hive-desktop

Hive Desktop is a local desktop application that gathers the work demanding attention -- pull requests, issues, review requests, notifications, firing alerts, and anything able to POST a webhook -- into a single queue. Rather than juggling many browser tabs, users configure flows that filter incoming items, route them into feeds, hand them to a coding agent, or trigger a command, aiming for one notification per genuinely new change. All processing happens on the user's machine: triaging inside Hive does not write back to GitHub, and configuration is plain YAML that can be kept in dotfiles. Installation on macOS is via a shell one-liner from hivedesktop.com that detects OS and CPU, resolves the newest build from the same channel manifest the in-app updater uses, verifies the SHA-256 checksum before writing to disk, installs the app, and symlinks the `hive` command onto the PATH. Linux support is described as wired into the installer but awaiting published builds; the project states macOS is supported today. Key capabilities described in the README: - Sources beyond GitHub: point a source at any GitHub search or notification inbox, across multiple accounts, or pair it with Grafana alerts, a PromQL expression, or anything that can POST JSON to a local endpoint. - Rules as programs: sources are wired through filters and functions on a canvas. Filters are declarative and their reject branch can be wired onward. Functions are user-written JavaScript with durable per-node memory, allowing one alert to fan out into a tracked item per firing entity. - Notification discipline: re-polling the same pull request does not re-fire; a cooldown limits how often one item can reach the user; stale pings are dropped rather than replayed as a burst; a restart recomputes every feed without double-notifying. - Agent-authored rules: Hive renders skill files from its node registry into `~/.claude`, `~/.codex`, `~/.pi` and `~/.agents`, keeping them in sync, so a user can describe what should be surfaced and have a coding agent write the flow. - Triage to action: an item can launch a coding-agent session from a prompt template, run a shell command, publish a message for another session, or render to the clipboard. - Owned configuration: flows and actions live as YAML in the config directory; Hive validates on save and reloads live, keeping the last good version in service if a file fails to build. Two additional surfaces ship switched off while they are still being settled: Code, a tmux-backed terminal for the sessions feeds launch, and Agents, named workspaces that generate the configuration a coding agent reads. Documentation is hosted at hivedesktop.com/getting-started (also served as llms.txt, with source under web/docs/), covering install and first run, the Inbox, Code and Chats areas, building from source, and troubleshooting. The repository also contains docs/architecture.md, docs/source-pipeline.md (ingestion, flows, membership replay, retention, actions), architecture decision records under docs/decisions/, plus development and distribution guides. Contribution guidance is in docs/development.md and CONTRIBUTING.md, and issues are tracked on GitHub. The project is licensed under MIT.