What Is Arvre? The Beginner's Guide to Managing Claude Code Sessions in the Terminal
CodexFuse Team · 2026-11-04T23:43:18.012Z · 12 min read min read
If you have been wondering **what is arvre**, here is the short answer: Arvre is a terminal graph of your Claude Code sessions and their forks. It shows every session as a tree, so you can navigate through them, resume sessions, see which session was forked from which, and jump between parents and children — all without leaving the terminal. Those two sentences are nearly verbatim from the project's own PyPI page (pypi.org/project/arvre, fetched 2026-11-04), because at three days old there is no better definition than the author's. This arvre explained guide covers what the tool is, the launch story, the problem it solves, how its fork detection works, the install arc, every command and key, and the honest caveats you should know before adopting a week-old project. If you run your coding agents on a Linux box that stays on, start with our VPS hosting guide — a small always-on server is the natural place to park long-running Claude Code sessions so they keep building while your laptop sleeps. ## What Is Arvre, Exactly? Arvre is an open-source terminal user interface (the arvre TUI) that draws your Claude Code sessions as a tree — the arvre claude code session graph the tagline promises. The project's tagline is "Terminal graph of Claude Code sessions and their forks," and the full description puts it plainly: "arvre shows your Claude Code sessions and their forks as a tree. Navigate through them, resume sessions, see which session was forked from which, and jump between parents and children." (pypi.org/project/arvre) The launch arc is part of the story: per the launch coverage on emergent.sh, Arvre "officially launched November 2, 2026 on GitHub" and "Creator Victor Carrico vibe coded the tool over one weekend to manage multiple concurrent Claude Code task workflows," going "from concept to public release in approximately 48 hours." One date note: the GitHub repo (github.com/victorcarrico/arvre) was actually created October 31, 2026, and version 0.1.0 hit PyPI the same day — so November 2 is the public launch date per the news piece, with the repo predating it by two days. As of this writing (2026-11-04), the current release is 0.2.1, uploaded November 3. The creator-name record is worth getting right: PyPI's author field says "Victor Carrico," while the GitHub account victorcarrico shows the display name "Victor Carriço Santos." The project is MIT-licensed, built on Python's Textual framework, and free. ## What Problem Does Arvre Solve? It solves fork sprawl: navigating arvre session forks without a map, where dozens of Claude Code sessions accumulate and you cannot remember which fork holds the working version of an idea. Claude Code keeps its own session list, but it is flat — there is no record of which session was forked from which, so a forked experiment and its parent look like two unrelated entries. Say this once, because it frames everything else: **arvre manages sessions — it does not run or spawn coding agents itself.** It is a navigator over history Claude Code already wrote to disk, not an agent runner. That is also what makes it different from the web dashboards some developers build: the news piece describes Arvre as targeting "developers juggling numerous AI-assisted coding sessions who prefer terminal-based navigation over web interfaces." If you live in the terminal, the tree is where the sessions live. ## How Does Arvre Figure Out Which Session Was Forked From Which? It reconstructs the family tree from message fingerprints, because Claude Code never writes down the answer. The project ships a design document, docs/forks.md, that explains the mechanism with unusual honesty — including what its author verified and what he has not. The key fact, verified by the author "with Claude Code 2.1.314 by creating sessions in a scratch project": "No explicit parent field is written (`forkedFrom`, `parentSessionId`, etc. are absent)." When you fork a session, Claude Code simply copies the parent's messages: "The fork file is a copy of the parent's records with the same message `uuid`s, with `sessionId` rewritten to the new id." In the author's test, 15 of the fork's 20 uuid-bearing records were copied from the parent. Arvre's parent rule follows from that: "For each session X with first timestamp `t(X)`: Candidates are sessions Y where `first(Y) < first(X)` and `uuids(Y) ∩ uuids(X) ≠ ∅`. Parent = candidate with the largest intersection. Ties go to the earliest `first(Y)`. Sessions with no candidate are roots." The document even anticipates the edge case: "Known limitation: siblings forked from the same parent share