Chat-driven development
Describe what you want in natural language. The agent plans the change, edits files, runs commands, reads logs, and iterates until it works.
Everything around the agent — on one surface.
ShipIt runs Claude Code or Codex through the full ship-real-software loop. Build, plan, review, and ship — without leaving the browser, without juggling tabs, and without losing context between tools.
FIG. 1 — BUILD
Build
Everything you need to go from prompt to working code, with the agent doing the bulk of the typing — and a live app to test against.
Pluggable agent backend
Pick Claude Code CLI or Codex CLI per session — agent-agnostic, with more runtimes on the way. Connect several accounts per provider, and ShipIt moves to the next one before a usage limit stops you, without interrupting sessions already in flight.
Compose-native live preview
Embedded iframe shows your app updating in real time, with HMR proxied through ShipIt, multi-port support, and Docker Compose services managed per session.
Agent-controlled services
The agent starts, stops, restarts, and tails any Compose service your project declares — including the manual ones it needs on demand, like a database to migrate or a cache to flush. A debug dependency never waits on a click from you.
Android build, test & preview
The session image bakes in a JDK, the Android SDK, and Gradle. The agent builds Gradle projects (web/Android monorepos included), runs JVM and Paparazzi snapshot tests, and reads the PNG diffs — and an emulator declared as a Compose service shows the running app live in the preview panel.
Git LFS support
Repos with large assets just work: LFS files arrive as real files, not pointer stubs, they're shared across sessions instead of re-downloaded every time, and LFS images show up as images in the diff viewer.
Project templates
Quick-start scaffolding for React, Vue, Next.js, Svelte, and more — start from a clean baseline in seconds.
File & image input
Drop files into the chat. The agent reads them as additional context for the next turn — PDFs, screenshots, design exports, logs.
File viewer with visual diffs
Browse the workspace with syntax highlighting and review changes as inline diffs — including image and SVG diffs rendered visually instead of as walls of text.
Presented artifacts
The agent renders diagrams, charts, mockups, prototypes, and formatted markdown into a dedicated Present tab — no dev server required, browsable as a gallery.
Agent Interface SDK
JavaScript in a preview or a presented artifact can compose and send messages back to the agent that owns the session, so an agent-built interface is something you interact with, not just look at.
Sub-agent transparency
When the agent fans work out to sub-agents, their prompt, work timeline, and final report render inline instead of an opaque tool call — and any in-flight tool call opens a live-updating output dialog.
Interactive terminal
Full PTY (xterm.js) inside the session container for ad-hoc debugging, log tailing, or one-off commands.
Persistent logs
Agent-container and preview-service logs are kept in a durable, disk-backed store, so full history survives container restarts, idle eviction, and orchestrator restarts.
MCP integration
Connect Model Context Protocol servers to give the agent more tools — internal APIs, observability platforms, vector stores, whatever you need.
FIG. 2 — PLAN & TRACK
Plan & track
Linear and GitHub Issues live inside ShipIt — so "what should I work on next?" and "start a session for it" are the same place.
Inline Issues tab
Linear and GitHub Issues in one priority-sorted list, with a sub-tab per tracker. Set status (both trackers) and priority (Linear) inline from the list or detail view.
Filters & search
Narrow by status, priority, and assignee (multi-select) or free-text search, applied across every connected tracker.
Extra tracker tabs
A repository can declare additional GitHub issue trackers in its shipit.yaml, each getting its own tab — so a separate planning repo sits alongside the code repo's issues instead of in another browser window.
Start a session from an issue
Kick off an isolated session straight from an issue row, with the issue as context — instead of copy-pasting the body into chat. Pick which repository it starts in when the issue isn't tracked in the one you're looking at.
Agent issue access
The agent reads and writes issues (view, list, comment, edit, set status and assignee, create issues and labels, nest Linear sub-issues) through a tracker-neutral, ShipIt-brokered interface, so tracker tokens never enter the session container.
Requirements before code
Opt a feature into requirements discipline and the agent writes down what it must do in your words first, asks its open questions as one batched prompt, and holds off on implementation until you've answered.
FIG. 3 — REVIEW & SHIP
Review & ship
PRs, CI, deploys, and review comments all render inline. The diff doesn't leave the conversation, and failures loop back to the agent.
Inline PR lifecycle card
Title, description, CI checks, review threads, deploy status, and merge state all render in chat as a single live card. No GitHub tab required.
AI PR descriptions
When you open a PR, ShipIt generates the description from the actual diff — accurate and consistent on day one.
