The window where most of your work happens
The primary way you drive Ship is from your OWN agent — Claude Code or Claude Desktop — connected to Ship over MCP (run claude mcp add ship <url> -t http, log in through the browser, and approve your workspace; no token to paste). That external agent is your front door: it creates tickets and projects, kicks off planning and reviews, and answers "what's on my plate."
The Navigator chat inside the console is the FALLBACK surface for the same conversation when you don't have your agent handy. It is not the primary surface — it mirrors what your agent can do, and most operators live in their agent, not here.
The console gives you a hub, a per-item approval page (/approve/{id}), Settings, and this fallback chat — there is no Inbox mailbox screen anymore. Open questions and approvals come to you through your agent (or a deep-linked approval page), not a mailbox you scroll. Knowledge and Process stay reachable under Settings.
What you can actually do here
Four things, day to day.
Plan. You describe an initiative. The conversation lands as a project description on your tracker — motivation, scope, constraints, and decisions in one body of text. Child tickets stay short and link back to the project. The pattern is project-first; the next page explains why.
Run the workspace. "What's on my plate this morning?" — ask your agent and Ship reads your open questions, in-flight work, and open pull requests and answers in one paragraph. "Snooze that improvement until next week." "Why did that routine fail last night?" Everything the console can show is also something your agent can fetch over MCP, and the agent path is usually faster. (The console chat can do the same when you're already in the browser.)
Think out loud about product. You float an idea. Ship checks the standing knowledge — has the team already decided this? If the idea is worth a ticket, it gets one. If it belongs in an existing project, the description grows. If it's a clarification you should answer first, it comes back to you as a clarification — answer it from your agent (or the deep-linked approval page) and Ship resumes the work automatically.
Look things up. Where is this function defined? What does our brand voice doc say about pricing language? When was this routine last green? Ship has read access to your code, your knowledge, and your audit log, and the chat is the fastest way to surface anything from any of them without changing tabs.
What the team sees
The chat is private to you. Nobody else in the workspace reads your conversations. What the team sees is the result: the ticket you created, the project description that got longer, the approval you granted, the pull request that got opened. Those carry the trail even though the conversation that produced them is yours alone. Your half-formed thoughts about what to build next don't belong on a shared board; the decisions that came out of them do.
What it won't do without you
Ship pauses before anything irreversible. Creating a ticket, updating a project body, parking a draft, kicking off multi-step work — for every action that changes the state of the workspace, Ship describes what it's about to do and waits for an OK. That's the difference between "the system did something" and "I asked for that, here's the proof."
If you give a direct instruction ("create a ticket for the login bug") Ship moves on it. If your message was a question or ambiguous ("we should probably do something about the login flow") Ship answers and waits. The bar is roughly: would a thoughtful colleague hit save, or would they sketch it on the whiteboard first.
What it won't pretend to know
Ship reads instead of guessing. Ask "how many open inbox items?" and the answer comes from your actual Inbox. Mention a ticket id and Ship looks at the ticket. Mention a file path and Ship reads the file. When something isn't available — a doc that doesn't exist, a metric that wasn't tracked — Ship says so plainly. "I don't have that" is a better answer than a confident guess, and it's the answer you'll get.
The next three pages walk the three behaviors that carry across whichever surface you use — your agent or this fallback chat: project-first planning, memory across sessions, and switching direction mid-conversation.