https://github.com/jyao97/xylocopa
Manage your AI worker bees easily β all in one, all at once. Could possibly be your handy workspace for the AI agent era π
https://github.com/jyao97/xylocopa
agentic-coding ai-agents claude claude-code developer-tools mobile-first multi-agent pwa task-management tmux vibe-coding
Last synced: 21 days ago
JSON representation
Manage your AI worker bees easily β all in one, all at once. Could possibly be your handy workspace for the AI agent era π
- Host: GitHub
- URL: https://github.com/jyao97/xylocopa
- Owner: jyao97
- License: apache-2.0
- Created: 2026-02-24T12:55:33.000Z (3 months ago)
- Default Branch: master
- Last Pushed: 2026-04-19T00:29:50.000Z (about 1 month ago)
- Last Synced: 2026-04-19T00:32:26.280Z (about 1 month ago)
- Topics: agentic-coding, ai-agents, claude, claude-code, developer-tools, mobile-first, multi-agent, pwa, task-management, tmux, vibe-coding
- Language: Python
- Homepage:
- Size: 8.43 MB
- Stars: 20
- Watchers: 0
- Forks: 4
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
- Code of conduct: CODE_OF_CONDUCT.md
Awesome Lists containing this project
README
# Xylocopa
[](LICENSE)
[](https://www.python.org/downloads/)
[](https://react.dev)
[](https://fastapi.tiangolo.com)
> [**The Loop**](#the-loop) Β· [**Getting Started**](#getting-started) Β· [**Features**](#features) Β· [**Architecture**](docs/ARCHITECTURE.md) Β· [**Contributing**](CONTRIBUTING.md) Β· [**ζ°ζε
₯ι¨οΌδΈζοΌ**](docs/getting-started-zh.md)
_Named after [Xylocopa caerulea](https://en.wikipedia.org/wiki/Xylocopa_caerulea): the blue carpenter bee._
Xylocopa aims to help you navigate attention across many [Claude Code](https://docs.anthropic.com/en/docs/claude-code) projects. Inspired by [GTD](https://gettingthingsdone.com/what-is-gtd/) β except here, agents do the work, not you.
If you find Xylocopa useful, a star helps others find it :)
| 
| 
| 
|
|:---:|:---:|:---:|
| Inbox | Projects | Agents |
| 
| 
| 
|
| Chat | Git | Monitor |
## The Loop
Classic [GTD](https://gettingthingsdone.com/what-is-gtd/) has one human do all five steps (capture, clarify, organize, reflect, engage). Xylocopa keeps the loop but rewires the executor: **you capture and decide, agents execute, the system remembers.**
### 1. Capture
Get ideas into the system, fast, from anywhere.
- **Inbox**: a persistent queue across all your projects.
- **Voice input**: speech-to-text for quick ideas on the go.
- **Lightning input**: title, project, go.
- **Draft persistence**: every keystroke cached locally, survives app close or network drop.
### 2. Dispatch
Turn tasks into agents and let them run.
- **Task β Agent**: one click. Pick a model (Opus/Sonnet/Haiku), optionally enable **Auto mode** (`--dangerously-skip-permissions`; destructive commands still blocked by the [safety hook](#safety-guardrails)).
- **Parallel execution**: run many agents at once, each in its own isolated git worktree.
- **AI batch dispatch**: one click to triage and fire off a pile of inbox tasks.
- **RAG-powered context**: new agents are seeded with relevant lessons from past sessions in the same project.
- **Cross-session reference**: tell an agent "check xy session ``" and it reads another agent's curated display file via a built-in [MCP server](orchestrator/mcp_server.py), ~54Γ fewer tokens than raw JSONL.
### 3. Monitor
Watch everything in real time, from desk or phone.
- **Mobile-first PWA**: add to Home Screen on iOS/Android.
- **Split screen**: 2/3/4 panes side by side, each navigating independently.
- **Attention button**: draggable FAB with a cyan unread badge. Tap for the oldest unread (FIFO), long-press for split screen.
- **Rich chat**: markdown, inline media, interactive cards for tool approvals and plan review.
- **Compact chat header**: status dot next to the title, tinted Stop / Resume / OK pills on row 1; project / worktree icon-pill / Auto / Task / 4-char id pill collapsed into one tag row underneath. Hover or long-press the id pill for a `xylo id:` popover, double-click to copy; same interaction on the worktree pill.
