Selora AI: Library of Recipes

Overview

Home Assistant blueprints let users share and reuse automations. Selora AI recipes take the same idea one level up: each recipe is a declarative package that describes an end-to-end smart-home outcome, and Selora AI configures everything for the homeowner.

A recipe ships as a directory containing presentation copy, a declarative manifest (the required integrations, devices, roles, inputs, and integration setup), and a package/ of Jinja-templated YAML the deterministic pipeline applies to Home Assistant.

Anyone can contribute a recipe. The community curates, reviews, and improves the library over time, similar to the Home Assistant blueprints exchange but purpose-built for AI-driven home automation.

Example

User: “I want my home to save energy when nobody’s around.”

Selora AI: “There’s a community recipe called Away Mode Energy Saver that uses your presence sensors and smart plugs to cut standby power when the house is empty. Want me to install it?”

User: “Yes, set it up.”

Selora AI: “Done. I’ve configured it with your motion sensors and smart plugs. It will activate after 30 minutes of no presence. You can adjust the delay anytime.”

Customer value

• Blueprints for AI: like Home Assistant blueprints, but instead of static YAML automations, recipes give Selora AI contextual intent. The agent adapts to device availability and handles edge cases conversationally.
• Community-driven: anyone can author and share recipes. The library grows organically with real-world use cases contributed by homeowners, installers, and developers.
• No code required: users discover, install, and configure recipes entirely through natural language in the Selora AI chat. No YAML, no templates, no dashboards.
• Reversible by contract: the pipeline tracks every artifact it writes under selora.recipes.<slug>.* so uninstall reverts cleanly. Recipes cannot opt out of reversibility.
• Hardware-aware: recipes declare their device and integration requirements, so Selora AI only suggests recipes compatible with the user’s actual setup.

Scope (first iterations)

• Recipe specification: define the package format and manifest schema (roles, inputs, integrations with auto-setup, package files). Required files are index.md, manifest.yaml, and a non-empty package/; CHANGELOG.md is optional.
• Catalog and discovery: a public catalog of recipes browsable on the marketing site, served as JSON so the agent can consume it programmatically alongside its companion device, role, and integration catalogs.
• Recipe tarballs: each recipe is also published as a downloadable tarball so the agent can fetch and apply it as one atomic package.
• Agent ingestion: the Selora AI agent on each instance downloads recipes on demand and executes the directives locally.
• Conversational discovery and deployment: users browse, install, and configure recipes directly from the Selora AI chat interface. One conversation, zero YAML.
• Versioning: recipes follow Semantic Versioning. The pipeline uninstalls the old version and installs the new one on major bumps; minor and patch bumps reapply the rendered package safely.

Architecture (directional)

• The marketing site ships the canonical recipe library: one Hugo page per recipe plus the catalog JSON endpoints (/api/recipes.json, /api/devices.json, /api/roles.json, /api/integrations.json).
• Per-customer install-day recipes are generated by Connect from a homeowner’s installation manifest. They are packaged in the same recipe format and reference canonical recipes from the site library, plus manifest-specific glue (pairing steps, integration credentials, room mappings).
• The Selora AI agent on each instance consumes either source: a public recipe pulled from the site, or a generated install-day recipe pulled from Connect.
• The pipeline applies the rendered package/ files through its registered Skills . Recipes stay declarative; the agent’s capabilities live separately.

Target customers

• Homeowners: access advanced AI behaviors with a single conversation, no technical knowledge needed.
• Installers: deploy proven, standardized recipe sets across client installations for consistent results. Connect generates install-day recipes from the customer’s manifest automatically.
• Developers and contributors: share expertise with the community and build reputation through a public contribution model.

Open questions and risks

• Security and validation: how to review community-contributed recipes before they reach user instances, especially recipes that interact with security-sensitive devices (locks, alarms, cameras). Reviewers focus on manifest.yaml and the package files since those are the user-visible promise of what the pipeline will do.
• Recipe quality: establishing review criteria and a curation process to maintain library quality as contributions scale. A two-tier model (Verified vs Community) lets the homeowner opt in to broader content.
• Proactive recommendations: integration with the Suggest Integrations & Devices feature to proactively recommend recipes based on the user’s hardware.
• Attribution and incentives: recognition or incentive models for active contributors.
• Offline catalog: how to handle recipe discovery and updates when the instance has limited or no internet connectivity.

Related

• Selora AI: Skills (Epic TBD): Roadmap — the capabilities the agent uses to execute recipe directives. Recipes are declarative; Skills are imperative.
• Selora AI: Embedded Claw (Epic #45): Roadmap — the runtime that ingests recipes and runs Skills on each Selora instance.
• Selora AI: Suggest Integrations & Devices (Epic #51): GitLab epic #51 — proactive hardware recommendations that can trigger recipe suggestions.
• Selora AI: Proactive Automation Suggestions (Epic #47): Roadmap — complementary feature that suggests automations; recipes provide the reusable building blocks.
• Selora AI: Create Scenes (Epic #46): Roadmap — recipes can bundle scene-creation capabilities for specific use cases.