Every Exponential release, newest first. Upgrade with curl -fsSL https://go-exponential.dev/install.sh | bash, or see the installation guide.
Talk to Claude Code or Codex from inside xpo board, then hand it an issue and watch it do the work, with you approving every step that matters.
xpo serve, xpo login and xpo logout. Issue data travels with your code through git, so teammates get the latest copy when they sync. The remote.* config keys are ignored.blocked_by link counts as resolved only when the blocker is DONE. Remove the link, or point it at the issue that replaced it.Press ⌘J anywhere on the board to talk to your local Claude Code or Codex about the project. Chat in a popup, or open the full-page Agent view (g a). Opening an issue adds it to the conversation as context. Attach files and images by pasting or dropping them. Read the docs.
A chat starts in Chat Mode, where the agent reads the repo and manages issues. When it's time to build, it asks to enter Work Mode on an issue. From then on it works in that issue's own worktree, and the conversation carries over. Start Work on any issue opens a new Work Mode chat, and Continue Work picks up where it left off. Read the docs.
Before the agent runs anything risky, such as a merge or an edit outside its worktree, the chat shows an approval card. Allow it once or for the whole chat, or deny it, all from the keyboard. Claude can also ask you questions with multiple-choice cards. Live status dots, a favicon badge and optional desktop notifications tell you when a chat needs you. Read the docs.
null in MCP update and xpo update --json works for assignee, cycle, priority, estimate and parent.XPO_* environment variables set every config key, as documented.Linking issues takes one menu instead of a modal, and the board makes it clear who did the work, you or your agent.
overwrite, which changes the link's kind. Existing pairs keep working until something changes them.xpo init or xpo doctor --fix to refresh your agent instructions.One Relationships button in the issue sidebar opens a menu for every kind of link: parent, sub-issue, related, blocked by, blocking and duplicate. The Relations card groups links by kind, shows a preview when you hover a linked issue, and lets you change a link's kind in place with a right-click. Read the docs.
Board cards and Backlog rows have an assignee picker. Click the avatar, or press a on a focused card or row. Unassigned issues show a placeholder avatar, so the column stays aligned. Read the docs.
When an agent acted for you, your name gets a dashed underline that reveals the agent on hover. Direct actions render plain. The same rule applies across the activity timeline, merge view, dashboard, timeline and inbox.
s, e and l appear next to the focused card instead of the board's top-left corner.Every issue now gets its own git worktree, xpo board stays on your machine unless you say otherwise, and setup takes a single wizard.
merge always runs from the main checkout. Before upgrading, merge or park any issue you started in branch mode, and check out your default branch in the main checkout. xpo start --mode and xpo merge --no-wt no longer exist, and the worktrees config key is ignored.xpo board binds to 127.0.0.1. To reach it from another machine, use ssh -L, or pass --host and sign in with the per-run access token it prints.xpo init mcp and xpo init skill are merged into xpo init.xpo history --json returns a single JSON object instead of JSON lines.Every issue you start gets an isolated git worktree under .xpo/worktrees/, so several agents and sessions can work in parallel without stepping on each other. merge runs from your main checkout and cleans up after itself. Running start again from inside an issue's worktree is safe.
xpo board only listens on your own machine by default, and rejects requests that try to reach it through DNS rebinding. Exposing it on a network now requires a per-run access token.
An issue is always assigned to a person, and work an agent does is recorded on that person's behalf. A new "Worked by" filter separates agent work from human work. Read the docs.
xpo init walks you through the prefix, your agents and a preview of every change before it writes anything. Agent instructions live in managed blocks that refresh safely without touching your own text. xpo doctor --fix repairs what it can. Read the docs.
--json command shares one input and output format with the MCP tools.unlink command and MCP tool remove a single relationship.j and k in an issue walk the list you opened it from.xpo merge works from inside an issue's worktree, and a failed merge leaves the issue as it was.merge on uncommitted work.xpo now configures Claude Code, GitHub Copilot, Cursor, Codex and OpenCode, and the merge view shows changes before they're committed.
xpo init sets up MCP and the xpo workflow skill for Claude Code, GitHub Copilot, Cursor, OpenAI Codex and OpenCode, each in its own format.xpo init skill --global (macOS and Linux).xpo doctor groups its checks into sections, each with the command that fixes it.An interactive dependency graph, searchable design history, and project health at a glance from the terminal.
Explore how issues depend on each other in an interactive graph, or work through them in a filterable table. Open it from any issue's sidebar to start from that issue. Read the docs.
xpo rationale searches past specs and walkthroughs, so agents find earlier decisions before they change code. It's available as an MCP tool too. Read the docs.
The merge view opens on the issue's walkthrough when it has one, so reviewers read the explanation before the diff. Read the docs.
xpo pulse shows velocity, cycle time, work in progress and blockers at a glance. xpo history draws a timeline of events, commits and merges for one issue or the whole project.
[ and ].xpo merge no longer blocks on unrelated untracked files.Several agents and sessions can work on different issues at once, and the Backlog gains new statuses and views.
xpo start gives each issue its own git worktree under .xpo/worktrees/, and xpo merge cleans it up when the work lands. Your main checkout stays on main, so several agents can work in parallel without conflicts, and the board shows every agent's progress live.
Two new closed statuses sit alongside Done. Like Done, they unblock dependent issues and can close a parent, but only Done counts toward velocity. Read the docs.
A Done tab lists finished issues, newest first. Switch any tab between a nested tree of parents and sub-issues and a flat list, and expand or collapse every parent at once. Read the docs.
xpo init can be re-run safely. It refreshes agent instructions without touching your config or issues.xpo merge deletes the issue branch by default.list tool returns issues in your manual order.A formatting bar in every Markdown editor, a guard against losing unsaved issues, and stricter validation everywhere.
p.xpo init detects your coding agents and configures them, and xpo drive shows what it's doing as it happens.
drive config changed: supervisor and coder are now objects with agent and model keys.xpo init detects installed coding agents and writes MCP config, agent instructions and the xpo workflow skill for each one.xpo drive streams the agent's activity live, with collapsible phases.drive.supervisor.model lets planning and review run on a faster, cheaper model.xpo drive reuses agent sessions, which roughly halves token usage..tar.gz and .zip archives.Generated agent instructions now lead with the MCP tools, and building from source no longer needs a frontend build first.
The first public release. Beats is now Exponential, with the xpo command, under the MIT license.
xpo, project data lives in .xpo/, environment variables use the XPO_ prefix, and MCP tools are named xpo_*. There is no backward compatibility with Beats projects.The git-native issue tracker for human-agent teams is open source under the MIT license. Issues, specs and walkthroughs live in your repository next to the code, and your agents work with them through MCP. Get started.