- **Dual-directional CLI sync**: CLI sessions show up in the web app; web sessions are resumable from the CLI via `tmux attach -t xy-` (legacy `ah-` still recognized).
- **Smart notifications**: Web Push, suppressed when you're already viewing the agent (WebSocket or tmux). Permission requests always cut through.
- **Context usage pill**: per-agent context-window meter on the chat header β live percentage with a tap-to-expand breakdown (system / tools / MCP / messages / cache split, free vs. used). Inline suggestion when usage gets high. Counts come straight from the Claude Code session JSONL, not estimated.
- **System monitoring**: disk, memory, GPU, token usage. Per-agent **lifetime cost** (cumulative spend across resumes, deduped by message id; Opus / Sonnet / Haiku priced separately, with 5m/1h cache split).
- **E-ink display mode**: a Settings β Display toggle (and auto-detect for BOOX / Onyx / Kindle / Bigme / Hisense / Meebook / iReader user agents) that flattens glass effects, collapses colored badges to grayscale, and switches saturated bubbles to outlined style for readable contrast on e-paper screens.
### 4. Review
Check results, give feedback, grow the knowledge.
- **Mark done**: approve and close the task.
- **Try β Summarize β Retry**: stop a miss, Xylocopa auto-summarizes what was tried, the next agent picks up from there.
- **Git operations**: diffs, commit history, branch status per project. One-click cleanup and push.
- **Curated project memory**: lessons accumulate in a per-project `PROGRESS.md` that you edit and review; relevant entries are retrieved (top-k) for future agents.
### 5. Remember
Knowledge compounds across sessions, projects, and time.
- **Project memory**: per-project `PROGRESS.md`, fully UI-managed.
- **Session archive**: every conversation persisted, searchable, starrable.
- **Bookmark messages**: double-tap a chat bubble β **Bookmark** to save standout turns. Each bookmark stores your optional note plus an LLM-generated summary + emoji; a per-project **Bookmarks** section lists them, tap a row to jump back with a yellow focus-flash on the original turn.
- **Resume anytime**: pick up any agent where it left off.
- **Full-text search**: across tasks, messages, and sessions.
- **Weekly progress stats**: see the trend, not just the backlog.
- **Automatic backups**: DB + session history + project configs, on a configurable schedule.
## Why Xylocopa?
### Why not just use `claude`?
Vanilla `claude` is fine for one-off sessions. It frays once you run several in parallel, across multiple projects, over multiple days.
- **Attention across agents**: one **Attention button** with a badge for unread/waiting agents. Tap for the oldest, long-press for split-screen. Per-project dashboards with weekly stats, LLM-generated recaps, and an in-project search bar.
- **Capture on the go**: a PWA with voice input (Whisper), β‘ quick-save to inbox, triage at the desk later. Every keystroke auto-drafts locally across 13+ input surfaces.
- **Find and resume old work**: full-text search across every session, with star-to-pin. One-click resume brings STOPPED/ERROR agents back (re-sync existing tmux, or relaunch via `claude --resume`).
- **Retry instead of rewrite**: Try β Summarize β Retry auto-generates what was tried; the next agent picks up with it in context. Durable lessons roll into per-project `PROGRESS.md` and are re-surfaced via RAG.
- **Rich content + cheap cross-references**: inline rendering for images, PDFs, media, and LaTeX (KaTeX). Agents reference each other's sessions via a built-in MCP server, ~54Γ fewer tokens than raw JSONL.
`claude` still runs the show. Xylocopa is the task, attention, and memory layer around it.
### Lessons Compound
**Agents miss. Xylocopa makes the loop after the miss part of the workflow.** Most agent tools assume the agent gets it right. Xylocopa assumes it won't. One click turns a miss into a summary, the next agent picks up from there, and the durable lessons accumulate per project in `PROGRESS.md`, not per session.
### Zero Migration Cost
Xylocopa wraps the same `claude` CLI you already use, launched inside tmux sessions on your machine. Your CLAUDE.md files, project setup, and workflow carry over. The only new dependencies are **tmux** and optionally **Tailscale** for remote access. No new APIs, no lock-in.
### Built for Reliability
Xylocopa hooks into Claude Code's native event system (no polling, no heuristics). Message delivery uses stop-hook dispatch with guaranteed ordering; session lifecycle is tracked via SessionStart/SessionEnd. Each agent runs in its own tmux session on a dedicated git worktree. A deterministic `PreToolUse` [safety hook](#safety-guardrails) hard-blocks destructive operations even under `--dangerously-skip-permissions`.