Cross-agent second opinions
Opt in to let the session agent consult a different model for a one-shot review or sub-task — inline in the same turn, with full context, no separate session required. A long consult runs in the background as a card you can watch, cancel, or come back to, and survives an orchestrator restart.
Inline diffs
File changes display as collapsible red/green diff blocks right in the conversation, so context never leaves the chat.
Auto-deploy on push
Deploy status surfaces inline on the PR card via the GitHub Deployments API — the same place your CI checks live.
PR comment sync
Review threads from GitHub appear inline in the conversation, so reviewer comments become first-class context for the agent.
CI failure loop
Failed GitHub checks and logs are surfaced to the agent so it can inspect the failure and, when enabled, attempt a fix on the next turn.
Preview failure loop
A crashed Compose service is detected and auto-retried, with its logs surfaced inline so the agent — and you — can act without leaving chat.
Auto-resolve merge conflicts
When your branch conflicts with its base and the agent is idle, ShipIt auto-rebases and runs a turn to resolve the conflicts for you.
Stay current with the base
A behind-the-base branch gets an inline nudge and a one-click sync that also advances the session's local base ref — with the rebase's progress shown live in chat and a persistent card recording that it happened.
Approval gate & merge-on-green
Merge eligibility reflects GitHub's review-approval status, and you can arm a trivial task to auto-merge once checks pass — set right at session start.
Ship a chain of PRs
A session can opt into being woken the moment its PR lands: it resets to the freshly merged base and keeps going on its own, so stacked work ships without you shepherding each merge by hand.
FIG. 4 — ITERATE SAFELY
Iterate safely
Auto-commits, fully isolated sessions, and live steering — so trying things never costs you the last hour of work.
Git as undo
Every agent turn auto-commits. Rewind to any previous state and fork into a new branch from any point in history.
Parallel PR-shaped sessions
Spawn separate workspaces with their own branch, container, and chat history, and review each as its own PR. A spawned session can push a finding back to the session that started it — and its siblings — as a card plus a queued turn, so a blocker one agent hits reaches the others.
Work survives updates & crashes
Update ShipIt and your running sessions keep going — a turn already in flight is picked up and finished, commit, push, and PR included. If a turn dies unexpectedly, its work is committed rather than left stranded.
Guardrails on risky moves
A commit that would leak a recognized secret is blocked and shown to you with what to do about it, and destructive git on already-merged work is routed through a safer reset that leaves a record behind.
Fully isolated sessions
Every session on the same repo gets its own clone and its own containerized environment, so its agent and services never share state with another session.
Sandbox sessions
Start a repo-less session from an empty workspace; the agent clones what it needs, with Git and session-scoped Docker granted as explicit capability toggles at creation.
Permission modes
Choose how much autonomy the agent has per session — from confirm-every-edit to fully autonomous.
Live steering
Interrupt and redirect the agent mid-turn without losing context. Course-correct as you read along.
Session sidebar
Pinned sessions, status indicators, AI-generated names the agent keeps current as the work changes (until you rename one yourself), and a hide toggle for repositories you're not working in right now.
FIG. 5 — EVERYWHERE
Everywhere
Mobile-first, voice-enabled, and remotely updatable. ShipIt comes with you.
Mobile-first
Tab-based layout on small screens, resizable split panels on desktop. Genuinely usable from a phone — not a cramped afterthought.
Installable app
Install ShipIt to your phone's home screen and it runs standalone and full-screen with no address bar — and always boots the latest code rather than a cached build.
Voice in and out
Dictate prompts, hear a spoken note when an agent needs you, and tap play to hear a completed turn read aloud — usable hands-free. Notes can play in-app, go to a webhook, or both, so "the agent needs you" can reach whatever notifier you already run.
Background notifications
Optional browser notification and sound when the agent finishes a turn — keep working elsewhere and come back when it's done.
Quick capture
A global hotkey opens an overlay that captures a prompt and spawns a new session in the background, without leaving what you're doing.
Software updates
VPS installs update and restart from Settings → Advanced; local Docker installs choose a channel there, then apply with the update script.
Usage & spend at a glance
Header badges show your Claude/Codex rate-limit usage — 5-hour window, weekly cap, reset clock — named per account when you've connected several, with weekly and per-month spend trends in the detail view.
FIG. 6 — COMPARE & CATEGORY
How ShipIt compares
Evaluating ShipIt against other tools, or trying to place the category? These pages are factual and balanced — including where ShipIt is the weaker pick.
FIG. 7 — NEXT STEP
Ready to try it?
Run one command and ShipIt is live on your machine at localhost:4123.