# Installation
Bento is a command-line daemon for macOS and Linux. The installer selects the published binary for your machine and installs it in user space.
## Supported platforms
M
Automatic
macOS · Apple Silicon
M-series Macs · arm64
bento-darwin-arm64
M
Automatic
macOS · Intel
Intel Macs · x86\_64
bento-darwin-x64
L
Automatic
Linux · x64
64-bit Linux · x86\_64
bento-linux-x64
L
Automatic
Linux · arm64
64-bit Linux · arm64
bento-linux-arm64
Bento does not support Windows.
## Requirements
Install these before you install the daemon.
* [Docker](https://docs.docker.com/get-docker/) — runs the Postgres database and the sandboxes that hold agent workloads. On macOS, [OrbStack](https://orbstack.dev/) is an alternative Docker runtime.
* [Claude Code](https://claude.ai/code) (`claude`) — the default agent runtime.
* [GitHub CLI](https://cli.github.com/) (`gh`) — authenticates the GitHub integration.
* [QMD](https://github.com/tobilu/qmd) (`qmd`) 2.8.3 or later — the default local retrieval engine for agent knowledge. Install it with `npm install -g @tobilu/qmd`. You can use [Chroma](https://docs.trychroma.com/docs/overview/introduction) instead with `knowledge.retrieval.engine: chroma`. Chroma also needs [uv](https://docs.astral.sh/uv/) on PATH, because its embedder runs on the host.
* SQLite — on macOS, install it with `brew install sqlite`.
Three runtimes are optional, each an alternative to Claude Code: [Codex CLI](https://developers.openai.com/codex/cli) (`codex`), `pi`, and `opencode`.
## Install the daemon
The one-line installer supports every platform listed above. It verifies the release checksum, then installs the binary under `/usr/local/bin`, `/opt/homebrew/bin`, or `~/.local/bin`, depending on directory permissions.
```bash
curl -fsSL https://install.getbento.sh | sh
```
:::warning\[Do not use sudo]
The daemon runs in user space and does not need elevated permissions. If the command is not found after installation, check that `~/.local/bin` is in your `PATH` and that the binary is executable.
:::
To verify the installation, print the version:
```bash
bento --version
```
## Install a specific version
For a reproducible install or a rollback, set `BENTO_RELEASE`:
```bash
curl -fsSL https://install.getbento.sh | BENTO_RELEASE=0.6.5 sh
```
## Bootstrap a project
Go to your project directory and run:
```bash
bento init
```
The command writes six files: `.bento/daemon.yaml`, `.bento/.gitignore`, `.bento/pipelines/pr-review.yaml`, `agents/reviewer/PERSONA.md`, `CLAUDE.md`, and `.mcp.json`. It reads the `origin` git remote and the `gh` login to fill in `repos:` and `github.username`.
## Set up the database
The daemon looks for Postgres at `postgresql://bento:bento@localhost:8421/bento`. To change that address, edit `.bento/daemon.yaml`. To start a Postgres instance at the default address, use Docker:
```bash
curl -fsSL https://install.getbento.sh/services.yml -o docker-compose.yml
docker compose up -d
```
`docker compose down -v` removes the volume and all data.
## Start the daemon
Run the daemon in the foreground:
```bash
bento serve
```
To register it as a background service instead, install it and start it:
```bash
bento daemon install # registers launchd (macOS) or systemd (Linux)
bento daemon start
```
The daemon reports a problem loudly. If it starts and prints no error, it is healthy. To confirm, run the check:
```bash
bento doctor
```
## Your first pipeline
`bento init` scaffolded a `CLAUDE.md` and an `.mcp.json` into your repository. Open the repository in an AI coding agent such as Claude Code, Codex, or Cursor. The agent reads `CLAUDE.md` to learn how bento works, and connects to the `bento` MCP server declared in `.mcp.json`. Ask it to build your first pipeline: an agent persona, the pipeline YAML, the webhook registration, and a test run.
For the concepts behind agents, skills, and pipelines, see [Concepts](/concepts). For the full pipeline field reference, see [Pipelines: Config](/pipelines/config).
## Configuration reference
`.bento/daemon.yaml` controls daemon-wide settings: the database URL, the ports, the sandbox backend, and the pipeline lifecycle hooks. Each file under `.bento/pipelines/` is one pipeline definition. See [Configuration](/configuration) for the full field reference.
**Runtime state** (the PID, the log, and the agent directories) lives under `~/.bento//`, where `name` is the `name:` field in `daemon.yaml` and defaults to `bento`. Workspaces are shared at `~/.bento/workspaces/`. Neither directory is committed to git.