### Durable by Default
Nothing you run through Xylocopa is ephemeral. Every layer is designed to survive restarts, crashes, and process kills:
- **30s incremental session cache** ([`session_cache.py`](orchestrator/session_cache.py)): active JSONL is append-only-cached like git packfiles; truncated lines auto-repaired on restore.
- **Unlimited retention**: sets `cleanupPeriodDays=36500` in `~/.claude/settings.json` so Claude Code never deletes history.
- **Tmux-anchored recovery**: agents with live tmux panes are re-linked without interruption on restart, your agents survive the web app.
- **One-click resume** ([`routers/agents.py`](orchestrator/routers/agents.py)): STOPPED/ERROR agents resume via re-sync or `claude --resume`.
- **Periodic backups** ([`backup.py`](orchestrator/backup.py)): DB + configs + session history, runtime-configurable interval and retention.
- **Local draft persistence** ([`useDraft.js`](frontend/src/hooks/useDraft.js)): every text input caches to `localStorage` across 13+ surfaces.
- **Orphan cleanup** ([`orphan_cleanup.py`](orchestrator/orphan_cleanup.py)): stale worktrees, zombie tmux, and tempfiles from dead processes are swept periodically.
> Every bullet is open source and linked to its implementation. Audit it, don't trust it.
## Features
### Highlights
- **Try β Summarize β Retry**: when an agent misses the mark, one click captures what was tried; the next dispatch picks up from there instead of starting cold.
- **RAG-powered context**: new agents are seeded with relevant lessons from past sessions in the same project, retrieved automatically at dispatch time.
- **Dual-directional CLI sync**: every agent runs in a tmux session you can attach to from your terminal; sessions you start in the CLI also appear in the web UI.
- **Crash-proof by design**: 30s incremental session cache, partial output salvage on restart, unlimited session retention, and one-click resume of stopped agents. Your work survives the app. See [Durable by Default](#durable-by-default).
- **Deterministic [safety hook](#safety-guardrails)**: `PreToolUse` hard-blocks destructive commands (`rm -rf`, force-pushes, `DROP TABLE`, out-of-project writes), even when agents run with `--dangerously-skip-permissions`.
### Full feature list
| Category | What you get |
|---|---|
| **Smart Notifications** | Hook-based notification system with dual-channel in-use detection, automatically notifies when you're away and stays quiet when you're present. Web Push (VAPID). Per-agent mute, global toggles. |
| **Task Management** | Inbox with drag-to-reorder. Voice input. Lightning capture. Draft persistence. Per-project organization. Retry with auto-summarization. |
| **Agent Control** | Start, stop, **one-click resume** of STOPPED/ERROR agents (re-sync to existing tmux or relaunch via `claude --resume`). Per-agent model selection (Opus/Sonnet/Haiku). Configurable timeouts and permission modes. AI batch dispatch. RAG-powered context from past sessions. Cross-session reference via MCP, agents read each other's curated display files (~54Γ fewer tokens than raw JSONL), keeping cross-references fast and context-window-friendly. **Context usage pill** with breakdown (system/tools/MCP/messages/cache) read from session JSONL. Per-agent **lifetime cost** tracking. **Subagent visibility**: `Agent`-tool sub-sessions discovered under `/subagents/` and surfaced in a Task β Xylo β CC β Sub-session hierarchy. **System / meta-agents** (Task-AI, merge, etc.) hosted on a synthetic `.xylo-internal` placeholder, no longer required to be bound to a real project. |
| **Chat Interface** | Rich markdown rendering (code blocks, tables, images). Inline media preview. Plan mode with approve/reject. Interactive tool confirmation cards. |
| **Monitoring** | Split screen (up to 4 panes). Real-time WebSocket streaming. System monitor (disk, memory, GPU, tokens). Weekly progress stats. |
| **Mobile PWA** | Add to Home Screen on iOS/Android. Full functionality, voice input, push notifications, task management. **E-ink display mode** (manual toggle in Settings β Display, plus auto-detect for BOOX / Onyx / Kindle / Bigme / Hisense / Meebook / iReader user agents) for high-contrast rendering on e-paper readers. |
