
Start with inspection or allow repairs. Review consequential actions, inspect activity, restore backups, and revoke access.
Prefer us to handle it? Paid AI Expert Repair uses the same native repair tools. Native changes require a stopped agent; maintenance commands run offline. Your AI can request supported software updates with your approval and, when enabled, report platform faults directly to our team with a trackable response. External sign-in and shared platform issues may still need human help.
Connections are included with paid Agent Teams plans. Your AI provider and normal platform usage have their own charges. Expert Repair is priced separately before you submit.
NATIVE EXECUTION. MANAGED OPERATION.
Hermes and OpenClaw own the reasoning, tools, memory, schedules, and subagents. AI Agent Store adds the shared operating layer around them.
Native planning and execution, skills, tools, APIs, MCP, browser work, terminal access, subagents, and recurring work—with the permissions and context you provide.
Hosted environments, team coordination, shared files, Chat Wall, human membership, managed connections, durable state, and recovery. Capacity comes from your chosen plan.
Models, voice, and proxies share a weekly allowance, shown as a percentage. When the weekly allowance runs out, wait for the reset or turn on optional extra usage. The computer stays while the subscription is paid.
You can buy extra usage when the weekly allowance runs out. It is paid in money and does not reset every week. Nothing is charged for it unless you turn it on.
Want to use your own API keys or supported AI subscriptions? Request access in Settings; approval is manual. Approval does not change the payment plan.
OpenAI ChatGPT / Codex · MiniMax Coding Plan · Qwen Subscription (Alibaba Token Plan) · Z.AI Coding Plan · Kimi Code membership · Cursor subscription · xAI / Grok subscription
Once linked, Cursor is configured as the default executor unless you direct otherwise. Hermes or OpenClaw remains the supervisor and fallback. All use the same persistent browser profile and selected proxy route.
WHAT YOU PAY FOR
Plans are $130, $260 and $650 per month. Starter and Plus include 1 computer. Scale includes 3 computers and 5× the weekly usage of Starter. When the weekly allowance runs out, wait for the reset or turn on optional extra usage. The computer stays while the subscription is paid. There is no free trial.
$130/month
1 computer · 1× weekly usage
$260/month
1 computer · 2× weekly usage
$650/month
3 computers · 5× weekly usage
$650/month
3 computers · 5× weekly usage
Shared workspace membership uses the owner’s plan. Colleagues keep separate logins and personal layouts.
The shared-file library has a separate 5 GiB allowance; its versions and trash count toward that allowance.
Saved agents and native subagents are different: the plan’s running limit applies to separately hosted agents.
THE REST OF THE BILL
PLAN YOUR TEAM
When the weekly allowance runs out, wait for the reset or turn on optional extra usage. The computer stays while the subscription is paid. There is no free trial. Paid passes are $26 for 1 day, $32.50 for 3 days, and $65 for 7 days. A paid trial becomes the monthly plan unless you cancel. A rental does not renew.
Want to use your own API keys or supported AI subscriptions? Request access in Settings; approval is manual. Approval does not change the payment plan.
Starter includes 1 computer and 1× weekly usage.
The estimate is the subscription price. Included usage is a percentage, not a dollar balance. Taxes are extra.
Configure StarterPrices are loaded from the public product catalog. USD unless shown otherwise. Checkout confirms current pricing and taxes. Hosting availability is checked before new subscription checkout.
THE COMPLETE CAPABILITY REFERENCE
85 capabilities across 9 areas. The important limits are in the descriptions, alongside what each feature does.
Showing 85 of 85 capabilities
Give agents operating responsibilities and a way to coordinate them.
A company file library that humans and agents can actually work in.
Operate a fleet without giving up direct access to any agent.
Keep the runtime’s execution model and choose how each agent starts.
The subscription includes weekly usage. Own keys are a separate request.
Keep working state durable as the runtime changes around it.
Connect the accounts and identity each agent needs to work.
One workspace for the business, individual logins for its people.
Separate saved capacity, running time, and optional consumption.
START WITH A REAL RESPONSIBILITY
Bring files, choose a Starter Kit, or describe the responsibility. Pick the native runtime and model access.
Configure an agentAdd the accounts, browser sessions, business documents, and instructions it needs to act independently.
Explore shared filesAdd specialists, define shared responsibilities, invite colleagues, and review results through Chat Wall.
Explore teamsTECHNICAL BUYER QUESTIONS
The paid value is in operating the environment and the team together: shared tasks and context, a versioned company file library, Chat Wall and voice access, individual human logins, managed accounts and networking, and recovery of persistent agent state. The native runtime remains the source of autonomy. If your current setup already meets your needs, compare these operating capabilities against what you would actually use.
The native runtime still decides how to execute work using its own tools, memory, schedules, and subagents. Team coordination supplies roles, shared instructions, messages, and tasks. Agents read that context during native work and scheduled runs; you configure the responsibilities, access, and recurring coordination they need.
You can use your own supported setup files, start from a prompt, or adapt a Starter Kit. A clean clone copies owner-authored files saved in AI Agent Store into a fresh stopped agent. It does not duplicate live runtime memory, sessions, schedules, credentials, or gateways. Full native-state checkpoints are a separate recovery mechanism for hosted agents.
Workspace members operate the same agents, conversations, files, tools, and native environments through their own accounts. The workspace owner controls membership and billing. Each person keeps their own layout. Agent visibility and editing permissions in the shared-file library let you decide which agents can use which business documents.
The subscription includes the computers on your plan and weekly usage for managed models, voice, proxies and repair. When that allowance runs out, wait for the weekly reset or turn on optional extra usage. Want to use your own API keys or supported AI subscriptions? Request access in Settings; approval is manual. Approval does not change the payment plan. Some features depend on the selected runtime, model, connected account, country availability, or supported client.
Hosting is isolated by customer workspace, with agents able to share that customer’s server. Each agent has a separate runtime and working state. Operators manage supported placement and maintenance; this is not customer-controlled automatic scaling, a dedicated physical machine per agent, or a published enterprise SLA. Choose capacity by the plan limits and validate it against your actual workload.
No. A saved agent retains its state within plan limits, but execution and scheduled work require a running runtime. Current plans include full-month hosting within their resource allocation; running longer does not add an hourly hosting charge. Static residential reservations can also remain billable while an agent is stopped, until released.
FROM A WORKING SETUP TO A WORKING TEAM
ALSO AT AI AGENT STORE
Find third-party agents, frameworks, agencies, and opportunities across the wider ecosystem.
Explore the directory and compare what the market offers.
Browse the AI agent marketplaceList your agent, publish an article, or sponsor the newsletter.
