Skip to main content
A Computer Agent uses the same published Agent configuration model as other Vida Agents. Its model requests use agentModel; advanced heartbeat settings do not replace that primary model selection.

​
Durable Agent configuration

Use the normal staging and publish lifecycle for Agent-owned settings such as:
  • instructions and model selection
  • heartbeat instructions and supported heartbeat settings
  • knowledge links
  • assigned skills
  • Computer delegation
Read staging, update only intended writable fields, verify staging, test, publish, and verify live.

​
Computer Agent operations

Use /api/v2/computer/accounts/{targetAccountId}/... for runtime and operational resources, including:
  • provisioning, restart, upgrade, backup, restore, and deprovisioning
  • health, activity logs, service logs, diagnostics, and confirmed repairs
  • redacted runtime configuration inspection and bounded patches
  • channels, skill installation and setup, managed secrets, memory, sessions, and tool policy
  • workspace files, browser access, recordings, and reusable helper execution
Use the target Computer Agent’s stable account ID in the route. Do not use an agentConfigId as targetAccountId. Continue with the focused guides for:

​
Delegation from another Agent

A Computer Agent uses its own Computer automatically. Set computerDelegateAccountId only when one Agent must deliberately use a different Agent’s Computer. Discover functions again and require the Computer function to report allowed: true before publishing. Use Browser for ad hoc work in the live Browser; registered reusable helpers are exposed through Computer. A reusable helper is appropriate for one known operation with an exact input and result contract. Computer delegation is appropriate for bounded multi-step work across skills, browser interaction, files, memory, or account operations. Operational resources can require user participation for credentials, OAuth, device login, or destructive actions. Complete and verify those resources before depending on them in a live conversation.

​
Reusable helpers

Helpers are exact, reusable functions owned by a Computer Agent skill. They can automate a stable Browser workflow or run without Browser access for API, file, and data work. Create a helper when a conversational Agent needs a stable operation that is not already a built-in function. This is usually better than free-form delegation for a voice-call operation whose inputs, side effects, and result must be predictable. Use ordinary Computer delegation when the work is novel, exploratory, or requires the Computer Agent to plan several steps.
  • List the selected Agent’s contracts with GET /api/v2/computer/accounts/{targetAccountId}/helpers.
  • Execute an exact registered name with POST /api/v2/computer/accounts/{targetAccountId}/helpers/execute and an optional arguments object.
  • Check requiresBrowser to know whether execution needs the live Browser. Registered helpers are still exposed through Computer. Use skillSlug to identify the owning skill.
  • Configure every ID in requiredSecrets through the selected Agent’s managed-secret API. Do not send secret values in helper arguments; Vida supplies only the declared values during execution.
Reusable source belongs under skills/{skillSlug}/helpers/. Browser-backed source uses @browser_function(...); Browserless source uses @computer_function(...). Generated registry files are runtime state and should not be edited directly. Skill install, update, and uninstall refresh registration automatically. After a lifecycle change, list helpers again and run a safe representative call before depending on one in production. To make a helper available during voice or messaging conversations:
  1. Add the helper and its usage instructions to one owning skill on the provisioned Computer Agent.
  2. Configure every declared Agent managed secret and verify the helper through /helpers.
  3. If the helper belongs to another Agent’s Computer, set computerDelegateAccountId to that Computer Agent account. Otherwise leave it unset.
  4. Read GET /api/v2/agent/functions?targetAccountId=... and configure the exact computer action returned by the catalog.
  5. In the action instructions, define when the helper should run and map conversation values to its exact registered arguments. Helper input values must be supplied as helper arguments; descriptive task or success text does not fill them automatically.
  6. Verify staging, test the staged Agent with safe data, publish with explicit intent, and verify the live configuration.
Browser workflow generation accepts an optional skillName. Use a neutral reusable name when a recorded domain contains customer-specific information or would create a poor skill identity. Vida normalizes it and returns the actual skillSlug; the domain remains attached to Browser helper metadata and recording evidence.