| **CLI Session Sync** | Dual-directional: CLI sessions in the web app, web app sessions resumable from CLI. |
| **Git Integration** | Commit history, diffs, branch status per project. Agents work in isolated worktrees. One-click cleanup and push. |
| **Session History** | Every conversation persisted and searchable. Star sessions. Resume any agent anytime. Full-text search. |
| **Bookmarks** | Double-tap a chat bubble β **Bookmark** to save standout turns. Each bookmark stores your optional note plus a `gpt-4o-mini`-generated summary + emoji, plus media references extracted from the bubble and Β±2 neighboring turns. Per-project **Bookmarks** section, tap a row for focus-flash navigation back to the original turn. |
| **Security** | Password auth with exponential-backoff rate limiting. Inactivity lock. HTTPS encryption. |
| **Safety Guardrails** | Deterministic `PreToolUse` hook hard-blocks destructive operations, `rm -rf`, `git push --force`, `git reset --hard` outside worktrees, `git clean -f`, `git checkout -- .` / `git restore .`, `DROP TABLE` / `TRUNCATE`, and any `Write`/`Edit` to paths outside the project directory. Enforced even when **Auto mode** (`--dangerously-skip-permissions`) is on. |
| **Reliability & Recovery** | 30s incremental session JSONL cache (append-only, like git packfiles). **Unlimited retention**: `cleanupPeriodDays=36500` prevents Claude from deleting your history. Orchestrator-restart recovery re-links live tmux agents without interrupting them. Automatic periodic DB + config + session backups (runtime-configurable interval & retention). Truncated JSONL auto-repaired. Orphan worktree/tmux cleanup. See [Durable by Default](#durable-by-default) for source pointers. |
## Before You Install
A few things worth knowing before running this on your dev machine.
### Where does my data live?
- **SQLite DB**: `data/orchestrator.db` in the install directory (tasks, projects, agent metadata, configs)
- **Agent sessions**: `~/.claude/projects//*.jsonl` (Claude Code's native session JSONL; Xylocopa doesn't duplicate these)
- **Per-project memory**: `/PROGRESS.md` inside each project's git repo
- **Backups**: `backups/` (rolling DB + session snapshots, see [Durable by Default](#durable-by-default))
- **Uploaded files**: `~/.xylocopa/uploads/`
To capture everything in one snapshot, back up the install directory and `~/.claude/projects/` together.
### How do I uninstall it?
```bash
# Stop the services
pm2 delete xylocopa-backend xylocopa-frontend && pm2 save
# Remove the install
rm -rf ~/xylocopa-main # or wherever you cloned it
rm -rf ~/.xylocopa # uploaded files
# Optional: remove your project directories too
rm -rf ~/xylocopa-projects
# Optional: restore Claude Code's default session-cleanup window
# (Xylocopa sets cleanupPeriodDays=36500 in ~/.claude/settings.json)
```
Project code, git history, and Claude Code session JSONL files in `~/.claude/projects/` are untouched by the uninstall.
## Getting Started
### Host Setup
#### Prerequisites
- **Linux** or **macOS** host (Ubuntu 22.04+ / macOS 13+ recommended)
- **Node.js** 18+ and npm
- **Python** 3.11+
- **tmux** (usually pre-installed; `sudo apt install tmux` if not)
- **Claude Code CLI**: `npm install -g @anthropic-ai/claude-code`, then run `claude` once interactively to log in (Xylocopa reuses the credentials in `~/.claude/`). On a headless server with no browser, use `claude setup-token` instead.
- **Claude subscription**: Claude Max or Pro (uses your existing subscription, no separate API billing)
- **OpenAI API key** _(optional, for voice input)_
#### Third-party / Local Models *(optional)*
Claude Code itself supports Amazon Bedrock, Google Vertex AI, and LLM gateways like LiteLLM (which can front local models such as Llama). Configure it the standard way β set the relevant environment variables in your shell or in `.env`. Xylocopa launches `claude` as a subprocess and inherits them transparently. A practical walkthrough (LiteLLM + local models with Claude Code) is at [unsloth.ai/docs/basics/claude-code](https://unsloth.ai/docs/basics/claude-code).
