Autonomous digital humanoid workers
Workers with a role, context, and tools to carry out an assignment. Managed cloud desktops are the default execution environment.
Build a LinkedIn agent. Put a worker on business operations. Give your product workers that research, use software, and return outcomes you can verify, even without APIs.
workerworkresourcesguardrailsoutcome
For enterprise teams, BPOs, growing companies, and developers. Run internal or outsourced operations, deliver work for clients, or build products and companies on the same worker infrastructure.
Build a LinkedIn agent around the work your customer needs: understand an account, research relevant people, and prepare a thoughtful first conversation. Use authorized access and keep outreach ready for review.
Example brief: Research 25 approved accounts, identify relevant people, and prepare a source-linked brief and outreach draft for each.
What comes back: prospect briefs with role, company context, source links, missing information, and outreach drafts
const run = await client.workers.run({ worker: { role: "Professional Research Agent", mode: "single", availability: "on_demand", }, work: "Research the approved LinkedIn account list, capture source links, and draft relevant outreach without sending messages", resources: { credentials: "vault_customer_linkedin", budget: 40, }, guardrails: { approvals: ["messages", "external_submit"], }, outcome: "Return prospect briefs with role, company context, source links, missing information, and outreach drafts",});Illustrative use cases and API assignments, not customer case studies or prebuilt integrations. Available through early access; each workflow depends on the access and tools you configure.
Every run has the same shape, whether it is a single ticket tonight or a team running a function for a month.
workerThe digital humanoid worker you spin up. A role, a mode, and an availability. One worker, a persistent worker, or a team.
workThe last mile in plain English: the job you need done in a portal, inbox, or system, stated like an ops brief.
resourcesWhat the worker is allowed to use: credentials, cards, files, and tools. Set a purchase budget; execution charges are separate.
guardrailsThe moments that need a human. Messages, external submissions, payments, and trades wait for approval.
outcomeDefine the result you need, then inspect the returned data, cost, available evidence, and verification status.
const run = await client.workers.run({ worker: { role: "Operations Team", mode: "team", availability: "24_7", }, work: "Clear the overnight operations queue across billing and vendor portals", resources: { credentials: "vault_operations", budget: 300, }, guardrails: { approvals: ["messages", "external_submit"], }, outcome: "Return completed cases, updated records, evidence, and exceptions",});The execution layer underneath: runtime, roles, device sessions, resources, guardrails, traces, and outcome delivery.
Workers with a role, context, and tools to carry out an assignment. Managed cloud desktops are the default execution environment.
Attach authorized credentials, files, tools, and spending limits. Workers interpret interfaces as they work and surface blockers when they cannot continue.
Run a single job, keep an always-on worker, or coordinate a team that owns an entire function.
Inspect structured results, cost, available screenshots, and verification status. Review missing evidence and exceptions before relying on an outcome.
Like a person, a worker can run a job alone, pair with your team on a shared screen, or train someone step by step, and it decides which the work calls for.
Use supported apps and browser workflows to operate portals and authenticated services, even without APIs. Local browser access requires a separately configured connector.
A worker that uses authorized apps and accounts in supported execution environments to carry the assignment through to an outcome.
mode: "single"One workerOne assignment, one outcome, then done. The unit of work is a job, not a session.
mode: "persistent"Persistent workerA worker identity that keeps context and accepts new assignments as they arrive. Availability sets when work can run.
mode: "team"Worker teamSeveral roles coordinating on one function, with handoffs and escalation between them.
availabilityon_demandSpin up for one jobpart_timeScheduled windowsfull_timeBusiness hours for a function24_7Always-on operationsExample: prepare a food order within a customer’s preferences and budget, request approval, and return the receipt after checkout.
const run = await client.workers.run({ worker: { role: "Commerce Operator", mode: "persistent", availability: "24_7", }, work: "Prepare food order #9921 and request approval before checkout", resources: { credentials: "vault_food_ordering", card: "card_vault_xxx", budget: 75, }, guardrails: { approvals: ["payments", "external_submit"], }, outcome: "Return receipt, final cost, and delivery ETA",});Start with one workflow. Scale across departments, client accounts, products, or an entire business. One worker infrastructure, built around the outcome you need.