A clear path forward.

Your website account manages your organization subscription, AI packages and environment setup. Your workspace is where you configure agents, knowledge, tools and deployments.

Production and sandbox keep separate application data. The owning account can open a ready workspace from Account without copying its password into that workspace.

When access is available, verify your email and confirm your subscription and hosting arrangements with our team. Open your workspace after its setup is complete. New registrations are currently closed.

Registration alone does not create a running workspace. Provisioning and activation are separate steps, with the deployment and access details confirmed before use.

The account-to-workspace handoff is short-lived, single-use and bound to the browser that started it. Return to Account and open a fresh link in the same browser.

Also check email verification, subscription status and workspace readiness. If access cannot be verified, the workspace stays closed. Never share session cookies, access tokens or password-reset links.

The website handoff currently supports the owning account. Other workspace members use individually issued access tokens managed by the workspace owner.

Copying a team profile to sandbox does not copy a working token or grant website account access. General account invitations and shared team sign-in are still planned.

Cancellation at the end of a paid period keeps access until that period ends. Inactive or overdue subscription states suspend workspace access without deleting the database. Reactivation reuses the existing workspace.

Downgrading preserves existing records but can block new usage when you are above a resource limit. Work already underway may finish; suspension cannot recall a completed action.

No. Each environment keeps its own agents, knowledge, model connections, credentials and conversations. The account owner can selectively copy supported settings through Sync & promotion.

This is an explicit, reviewed transfer. Continuous synchronization is not enabled, and deleting a source record does not silently delete its destination copy.

Choose sandbox as the source in Sync & promotion, select the agent and explicitly map the destination model and knowledge libraries. Review the changes before approving.

The transfer creates or updates a draft. It does not publish the agent or change a live deployment. Test the destination draft and use the usual evaluation and deployment controls.

API keys, OAuth tokens, server URLs, conversations, knowledge contents, live releases and deployment bindings are not copied. Imported provider settings stay disabled until configured and verified at the destination.

Newly copied members need a fresh destination token. New agent imports have no tool grants; existing destination tool grants stay local.

You can cancel before approval. After approval, changes may already have committed at the destination, so the worker must finish or verify the outcome rather than report a potentially false cancellation.

If the outcome is being checked, avoid starting a duplicate transfer. The worker retries automatically. An expired approval cannot apply new changes, but an existing committed result can still be acknowledged.

Yes. Your platform subscription covers the features and resource allowances in your plan. Managed AI usage uses an optional recurring prepaid AI package and/or one-time top-ups.

With your own provider key, that provider bills you directly. Self-hosted models have hosting and capacity costs; they do not provide unlimited compute.

Yes. Enrolled managed inference in both environments draws from the parent organization's shared wallet. Usage retains environment attribution, and environments can have separate budgets.

Direct BYOK and local Ollama calls are outside the managed credit wallet and follow their provider or hosting arrangement. Switching environments does not create free credits.

No. The wizard uses a small reviewed reference catalog and your expected request and token volume to compare suitable connections. Unknown rates are not guessed, and stale estimates are withheld.

Actual managed usage and account rates determine managed billing. Confirm the selected provider's current availability and costs before relying on an estimate.

Check model enrollment, whether the connection is enabled, subscription status and environment budgets. Resource limits can also stop a request. Pending reservations reduce the credits available for other work.

Use account usage and request records to identify the cause before resubmitting an uncertain request.

File storage holds originals. Database storage holds imported content, configuration, conversations, tickets, audit records and other application data.

Your own S3-compatible bucket can move original-file costs to your storage provider. Imported content and application metadata still count toward database allowance.

Selected-file knowledge connectors are implemented for Google Drive and Microsoft OneDrive. You choose what to import, and imported content is drafted before publication. Connecting an account does not grant an agent general write access to that drive.

Provider credentials, consent and permissions must be configured and tested for the deployment. These connectors import knowledge; they are different from the bucket used to store platform file originals.

No. Website knowledge sources can scan a configured site and import its pages. Configure the source in Knowledge, review the imported content and publish the sources the agent should use.

Check crawl status and relevant source pages if an answer is missing or outdated. An agent's answer quality depends on the published content and its instructions, not simply the number of pages imported.

Ordinary grounded answers use ingested, indexed content rather than downloading the original for each question. Original-file tools may fetch remote content; bounded caching and direct downloads reduce repeated transfers.

Storage placement, provider latency and file size still matter. Test the actual connection from the environment where the agent runs.

Open Upload / activity to inspect status, resume, retry or cancel. Resume requires the same original file. Integrity checks prevent partial or corrupt uploads from becoming ready files.

Reserved storage may remain charged until abandoned objects or parts are successfully removed. Repeated uploads are not a workaround for failed cleanup.

Deployment tooling supports scheduled encrypted database backups and verification by restoring into disposable databases. Confirm your backup schedule, destination and restore arrangements with our team.

A database backup alone does not restore bucket contents, queue state or deployment configuration. Hosted recovery still requires independent storage, key protection and tested restore procedures. Local backups on the same machine are not off-site disaster recovery.

No. The platform also supports internal knowledge, sales engagement, documents and application workflows. Start with one useful task, then expand the agent's role after testing.

The setup wizard helps you choose a purpose, connected model, knowledge and behavior before creating a draft.

No. Capabilities depend on explicit grants, access policies and deployment configuration. Connecting a tool does not give every agent permission to use it. Actions that require approval keep that approval step.

Configure resource paths and credentials for the environment where the agent will run. A platform file path is not automatically a path on a customer's computer.

Include the workspace name, whether it is production or sandbox, the approximate time, the feature involved and any visible request or transfer ID. Describe what you expected and what happened.

Remove secrets and customer data from screenshots or error text. Never send passwords, API keys, session cookies, access links or full database connection strings.