devloop

Keep the local loop cheap.

devloop is a config-driven tool for running multi-process systems locally.

Most local setups become expensive to change:

• restarting everything
• losing state
• waiting for rebuilds
• coordinating multiple services

devloop keeps everything alive so you can change one thing at a time.

A concrete example

Working on a blog post with a public preview:

• change Rust -> the server rebuilds and restarts
• the browser reconnects and reloads
• cloudflared restarts -> new public URL
• the current page path is combined with the new tunnel URL
• the final URL is printed, ready to paste into LinkedIn validator

One change -> everything updates -> copy and paste once.

Without devloop:

• restarting the server does not automatically coordinate the rest of the loop
• you still need to manage CSS rebuilds separately
• the tunnel keeps the old URL unless you restart cloudflared yourself
• you rebuild the full public URL by hand every time

The pieces exist.
They just don’t know about each other.

Install

Install the latest published main branch directly from GitHub:

cargo install --git https://github.com/pasunboneleve/devloop.git

Tagged releases are also published automatically on GitHub with prebuilt release archives for Linux x86_64 and macOS Apple Silicon. Each supported platform publishes its release asset independently, so a failure on one platform does not block the other asset from being attached to the GitHub release.

Supported prebuilt release targets:

• x86_64-unknown-linux-gnu
• aarch64-apple-darwin

For local development from a checkout:

cargo install --path .

Usage

Run devloop in a repository with a devloop.toml config:

devloop run

The tool will:

• start declared processes
• watch configured paths
• execute workflows on change

Built-in reference docs are also available from the CLI:

devloop docs config
devloop docs behavior
devloop docs security

Design

The tool has three layers:

1. Engine Watches files, supervises processes, executes workflows, and stores session state.

2. Repository config Declares watch groups, named processes, workflow steps, and hook commands.

3. Repository hooks Small commands that answer project-specific questions such as "what is the current post slug?" or "what public URL should be printed now?"

Internally, devloop is being refactored toward a pure core plus an imperative shell: workflow orchestration is planned as explicit state/effect transitions, runtime scheduling and process-supervision decisions are planned the same way, and replaceable adapters interpret those effects at the edges.

The session state file is owned by devloop while it is running. External edits to that file are not merged back into the live session; restart the supervisor if you need to seed a different initial state.

Example use case

Used as the primary local development workflow for gcp-rust-blog-public.

The generic example config lives at:

examples/blog/devloop.toml

The real client config lives in the client repository itself:

gcp-rust-blog-public/devloop.toml

It models a blog workflow as configuration:

• rust changes restart the server, wait for health, refresh the current post slug, restart the tunnel, and publish the current post URL
• content changes refresh the current post slug, restart the tunnel, and republish the current post URL
• CSS is handled by a long-running Tailwind watch process started by the startup workflow

The example expects repo-owned helper scripts:

• ./scripts/build-css.sh
• ./scripts/current-post-slug.sh

At the same time, the tunnel itself is described as a managed process:

• cloudflared is started directly by the engine
• stdout is scanned with regex rules
• the matched tunnel URL is written into session state
• readiness waits for the state key to be populated
• restart policy keeps the process alive if it exits
• inherited process output is source-labeled without wrapper scripts

When you need to identify which managed process emitted a line in mixed output, inherited process lines include the executable first and the configured process name second. The label is color-coded per process, and the body style is configurable:

[process.tunnel]
command = ["cloudflared", "tunnel", "--url", "http://127.0.0.1:18080"]
output = { inherit = true, body_style = "plain", rules = [{ state_key = "tunnel_url", extract = "url_token" }] }

That renders inherited lines with the executable and process name, for example [cloudflared tunnel] ..., using ANSI color when stdout is a terminal and NO_COLOR is not set.

For the runtime behavior reference, see docs/behavior.md.

For the full configuration reference, see docs/configuration.md.

For the external-event trust model and push-versus-polling tradeoffs, see docs/security.md.

The client config can then compose derived values with write_state steps, for example:

step = { action = "write_state", key = "current_post_url", value = "{{tunnel_url}}/posts/{{current_post_slug}}"}

Workflows can also emit rendered log lines:

step = { action = "log", message = "current post url: {{current_post_url}}"}

For high-visibility output in a mixed process log, use the boxed style:

step = { action = "log", message = "current post url: {{current_post_url}}", style = "boxed"}

Repeated setup can be factored into helper workflows and reused with run_workflow, for example a publish_post_url workflow that waits for the tunnel and then writes the derived URL.

Hooks can also be observed on the runtime tick when external state changes are not represented by file edits. For example:

[hook.current_post_slug]
command = ["./scripts/current-post-slug.sh"]
capture = "text"
state_key = "current_post_slug"
observe = { workflow = "publish_post_url", interval_ms = 1000 }

That lets a helper hook refresh session state from something like a development server endpoint, and rerun the follow-up workflow only when the state actually changes.

For more precise local event flows, devloop can also accept capability-scoped pushed events over a localhost HTTP server. A trusted client process can post a value to a configured event, devloop updates the mapped session-state key, and then runs the mapped workflow if the value changed. This is the preferred model for things like browser-path updates, while observed hooks remain a simpler fallback.

---

Known gap

Real working configs should live in the client repository, not under devloop/examples/. The example here is intentionally generic.

---

Development

Quality gates:

cargo fmt
cargo test
cargo clippy --all-targets --all-features -- -D warnings

Git hook setup:

git config core.hooksPath .githooks

That enables the versioned pre-commit hook in .githooks/pre-commit, which runs cargo fmt before each commit.

Task tracking:

bd ready
bd show <issue>
bd update <issue> --status in_progress
bd close <issue>