forgewatch

A single-binary, local-first CI runner. Define a pipeline in YAML, run it from your terminal, or watch it re-run on every save — with a live web dashboard that streams logs as they happen. No Docker, no cloud, no daemon fleet. One Rust binary, one SQLite file, zero external services.

Think of it as the CI you actually run on your machine: the same pipeline that runs in the cloud, but fast, offline, and observable — great for tightening the edit → build → test loop before you ever push.

┌──────────────────────────────────────────────────────────────┐
│  forgewatch serve --watch                                      │
│    • parses forgewatch.yml                                     │
│    • schedules jobs as a DAG (needs:)                          │
│    • runs steps as subprocesses, streams stdout/stderr         │
│    • caches unchanged jobs by content hash                     │
│    • serves a live dashboard at http://127.0.0.1:7878          │
└──────────────────────────────────────────────────────────────┘

Features

• DAG scheduling — jobs declare needs:; independent jobs run concurrently, dependents wait, cycles are rejected up front.
• Content-addressed cache — a job's cache_key globs are hashed; on a match the job is skipped and its cache outputs restored. Rebuild only what changed.
• Live dashboard — a self-contained web UI (zero external requests) streams logs over Server-Sent Events, with per-job status cards and full run history.
• Watch mode — edits matching watch: globs re-trigger the pipeline, debounced, so saving a file kicks off a build automatically.
• Full history — every run, job, and log line is persisted in a WAL-mode SQLite database, so you can replay any past build in the dashboard.
• Single binary — Rust + Tokio + axum. Copy it anywhere and go.

Install

cargo install --path .
# or
cargo build --release   # -> target/release/forgewatch

Quickstart

forgewatch init                 # writes a starter forgewatch.yml
forgewatch run                  # run once, stream logs to the terminal
forgewatch serve --watch        # dashboard at http://127.0.0.1:7878, auto-rerun

Pipeline format

name: demo-ci

# Files that re-trigger the pipeline in `serve --watch`.
watch:
  - "src/**/*"

# Environment inherited by every job and step.
env:
  APP: forgewatch

jobs:
  - name: lint
    steps:
      - name: check
        run: cargo clippy --all-targets

  - name: build
    needs: [lint]              # runs after lint succeeds
    cache_key:                 # hash these inputs...
      - "src/**/*"
      - "Cargo.toml"
    cache:                     # ...restore these outputs on a cache hit
      - "target/release"
    steps:
      - name: compile
        run: cargo build --release

  - name: test
    needs: [build]
    steps:
      - name: unit
        run: cargo test
      - name: bench
        continue_on_error: true   # a non-zero exit here won't fail the job
        run: cargo bench

Step fields

Shell note: variable expansion follows the shell. $VAR works under sh/bash; on the Windows default (cmd) use %VAR%, or set shell: bash -c for a step.

Commands

State lives in .forgewatch/ next to your pipeline (runs.db + cache/).

Examples

Runnable pipelines live in examples/:

forgewatch run --file examples/parallel/forgewatch.yml

How it fits together

main.rs      CLI (init / run / serve)
config       YAML pipeline schema  ......... model.rs
engine       DAG scheduler + streaming steps  engine.rs
store        SQLite run/job/log history ..... store.rs
cache        content-addressed job cache .... cache.rs
watch        debounced filesystem trigger ... watch.rs
server       axum API + SSE + embedded UI ... server.rs + web/index.html

License

MIT — see LICENSE.