Updated 27 Sep 2026 · nothing on this page is usable yet
Roadmap.
Today an agent gets an address (Homingbox), a way to sign in (Shakehand) and a way to be reached (aSIM), with Customs checking everything in and out. Next come the things an agent needs to act on its own while a person stays in charge. The order is honest; we don’t give dates.
after the launch work
Next
What we build after the current launch work: billing, the Python SDK on PyPI and framework integrations.
- Approvals hubControl
- The agent asks, a person approves: sends, sign-ups and other actions wait for a yes from the dashboard, by email or with one tap on a phone. Builds on: Drafts for approval, human on call and the emergency stop.
- Activity ledgerControl
- A signed, append-only record of what each agent did across Homingbox, Shakehand and aSIM, which you can export. Evidence for your own transparency duties, such as those in the EU AI Act. Builds on: The Agent-Disclosure header, the audit trail and EU hosting.
- Outgoing mail checks in CustomsControl
- Mail from new workspaces is scored for phishing before it leaves, and suspicious mail is held for review instead of sent. Builds on: The injection and phishing scores Customs already gives incoming mail.
planned
Later
Planned, in roughly this order, after the items above.
- Policies and budgetsControl
- Per-agent rules: who it may write to, how much it may send, and which apps it may sign in to. Builds on: Scoped API keys, allow and block lists and plan limits.
- Calendar invitesCommunication
- An agent reads, accepts, declines and sends meeting invites from its own inbox. A full calendar (CalDAV) may follow. Builds on: Invites already arrive by email, as .ics attachments.
- SchedulerInfrastructure
- Wake an agent at a set time or on a schedule (“follow up in 3 days”) without running your own job queue. Builds on: Scheduled drafts, which already send at a set time.
- Webhook relayInfrastructure
- Forward webhooks from any service to an agent on a laptop or behind a firewall, with no public URL. Builds on: The WebSocket event stream, which already works behind NAT.
- Telegram bridgeCommunication
- An agent gets its own Telegram bot, and messages arrive and are answered like email, with the same checks.
not committed
Exploring
We want these, but each brings cost, regulation or risk we haven’t solved yet. They may change shape, or not happen.
- Secrets for agent toolsAccess
- Encrypted storage for the API keys an agent’s own tools use, or a connection to the vault you already run. We don’t plan to store passwords or two-factor secrets for other companies’ accounts: that would make us a target and would help fake-account farms.
- WhatsApp bridgeCommunication
- The Telegram bridge, for WhatsApp Business, where many customers prefer to talk. Paid per conversation and approved by Meta for each business.
- Phone numbers and SMSCommunication
- A real number for an agent, for services that still send codes by SMS. Carrier registration, per-country rules and SMS fraud make this costly to run safely.
- File storageInfrastructure
- S3-compatible storage for the documents an agent collects and produces, next to its mail. Attachments are already stored; we will add this only if agents need more than that.
- Persistent browser sessionsAccess
- Hosted browsers that keep cookies between runs, for sites with no API. Heavy to run, and easy to abuse; it needs the approvals hub first.
- Agent wallets and cardsMoney
- Virtual cards with spending limits, so an agent can pay for what it needs with a person’s approval. Card issuing brings identity checks, compliance and fraud work that we won’t rush.
hello@agentboxd.com
Tell us what you need
This list is shaped by what people building agents ask for. If something here would unblock you, or something is missing, write to hello@agentboxd.com. What ships is announced on the changelog.