Any device.
Any provider.
Anywhere.

Your agents can see the work around them.

Agents in the same project discover who is working on what, coordinate file access, and ask each other focused questions. Give separate chats shared awareness without relaying every update yourself.

• Discover peer work and current progress
• Coordinate edits with advisory file reservations
• Exchange focused questions and replies

Works within a project, including its Git worktrees. Claude, Codex, and Kimi support full participation; other harnesses may show presence only.

From your request to a result you can review.

With agent task management enabled, substantial requests become tracked outcomes. Agents maintain the steps, update progress, and record what they checked. Completed work stays available for your review and acceptance.

• Track outcomes across chats and devices
• See progress and verification evidence
• Keep human review separate from agent completion

Project tasks are shared across the project's worktrees and managed by your host.

Let the reset time handle the restart.

When a supported provider reports its usage reset time, your chat is set to continue one minute after it resets. No alarm to set. No return trip just to type continue.

• Enabled by default for supported limit notices
• Turn automatic resume off in the chat
• Keep the pending continuation across reconnects

The host must be running. A known reset time is required; pending approvals and active work take priority.

Move the work forward without starting over.

Choose a new provider and bring the conversation with you. Agents Dispatch transfers the full persisted conversation transcript by default, along with a source-chat reference and the available work checkpoint, so the next agent can pick up the task.

• Switch providers from the current chat
• Carry the conversation, decisions, and work progress
• Continue in the same project with your chosen model

Transfer and model context limits still apply. If the transcript exceeds the server limit, you can choose to transfer the most recent 100,000 characters. Provider tools and internal model state are not identical.

Line up the work while your agent is busy.

Queue a follow-up without interrupting the current turn, or schedule a message for later. Use recurring scheduled tasks for work you want agents to revisit regularly.

• Queue follow-ups from the composer
• Schedule a one-time run
• Set recurring work for your agents

Keep your host awake and connected for scheduled work to run.

Spend your effort on the idea, not the phrasing.

Use AI to sharpen a draft before it reaches your coding agent. Prompt improvement draws on the conversation and project context to clarify your intent while preserving the goal and constraints you already gave.

• Turn a rough draft into a focused request
• Bring relevant project context into the prompt
• See the original and improved prompt in the conversation

The rewrite is instructed to preserve your request and avoid inventing requirements.

Keep the decisions. Know where to pick up.

Ask the current agent for a standalone summary of the conversation. Capture the goal, constraints, decisions, completed work, open questions, and an explicit continuation checkpoint in one readable handoff.

• Review what happened without rereading every turn
• Preserve plan items and their current status
• Give the next agent a concrete starting point

Summaries are generated from the conversation. They do not independently verify the workspace or perform more work.

Switch projects without losing your place.

Keep active conversations across projects in a tab bar, an Active Chats sidebar, or both. See the project and provider at a glance, then jump straight into the conversation that needs your attention.

• Keep active chats visible across projects
• Choose tabs, the sidebar, or both
• Recognize provider and session status at a glance

Cross-project navigation keeps your conversations close; agent communication remains scoped to each project.