Xirp Review (2026): Pricing, Features & Honest Verdict

Xirp is an agentic development environment from Spotify that runs Claude Code, Codex and Gemini CLI side by side, each session in its own Git worktree, with your org’s context loaded in. Best for: engineering teams already running multiple coding agents on one codebase. Price: free in public beta, macOS only. Rating: 8/10.

What is Xirp?

Xirp is a macOS desktop app that sits above your coding agents instead of replacing them. You point it at a repository, spin up a session, and pick whichever agent you want for that job: Claude Code, OpenAI Codex, Gemini CLI, or an open-source model. Each session gets its own Git worktree, so a dozen agents can work on the same codebase at once without stepping on each other’s branches. The context you build up in a session belongs to Xirp, not to the agent, which means you can switch from Claude to Codex halfway through a task and the working state carries over.

Spotify Engineering opened it to public beta on 10 August 2026 after running it internally for months. The numbers they published are the interesting part: thousands of Spotify engineers adopted it organically across more than 36,000 agent sessions, and teams have run 50-plus concurrent sessions against a single codebase. This is not a launch-day prototype. It is internal tooling that got big enough inside a company with thousands of engineers that Spotify decided to do what it did with Backstage and push it outward. That track record is the strongest thing Xirp has going for it, and it is why we are rating it above most of what launches on Product Hunt in a given week.

What are Xirp’s key features?

Vendor-neutral session management

The core of the product is the session grid. Every task you hand to an agent becomes a session with its own worktree, its own terminal, and its own transcript. You can see all of them at once in grid view, which is the thing that makes running 20 agents feel like managing work rather than juggling 20 terminal tabs. Crucially, the agent is a per-session choice, not a global commitment. If Claude Code is rate-limited or a task suits Gemini’s context window better, you switch and keep going. Most agent wrappers pick a lane and lock you in. Xirp explicitly refuses to, and the marketing line for this is blunt: “Works with Claude, Gemini, Codex. Knows your org, not theirs.”

Institutional memory through Portal

Connect Xirp to Spotify Portal, the commercial Backstage distribution, and sessions start with your service catalog loaded: component architecture, dependency graphs, ownership data, architectural decision records. The agent knows that the service it is about to change has three upstream consumers and who owns them. That is the difference between an agent that edits a file correctly and one that understands why the file is written the way it is. Session transcripts then flow back into Portal so the next engineer, or the next agent, inherits what was learned.

Living documentation from sessions

Xirp generates and maintains documentation out of what actually happened in coding sessions rather than asking engineers to write it afterwards. This is the feature most likely to be oversold, and we would treat it as a nice byproduct rather than a reason to adopt. Auto-generated docs are only as good as the sessions behind them, and a messy exploratory session produces messy prose. But for the specific case of “why did we change this and what did we try first”, the transcripts are genuinely more useful than a commit message.

Rules, skills and persistent terminals

Underneath the session grid, Xirp carries the working furniture you expect from a real development environment rather than a wrapper. Persistent terminals survive across a session, so an agent that started a dev server does not orphan it when you switch away. Rules and skills are defined at the Xirp level and apply to whichever agent you point at the task, which is the practical payoff of vendor neutrality: you write your conventions once instead of maintaining a CLAUDE.md, an AGENTS.md and a GEMINI.md that drift apart. Files and worktree state are visible per session, so you can see what an agent has actually changed before you merge anything. None of this is exciting, and all of it is the difference between a demo and something you use every day.

Local projects with no Portal required

Worth being clear on this because the marketing blurs it: Portal is optional. Xirp runs general sessions with no project attached, and project sessions against any folder on your Mac. Local projects, persistent terminals, Git worktrees, files, rules, skills and grid view all work without Portal. You only need Portal for the org-context layer. So a solo developer can install it today and get the multi-agent orchestration part for nothing.

How much does Xirp cost?

Xirp is free during the public beta. Spotify has published no pricing page, no plan tiers, and no stated intention about what happens when the beta ends. You bring your own agent subscriptions, so the real cost of running it is whatever you already pay Anthropic, OpenAI or Google. Xirp itself adds nothing on top today.

The unpriced part is Portal, and that matters if you want the institutional-memory features. Spotify Portal is a commercial product sold on request with enterprise-style pricing, not a card-on-file signup. So the honest read on cost is two-tier: the local multi-agent environment is free and will probably stay cheap, and the org-context layer is an enterprise sale. Compare that to Conductor or Crystal, which wrap coding agents for free with no upsell, or to Warp at $15 per user per month for its agent features. Xirp is cheaper than the paid options right now and carries more risk of a future price, because a free beta from a company that also sells Portal is a funnel, not a gift.

Who is Xirp best for?

Use Xirp if you are already running more than one coding agent, on more than one task, in a codebase big enough that context is the bottleneck. That is the exact problem it was built for inside Spotify and it solves it well. It also fits teams that have deliberately not standardised on a single agent vendor, because switching cost is the thing Xirp is designed to remove. If you already run Backstage or Portal, the case is stronger again, since the catalog you have maintained suddenly becomes agent-readable.

Skip Xirp if you are on Windows or Linux, because it is macOS only and there is no published timeline for anything else. Skip it if you use one agent for one task at a time, since the session grid solves a problem you do not have and plain Claude Code in a terminal is simpler. And skip it if you need vendor stability today, because public beta from a music company means the support model, the roadmap and the eventual price are all unwritten. Spotify has a good record here with Backstage, which went to the CNCF and thrived, but a good record is not a commitment.

Best Xirp alternatives

Conductor is the closest free comparison. It runs multiple Claude Code sessions in parallel with Git worktrees on macOS, which is the same core mechanic. It is Claude-only, so you lose the vendor neutrality, and it has no org-context layer. Free.

Crystal does parallel agent sessions in worktrees as an open-source desktop app, with support for Claude Code and Codex. Open source is a real advantage over a beta from a large company, and it runs beyond macOS. It has a much smaller team and no institutional-memory story. Free.

Warp attacks the same territory from the terminal side, with agent management, multi-agent runs and a polished UI. Warp is a mature paid product with real support, starting around $15 per user per month, but it wants to be your terminal, and it has no service-catalog integration.

Final verdict: is Xirp worth it?

Yes, with a caveat about timing. Xirp is the most credible agent-orchestration tool we have looked at this year, because it is not a launch built to be launched. It is a tool thousands of engineers used 36,000 times before anyone outside Spotify saw it. The worktree-per-session design is the right primitive, the vendor neutrality is a real position rather than a slogan, and the Portal integration is a genuinely novel answer to the question of where agent context should live. Install it today if you have a Mac and more than one agent subscription. It costs nothing and it removes a category of tab-juggling that most teams have quietly accepted.

The caveat is that you are adopting beta software with no pricing, no SLA, and a hard macOS requirement, from a vendor whose most valuable feature is gated behind an enterprise product. Do not build a team process on it yet. Use it, get the parallel-session benefit, and keep your setup portable enough that a price announcement or a Windows-shaped gap does not strand you. We rate it 8/10 for what it does and how well it has been proven, discounted from higher only because the commercial terms do not exist yet.