the parent's prefix, so the rule links each sibling to the parent (not to each other). This is the desired result. Siblings are not linked to each other." Two caveats travel with the mechanism. First, "Session data is read from `~/.claude/projects`," and "This is an undocumented format, so a future Claude Code release could change it" — the README says it outright. Second, the author flags what he has not yet verified: "`/branch` and the interactive `/rename` paths were not run (non-interactive mode only)," so treat fork detection around those paths as untested until a later release confirms it. ## How Do You Install Arvre? The arvre install route is deliberately minimal: one command, one prerequisite. This arvre tutorial walks through the whole install in five steps, and it takes about a minute. First, meet the prerequisites: Python 3.11 or newer, and Claude Code installed with `claude` on your PATH — the README is explicit: "You need Claude Code installed, with `claude` on your `PATH`, for the "open in this terminal" action." If Claude Code itself is new to you, our Claude Code beginner guide walks through the first session. Second, install with the only route the documentation shows: "With pipx (macOS and Linux, Python 3.11+):" followed by `pipx install arvre`. Note the scope: the README documents macOS and Linux only — there is no documented Windows install route, and `pip install` or `uv tool install` are never shown, so do not assume them. Third, launch it from anywhere with `arvre` — by default it shows all sessions from every project as a flowchart. Fourth, run `arvre --help` for all options, since the README ends its usage section with exactly that line. Fifth, learn the two scoping flags you will reach for most: `--this-dir` for only the current directory's sessions, and `--last-day` for only the last 24 hours. ## What Do the Commands and Keys Do? The six invocation forms from the README, each with its documented meaning: | Command | What it does | | --- | --- | | `arvre` | All sessions from every project, as a flowchart | | `arvre --this-dir` | Only sessions from the current directory | | `arvre --last-day` | Only sessions from the last 24 hours | | `arvre --style compact` | Start with the compact tree instead | | `arvre --theme ocean` | Start with a colour theme | | `arvre --print` | Print the tree as text and exit | Inside the TUI, navigation is vim-flavored. The full key map, verbatim from the project's Keys table: | Key | Action | | --- | --- | | Up / Down (or k / j) | Move. In compact style, through the tree. In flowchart style, to the next session at the same depth. | | Left / Right (or h / l) | Go to the parent / first child | | Enter | Open the action menu for the selected session (or click it twice) | | a | Toggle last 24 hours / all time | | p | Toggle this directory / all projects | | v | Toggle compact / flowchart style | | t | Cycle colour themes (`classic`, `ocean`, `forest`, `sunset`) | | r | Reload | | q | Quit | Pressing `a`, `p`, and `v` inside the app mirrors the `--last-day`, `--this-dir`, and `--style` flags, so you rarely need to quit and relaunch to change the view. ## Flowchart or Compact: Which View Should You Use? Use the flowchart view when you are hunting a fork: it lays sessions out so parent-child links read left to right, and the view where the "which session was forked from which" question answers itself visually. Use the compact view when you just need the list: a dense tree of titles you can move through with the arrow keys. Toggle between them with `v` at any time, or start directly in compact mode with `arvre --style compact`. The four colour themes (`classic`, `ocean`, `forest`, `sunset`) cycle with `t` — purely cosmetic, but `ocean` reads best on dark terminals. ## What Can You Do From the Action Menu? Selecting a session and pressing Enter opens the action menu, and its two actions are the whole reason the tool exists. "Open in this terminal: runs `claude --resume <id>` in the session's directory, and returns to arvre when Claude exits." That is the resume loop: find the session in the tree, open it where you stand, and land back in the tree when you quit Claude. "Copy command: copies `cd '<dir>' && claude --resume <id>` to the clipboard, for pasting into another terminal." That covers the other workflow: you are browsing sessions on one machine or pane and want to resume the session somewhere else. Both actions depend on the prerequisite from the install section — without `claude` on your PATH, there is nothing to resume with. ## Arvre vs Worktrunk: Sessions or Worktrees? Arvre and Worktrunk solve adjacent problems in the same parallel-agent workflow, and drawing the line once saves confusion: **forks are conversation branches inside Claude Code; worktrees are filesystem checkouts.