Build your business's missing tools on a working foundation.

Give supported AI jobs business context and keep their progress, results and decisions visible. Use the configured model, scheduler and registered tools. Work keeps its source context and recorded results; current policy determines which changes need a decision or can use bounded internal permission.

Adapt the platform to track the information your process needs. Use supported settings where available. New fields, record types or lifecycles may require a source extension that reuses the existing identity and permission services.

Use supported jobs and workflow triggers to keep repeated work visible. Configure the required capabilities and schedules, then inspect job and action results. A new business rule may require a source extension rather than a setting.

Review AI proposals and workflow plans in one place, with their source context. Approval records the decision; execution records the result. Autonomous proposals are checked before they reach the queue, and each one explains why it earned your attention. Anything you asked for directly is never held back, and a background check that finds nothing finishes quietly instead of reporting it. Eligible autonomy changes need a separate human decision, and restricted external actions keep their approval requirements.

Propose guidance your team wants future work to follow. Learning Inbox keeps the proposed rule, scope and source available for review. Approved independent rules stay active together, and edited replies can propose reusable corrections. Cited references and later context receipts make their use inspectable.

Keep the reason for declining a proposal available with the work. A recorded rejection helps the team understand the decision. Propose reusable guidance through Learning Inbox when the correction should apply beyond that one action.

Compare recorded replies, stage changes and provider results with the work you requested. Activity and action receipts retain evidence of execution. Use those records to assess the result; a completed action alone does not establish a business outcome.

Every kind of action carries its own trust level, and standing permission takes a human decision to grant. Changing a level, a constraint, or the source clears the previous approval, so a shift in policy always comes back to a person. Revoking a permission keeps its audit history.

Start with decisions, follow-up and business changes drawn from available records. Today provides source-linked attention and supporting context. Open a finding to inspect its record before acting.

Today connects sourced business findings to the next action; the same tasks and approvals stay editable in Work. Nine business groups explain where to review the operation, manage customers, deliver client work, follow billing, communicate, use AI, connect Apps and adjust settings. Personal and shared Today arrangements keep view tools under More, wide desktop layouts keep independent columns, and Search and Ask AI stay available while you scroll. The mobile dock keeps the three daily destinations within reach and waits for notifications to finish closing before returning. Dismissing notifications returns focus to the bell; following a notification keeps focus with the destination. Notification panels skip animation under reduced motion. More keeps the page paused through closing, then opens Search or Ask AI with keyboard focus. Following a menu link focuses the destination heading.

Ask for the result you need, review exact changes in chat and follow the saved work. Ask AI reads records, prepares replies and tasks, and previews ordered jobs. Configured background work retains progress after the browser closes. You can approve exact proposals inside chat and grant bounded permission for named internal operations. Messages, publishing, billing and permission changes keep human review. Available tools depend on modules, providers and current access.

Use a connected agent to read records, prepare changes and manage saved work. Compatible clients use the same registered tools, record services and approval rules. An authenticated member connection can delegate ordered work and read or control its progress. Tenant bearer keys retain integration scope. Pending proposals link to the exact human review, and unsupported operations are reported plainly. A separate owner-only Site Studio connection handles website editing.

Today, Intake review, Pipeline, Records, and Setup all work from a phone browser, including the pipeline board. The board scrolls freely, keeps its column controls at every screen size, and supports dragging, touch, and a stage control for moving a card without dragging. Deployments with a dedicated app address also support installation, connection status and local drafts that you can read and reuse after reconnecting.

Build integrations against the platform’s supported operations. Use the documented tool, adapter and service contracts. A custom integration needs implementation and permission checks for the particular records and actions it exposes.

Extend the platform with custom records, lifecycles, workflows, integrations, AI tools, and screens that reuse your existing customer context. Configuration covers the supported settings today, and broader changes use source development. A general-purpose in-app App builder is planned.

Choose one of nine appearances or preview a custom palette, typography, and corner style, then save it for the workspace. Text contrast is checked before a theme saves, density is adjustable independently, and a theme follows the same permissions and revision checks as other branding. Delete, discard and archive confirmations use the theme's danger appearance. Nested confirmations dim the editor beneath them and reveal it again when closed. Phone navigation returns after the last dialog finishes closing. Dialogs open and close immediately under reduced motion. Demo choices stay separate from the live workspace.

Enable the optional parts of the workspace your business needs. Core capabilities remain available. Disabling an optional module removes its navigation and gates its routes and tools, while preserving existing records.

A plain-language request is enough for a coding agent to pick up a task, prepare an isolated workspace, and carry it through verification and commit. Coding-agent instructions reserve pickup for explicit backlog requests and keep ordinary continuation requests on the current task. The agent does not need a ticket key or a provider-specific command, and it never asks you to paste credentials. If pickup cannot use the approved source, it names the blocked card and the recovery step before claiming. With recovery enabled, a replacement agent can resume an interrupted task from its saved state. Verification renews the active claim, and checkpoints include safe unfinished source while excluding credentials and generated output.

Inspect who initiated a change and the result recorded for it. Activity and receipts link actions to source records. Sensitive content stays in authorized records rather than being copied into every audit entry.

Turn off a supported module or capability when your team needs to stop its work. Use the relevant configuration control and inspect any pending work. Historical records remain available according to the feature’s retention and access rules.

Keep business records separated by workspace access, or run a self-hosted installation on infrastructure you control. Managed workspaces use tenant-scoped records, active membership, and row-level policy. A self-hosted installation uses your own database and provider accounts; agree on export and handoff responsibilities before a managed implementation.

Keep business data behind active membership, roles and record permissions. The runtime rechecks access for supported operations. An installation can offer Google sign-in alongside email, while active membership still controls workspace access. Review your workspace’s configured permissions instead of assuming a job title automatically grants or hides particular screens.

Find missing setup, failed jobs and results that need recovery. Setup and operational health views show supported checks with their evidence. Follow the named recovery step and verify the specific connection or job afterward.

Have a coding agent set up and extend the MIT-licensed workspace using accounts you control. Give Claude Code or Codex the business workflow and review its working result. The agent follows setup and extension contracts, reusing customer identity, permissions and business services so each custom screen needs less shared infrastructure. Review its exact source and checks before release, and keep account ownership and recovery responsibilities clear. Setup shows the recorded core version separately from your fork commit and checks the supported stable upgrade path.