Display Name
Alfred
Category
Tooling
Sub-Category
Tooling: Orchestrators
Primary Link
https://github.com/luminik-io/alfred-os
Author Name
prasadus92
Author Link
https://github.com/prasadus92
License
MIT
Other License
No response
Description
A self-hosted runtime for autonomous Claude Code and Codex agents. Scoped GitHub issues and specs are intake; each firing runs in a per-firing git worktree and ships a reviewed pull request, follow-up tests, and a Slack summary. Engines are routable per role across Claude Code and Codex, on the operator's own CLI subscription.
Validate Claims
Clone https://github.com/luminik-io/alfred-os and run the bundled dry-run example with no host configuration at all:
git clone https://github.com/luminik-io/alfred-os.git
cd alfred-os
PYTHONPATH=lib python3 examples/bin/echo_summarise.py --dry-run
It runs the full firing lifecycle (preflight, pick, claim, worktree, invoke, act, release, report) with every side-effecting boundary stubbed - no LLM call, no spend, no Slack, no GitHub mutation, no real repo - and exits 0 with a narrated, step-numbered trace. The same pattern works for examples/bin/hello.py and bin/lucius.py. There is no auth requirement and no API key to obtain to see the full execution path.
For an end-to-end live run, the docs at https://alfred.luminik.io/getting-started/install/ walk through the 30-minute setup. The "Why Alfred exists" page summarises what Alfred coordinates that an interactive coding agent does not.
Specific Task(s)
Run the zero-config dry-run example so you can see the full firing lifecycle without installing anything beyond Python and cloning the repo:
PYTHONPATH=lib python3 examples/bin/echo_summarise.py --dry-run
Then, if you want to see the lifecycle with Claude Code involved (still in dry-run, no spend or side effects), run:
PYTHONPATH=lib python3 bin/lucius.py --dry-run
Specific Prompt(s)
No Claude prompt is required to verify Alfred. The dry-run prints a step-numbered narration of the actual firing lifecycle that the production runner executes. If you want to evaluate Alfred against your existing Claude Code setup, the docs at https://alfred.luminik.io/getting-started/install/ describe the install path; the dry-run is the lowest-friction way to confirm the runtime shape first.
Additional Comments
Alfred is a runtime for the work that keeps coming after the keyboard closes: planned features, follow-up tests, reviewer comments, dependency bumps, docs gaps, multi-repo rollouts. The differentiation against the interactive use of Claude Code (or Codex) is the runtime layer: scoped GitHub intake, per-firing git worktrees, role-based engine routing, hard spend caps, and a state machine that keeps concurrent agents from colliding. Self-hosted, MIT, runs on macOS launchd or Linux systemd, and uses the user's existing Claude Code or Codex CLI auth rather than a separate provider API key.
Recommendation Checklist
Display Name
Alfred
Category
Tooling
Sub-Category
Tooling: Orchestrators
Primary Link
https://github.com/luminik-io/alfred-os
Author Name
prasadus92
Author Link
https://github.com/prasadus92
License
MIT
Other License
No response
Description
A self-hosted runtime for autonomous Claude Code and Codex agents. Scoped GitHub issues and specs are intake; each firing runs in a per-firing git worktree and ships a reviewed pull request, follow-up tests, and a Slack summary. Engines are routable per role across Claude Code and Codex, on the operator's own CLI subscription.
Validate Claims
Clone https://github.com/luminik-io/alfred-os and run the bundled dry-run example with no host configuration at all:
It runs the full firing lifecycle (preflight, pick, claim, worktree, invoke, act, release, report) with every side-effecting boundary stubbed - no LLM call, no spend, no Slack, no GitHub mutation, no real repo - and exits 0 with a narrated, step-numbered trace. The same pattern works for examples/bin/hello.py and bin/lucius.py. There is no auth requirement and no API key to obtain to see the full execution path.
For an end-to-end live run, the docs at https://alfred.luminik.io/getting-started/install/ walk through the 30-minute setup. The "Why Alfred exists" page summarises what Alfred coordinates that an interactive coding agent does not.
Specific Task(s)
Run the zero-config dry-run example so you can see the full firing lifecycle without installing anything beyond Python and cloning the repo:
PYTHONPATH=lib python3 examples/bin/echo_summarise.py --dry-runThen, if you want to see the lifecycle with Claude Code involved (still in dry-run, no spend or side effects), run:
PYTHONPATH=lib python3 bin/lucius.py --dry-runSpecific Prompt(s)
No Claude prompt is required to verify Alfred. The dry-run prints a step-numbered narration of the actual firing lifecycle that the production runner executes. If you want to evaluate Alfred against your existing Claude Code setup, the docs at https://alfred.luminik.io/getting-started/install/ describe the install path; the dry-run is the lowest-friction way to confirm the runtime shape first.
Additional Comments
Alfred is a runtime for the work that keeps coming after the keyboard closes: planned features, follow-up tests, reviewer comments, dependency bumps, docs gaps, multi-repo rollouts. The differentiation against the interactive use of Claude Code (or Codex) is the runtime layer: scoped GitHub intake, per-firing git worktrees, role-based engine routing, hard spend caps, and a state machine that keeps concurrent agents from colliding. Self-hosted, MIT, runs on macOS launchd or Linux systemd, and uses the user's existing Claude Code or Codex CLI auth rather than a separate provider API key.
Recommendation Checklist