** Worktrunk is a CLI for git worktree management — one branch, one worktree, one agent, each with its own directory. Arvre navigates Claude Code sessions and their forks — the conversation history, not the files. They are never rivals. In practice they compose: Worktrunk gives each parallel agent its own worktree, and each of those agents accumulates sessions and forks that Arvre can map. If the worktree side is new to you, our Worktrunk beginner guide covers the parallel-agent pattern from the filesystem angle. ## Where Does Arvre Fit With Beacon and the Rest of the Ecosystem? The Claude Code ecosystem is splitting into lanes, and Arvre occupies exactly one. Beacon.sh is the memory lane: it captures finished sessions and distills them into reusable knowledge — what the agent remembers. Arvre is the navigation lane: finding and resuming the sessions themselves — where the agent's history lives. Our beacon.sh beginner guide covers the memory side in depth. The emerging picture is a small toolchain rather than one tool: the agent runs (Claude Code), its checkouts get organized (Worktrunk), its sessions get navigated (Arvre), and its lessons get remembered (beacon.sh). Each lane is one page, one job, no overlap. ## The Honest Caveats Before You Adopt Arvre Arvre is days old, and the numbers say so plainly: the repo carried 2 stars, 0 forks, and 6 commits when checked on 2026-11-04, with three releases (0.1.0, 0.2.0, 0.2.1) landing within four days. Early-stage software moves fast and breaks things — adopt it as a handy viewer, not as load-bearing infrastructure. The structural risk is the data source: everything Arvre knows comes from `~/.claude/projects`, "an undocumented format, so a future Claude Code release could change it." If Anthropic renames a field, the tree goes dark until the author ships a fix. Relatedly, fork detection was "verified with Claude Code 2.1.314" only — other Claude Code versions are untested by the author's own account. The author's own caveats deserve the last word: the `/branch` and interactive `/rename` paths "were not run," and whether `/resume` hides empty sessions is still an open question in the design doc. Read docs/forks.md before you trust the tree for anything important. And one naming trap to avoid, since it pollutes the search results: `aru-code` on PyPI is a different, unrelated LLM-agent package — not Arvre, not affiliated, not a fork. ## Sources - pypi.org/project/arvre — package page (tagline, description, install route, usage lines, keys table, action menu, fork-detection explanation, data-source caveat, version 0.2.1 metadata). Fetched 2026-11-04. - github.com/victorcarrico/arvre — repository (creation date 2026-10-31, 2 stars / 6 commits / 3 releases as of 2026-11-04, MIT license, creator account victorcarrico). Fetched 2026-11-04. - github.com/victorcarrico/arvre/blob/main/docs/forks.md — "How Claude Code records forks (Phase 0 findings)" (uuid-intersection parent rule, title priority, timestamp note, author caveats). Fetched 2026-11-04. - emergent.sh/news/arvre-tui-launches-claude-code-session — launch coverage (November 2, 2026 launch, weekend vibe-coded build, ~48 hours, Hacker News posting claim). Published 2026-11-03, fetched 2026-11-04. ## Frequently Asked Questions ### Is Arvre free? Yes. Arvre is open source under the MIT license — you can install and use it for free, with no account, no sign-up, and no cloud component. It reads session files already on your machine. ### Does Arvre run coding agents for you? No. Arvre manages sessions; it does not run or spawn agents itself. It draws the tree of sessions Claude Code already recorded and lets you navigate and resume them — the coding still happens in Claude Code. ### How does Arvre know which session was forked from which? Claude Code writes no parent link, so Arvre compares message fingerprints: a fork is a copy of its parent's messages with the same message IDs, so each session is linked to the earlier session that shares the most message IDs with it. The full rule, with its sibling edge case, is documented in the project's docs/forks.md. ### Is Arvre the same as aru-code? No. `aru-code` is an unrelated PyPI package — an LLM coding assistant built on a different framework. It shares a few letters of the name and nothing else. If a search result talks about Agno agents and model switching, you are looking at the wrong package. ### Does Arvre upload my sessions anywhere? Nothing in the documentation describes any upload, account, or sync. Arvre reads the session files Claude Code already stores locally under `~/.claude/projects` and renders them in your terminal. Your sessions stay on your machine.
