Integrations

Alfred core should stay boring: local Python runners, host-scheduler units, git worktrees, GitHub CLI, model CLIs, local Redis Agent Memory, FleetBrain, and optional Slack delivery.

The default model path uses local Claude Code / Codex CLI authentication. Alfred does not manage provider API keys.

ALFRED_HOME is the runtime root. A fresh install defaults to ~/.alfred. Alfred uses ALFRED_HOME only for its runtime path.

Bundling policy

Alfred may include adapters, examples, docs, and small optional CLIs that fail gracefully when a provider is missing.

Alfred should not bundle private memory stores, canon files, transcripts, a hosted vector database, a required external agent runtime, or third-party skills fetched at runtime without review.

Profiles

Profile	Use it when	Required pieces
Standalone Alfred	You want scheduled engineering agents on a local Mac or Linux host.	Alfred, Python, gh, git, Claude Code or Codex, local Redis Agent Memory, optional Slack webhook.
Alfred + personal memory	You want Alfred to consult your own notes or knowledge base as a fallback.	Standalone Alfred plus a read-only provider such as gbrain.
Alfred + control gateway	You want chat control, MCP registration, skills, durable task boards, or dashboards around Alfred.	Standalone Alfred plus a separately installed gateway/control layer.

The standalone profile is the only path the installer assumes.

Good next adapters

• alfred mcp serve for read-only fleet status and shipped summaries.
• alfred events export for sanitized memory/dashboard imports.
• alfred gateway bridge for GitHub issue/label handoff from an external board.
• alfred dashboards export for static fleet snapshots.

All should be additive. New users should still be able to run Alfred without installing any of them.