> **Heads-up on UI scope:** Xylocopa's model dropdown is hardcoded to Anthropic's `claude-*` model IDs (Opus / Sonnet / Haiku). Bedrock / Vertex generally just work because they reuse the same model names. For non-Anthropic backends (LiteLLM β Llama, etc.), you can still run them via the `CC_MODEL` default in `.env`, but per-agent model switching from the UI isn't wired up. Exposing a custom-model field is a small tweak (`frontend/src/lib/constants.js` + `VALID_MODELS` in `orchestrator/config.py`), but it's outside Xylocopa's maintained scope.
#### Installation
Fastest path (clones + runs the interactive installer):
```bash
curl -fsSL https://raw.githubusercontent.com/jyao97/xylocopa/master/setup.sh | bash
```
This installs into `~/xylocopa-main` and prompts for your projects directory, default Claude model, OpenAI API key (optional), and ports. It writes `.env`, generates SSL certs, installs Python and Node dependencies, and launches the services. No manual `.env` editing required.
If you prefer to clone manually:
```bash
git clone https://github.com/jyao97/xylocopa.git ~/xylocopa-main
cd ~/xylocopa-main
./setup.sh # same interactive prompts as above
./run.sh start
```
Verify by opening `https://:3000` on the host. Find your machine's LAN IP with `hostname -I` on Linux or `ipconfig getifaddr en0` on macOS.
> **Tip:** You can also run `claude` in an empty directory and tell it to set up Xylocopa for you :)
> **Tip:** Symlink the Xylocopa repo into `~/xylocopa-projects/` to personalize your experience, let agents improve the tool while you use it.
#### Auto-Start on Reboot (PM2)
Strongly recommended, not just for reboot survival: this step also moves the pm2 daemon out of the terminal session that spawned it. On Linux that matters because systemd-oomd can SIGKILL an entire terminal cgroup (e.g. GNOME Terminal's `vte-spawn-*.scope`) under memory pressure, taking backend+frontend with it. On macOS the equivalent benefit is that pm2 no longer dies if you close Terminal.app.
```bash
./run.sh startup
```
This runs `pm2 save` + `pm2 startup` with auto-detection (systemd on Linux, launchd on macOS). On Linux it will print a `sudo env PATH=... pm2 startup systemd ...` line, copy and run it exactly as shown. On macOS no sudo is needed.
To disable later: `pm2 unstartup` (same auto-detection).
#### Set Up Your Projects
Add projects in the app: **long-press the + button β New Project**: paste any GitHub URL or point to an empty folder. You can also manually create or symlink folders in the projects directory (`~/xylocopa-projects/` by default, configured via `HOST_PROJECTS_DIR` in `.env`).
### Client Setup
After setting up the host, visit `https://:3000` from any device with network access, that's it. Set a password on first visit.
#### Remote Access
For access outside your LAN, Xylocopa works with any tunneling or VPN solution, [Tailscale](https://tailscale.com), [ZeroTier](https://www.zerotier.com), [WireGuard](https://www.wireguard.com), [frp](https://github.com/fatedier/frp), Cloudflare Tunnel, etc. The author uses Tailscale:
1. Install [Tailscale](https://tailscale.com) on your server and phone
2. `tailscale up` on both devices
3. Access Xylocopa at `https://:3000`
No port forwarding, no public exposure, traffic stays in an encrypted tunnel between your devices.
#### iPhone PWA
If you want the full app experience on iPhone (home screen icon, fullscreen, push notifications):
1. Open `https://:3000` in Safari (bypass the certificate warning via **Advanced β Visit Website**, then refresh).
2. Follow the on-screen guide on the login page to install the CA certificate and the Xylocopa app.
#### Installing the CA Certificate
Xylocopa uses a self-signed SSL certificate. The host trusts it after setup, but other client devices will show a browser warning until you install the cert. iPhone/iPad users can skip this, the [iPhone PWA](#iphone-pwa) guide above already covers it.
For Android, macOS, Windows, and Linux, see [detailed instructions](docs/install-cert.md).
## Telemetry
Xylocopa sends **one anonymous event per day** β `daily_heartbeat` β to help me see if the project is being used. Payload: random `install_id` (UUID v4 generated locally), `version`, `platform` (`darwin` / `linux` / `win32`), `timestamp`. **That's everything.** No IPs, no prompts, no code, no file paths, no hostnames, nothing user-generated.
Client: [`orchestrator/telemetry.py`](orchestrator/telemetry.py) β sends to a Cloudflare Worker the author owns ([source](https://github.com/jyao97/xylocopa-telemetry)) that writes to a private D1 database. No third-party analytics.
