Xirp: Spotify's Tool for Managing Multiple AI Coding Agent Sessions

On August 10, 2026, Spotify Engineering opened Xirp to public beta. It's a vendor-neutral agentic development environment built internally to solve a specific problem: managing dozens of concurrent AI agent sessions across Claude Code, Gemini CLI, and Codex without losing track of what's running, where, and what it's working on. More than 1,300 Spotify engineers have been using it internally across 36,000+ sessions. It's now available as a free public beta, gated behind a Spotify Technology account.

This article covers what Xirp does, how its session management works, and what the notable features and tradeoffs look like.

What problem it solves

Running a single AI coding agent is straightforward. Running five simultaneously across two projects, with some blocked on others and some waiting for review, quickly becomes difficult to track with a collection of terminal windows. Xirp wraps the CLI harnesses (Claude Code, Gemini CLI, Codex) in a unified GUI where you can see all sessions, monitor their output, switch between them, and manage dependencies between them.

Each session runs in its own git worktree, which means dozens of agents can work concurrently on the same codebase without interfering with each other's file state. Context decouples from any single agent or harness, so switching models mid-session carries the full working state to the new agent rather than starting from scratch.

The other component is Portal integration. Xirp connects to Spotify's Portal (their internal developer platform built on Backstage), which means sessions start with awareness of service ownership, dependencies, architectural decisions, and documentation rather than just the files in the current directory. For external users, you get a Portal instance included with the beta.

Getting started

Xirp requires a Spotify Technology account, separate from a regular Spotify account. This is the first point of friction and a deliberate choice: the tool is hosted by Spotify, tied to their infrastructure, and subject to their data retention policies.

Once logged in, the home page presents a prompt box ("What are we shipping? Describe your goal or paste a ticket URL") and project management controls.

Adding a project takes two inputs: the absolute path to the local project directory and an optional friendly name. This defines the working directory for agent sessions and scopes their file access.

The project dashboard

The projects overview shows each registered codebase with its current git branch, the number of active sessions, total historical sessions, and when it was last modified.

Clicking a project opens its command center: a prompt box for launching new sessions and a card-based view of all active sessions for that project. The project view also has tabs for Git, Files, Skills, and Rules, covering the organizational context layer.

Session management

The core of Xirp is the session view. When you open an active session, you see a live terminal with the agent's full output, an input field at the bottom for sending new instructions, and a sidebar listing all projects and their active sessions.

Switching between sessions is a single click in the sidebar. The design is conceptually similar to tmux but without the terminal learning curve.

Grid view

For monitoring multiple sessions simultaneously, the grid view puts several terminal panes on screen at once. You can configure layouts (2x1, 2x2, 3x3), drag sessions from the sidebar into the grid, and resize panes by dragging their borders. Regular terminal sessions can also be added alongside agent sessions.

This is the most immediately useful feature for teams running several agents in parallel. Having one agent implement a feature, another run tests, and a third watching for regressions, all visible on a single screen, is much easier to track than tabbing between terminal windows.

Notable features

Session forking

Forking creates an exact duplicate of a session including its full conversation history and context. The fork lets you try a different prompting approach without touching the original session. If the experiment fails, close the fork; if it succeeds, continue from there. This is useful for testing different instructions on a complex task without risking the agent's accumulated context.

Agent swapping

Different models perform differently on different tasks. Xirp lets you switch the underlying agent mid-session. Start a task with Claude, then switch to Codex for review, and the full session context carries over automatically. The new agent picks up where the previous one left off without requiring you to re-establish context manually.

Dependent sessions

Sessions can be set as children of other sessions. A child session waits in a holding state until its parent completes, then starts automatically. This enables simple sequential pipelines: implement a feature in the parent session, run integration tests in the child session after it finishes. The dependency is configured at session creation time.

Tradeoffs worth knowing

Xirp is closed-source and hosted by Spotify. Your session data passes through their infrastructure, subject to their data retention policies. For teams with strict data handling requirements, this is a meaningful constraint.

The Spotify Technology account requirement gates access in a way that open-source alternatives don't. Tools like OpenCode Desktop or a custom tmux setup require no registration and run entirely locally.

The Portal integration is the most distinctive part of Xirp's pitch: sessions that start with awareness of service ownership, architectural decisions, and team context rather than just file contents. For external users without a large existing knowledge graph to plug in, this benefit is less immediately apparent and will take time to build up.

The tool launched on August 10, 2026. For anything beyond experimentation, watching the beta's stability over the next few months before relying on it for production workflows is sensible.