**Disable** (any one is enough): toggle off in **Monitor β Help improve Xylocopa**, or set `XYLOCOPA_TELEMETRY=0`, or write `telemetry: false` in `~/.xylocopa/config.yaml`.
## Gestures & Shortcuts
- **Short-press the + button** to quickly add a task. **Long-press** it to choose between adding a project, agent, or task.
- **Long-press a card** (Inbox task, Agent row, Project tile, or row inside a Project / Trash list) to enter multi-select mode β the pressed card is pre-selected. Tap others to add/remove, then use the bottom bar to bulk-act. Tap **Done** in the header to exit.
- **Inbox**: AI batch-process / Start / Delete the selection.
- **Agents** (and inside a Project's agent tab): mark Read / Stop / Delete the selection β including stopping or deleting agents in bulk (no per-row delete button; long-press is the entry point).
- **Projects**: Activate / Archive / Delete; the Activate and Archive buttons light up only when the selection is uniformly archived or active, otherwise only Delete is available.
- **Trash**: bulk Restore / permanently Delete deleted projects.
- **ID pill** in the chat header (4-char monospace, row 2): hover or long-press for a `xylo id:` popover with a Copy button; double-click to copy directly. Same pattern on the **worktree pill**.
- **Double-tap a message** in the chat view to open its action menu β Copy / Modify / Delete / Bookmark. Only one menu is open at a time; an outside tap dismisses. **Bookmark** opens an inline note prompt; skip it and the AI-generated summary is used as the title.
- **Double-tap a tab** in the bottom nav to scroll to the first unread / pending item.
## Agent control plane
Xylocopa-managed agents can call back into the orchestrator via a built-in
MCP server β list/create/dispatch tasks, read each other's sessions,
scaffold projects, check health. The surface is deliberately
non-destructive (verb whitelist + blacklist).
See **[docs/agent-mcp-tools.md](docs/agent-mcp-tools.md)** for the full
tool list, safety model, and what's intentionally not exposed.
## Troubleshooting
- **Conversation stuck or not updating?** Click the **refresh button** at the top of the chat view to re-sync the session from the CLI.
- **Agent shows IDLE after server restart but is still running?** Normal. Status restores to EXECUTING on the next tool call (heartbeat via the `agent-tool-activity` hook). Send a message to trigger activity if it's in a long thinking phase.
- **Don't name tmux sessions with the `xy-` or `ah-` prefix** β those are Xylocopa's managed prefixes (`xy-{id}`; legacy `ah-{id}` still recognized). User-created sessions with those prefixes won't be detected.
- **PWA stuck on a perpetual loading screen?** Stale Service Worker precache. Run `.venv/bin/python tools/push_reset.py` on the host for an interactive picker (lists devices with labels and last-ack times; `a` resets all, `q` quits). Then fully close the PWA on the device and reopen it. Direct forms `list`/``/`all` are supported for scripting.
## Contributing
We welcome contributions! See [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines on:
- Reporting bugs and suggesting features
- Setting up a development environment
- Running tests and submitting pull requests
## Migration from AgentHive
Xylocopa was previously named **AgentHive**. The upgrade is backward compatible, no manual migration needed:
- **CLI**: `agenthive` stays as a symlink to `xylocopa`.
- **Install dir / env vars**: `AGENTHIVE_DIR` and `AGENTHIVE_MANAGED` still honored alongside the new `XYLOCOPA_*` names. `~/agenthive-main` checkouts keep working.
- **Process names**: `pm2` processes are now `xylocopa-backend` / `xylocopa-frontend` (upgrade script removes the legacy `agenthive-*` entries).
- **MCP / tmux**: `.mcp.json` entry renamed to `xylocopa` on first agent start; new agents use `xy-{id}` prefix, legacy `ah-{id}` sessions are still recognized so in-flight agents survive.
- **Data dirs**: `~/.agenthive/uploads` auto-renames to `~/.xylocopa/uploads` on first backend start (if the new path doesn't exist).
- **Browser / certs / Web Clip**: `localStorage` keys auto-migrate on first page load; existing certs keep working; re-download `Xylocopa.mobileconfig` from the login page if you want the renamed Home Screen entry.
To rename your install dir: `mv ~/agenthive-main ~/xylocopa-main && cd ~/xylocopa-main && ./run.sh restart`.
## License
Apache 2.0, see [LICENSE](LICENSE) for details.