[Submission] Terrace Guardian – AI-Powered Outdoor Monitoring with Temporal + Pydantic AI

[danmarinescu]

Project link

https://github.com/danmarinescu/terrace-guardian

Language

Python

Short description (max 256 chars)

A durable Temporal workflow that monitors an outdoor terrace using Claude's vision capabilities via Pydantic AI. Analyzes photos for birds, dry plants, rain, and hazards — then autonomously plays sounds or sends notifications.

Long Description

Terrace Guardian is an AI-powered outdoor terrace monitoring system that combines Temporal's durable execution with Pydantic AI's agent framework.
A camera captures photos at configurable intervals. Each photo is sent to an LLM (Claude Sonnet 4.6) as a multimodal message through a Pydantic AI agent. Based on the analysis, the agent autonomously calls tools to take action:

• Bird detected → plays a scare sound via macOS afplay
• Plants look dry → sends a macOS notification to water them
• Rain + exposed cushions/furniture → sends an urgent notification
• Fire or other hazards → sends a safety alert

The agent's LLM calls and tool invocations are automatically wrapped as durable Temporal activities using Pydantic AI's TemporalAgent integration — making the entire AI pipeline retryable and observable.

The workflow supports pause/resume signals, configurable intervals, a status query, and continue-as-new to bound history size. A single ./run.sh script handles all setup and runs a 5-photo demo.

The agent is configured to use Claude Sonnet 4.6 by default, but can easily be swapped to any LLM supported by Pydantic AI (OpenAI, Gemini, etc.) — the model is defined in a single place.

Author(s)

Dan Marinescu @danmarinescu

[github-actions]

Hi, I'm ZiggyBot! 🤖 Here's my pre-evaluation of this submission:

---

Criterion	Assessment
Useful to Temporal users	✅ Yes — Demonstrates a compelling real-world use case combining Temporal's durable execution with Pydantic AI's agent framework for autonomous decision-making based on vision analysis.
Benefits clearly explained	✅ Yes — The README clearly explains the architecture, the specific scenarios handled (birds, dry plants, rain, hazards), and why Temporal's durability matters for the AI pipeline.
OSI Approved License	❌ No — No license file was detected in the repository.
README with working instructions	✅ Yes — Comprehensive README with quick start (./run.sh), manual setup instructions, CLI commands for control signals, and clear project structure documentation.

Overall: Needs review ⚠️

Notes:

• This is a creative and well-documented project that showcases an interesting integration pattern between Temporal and Pydantic AI's TemporalAgent wrapper for durable LLM interactions.
• The workflow demonstrates several Temporal best practices: signals for pause/resume, queries for status, continue-as-new to bound history size, and proper retry policies.
• The architecture diagram and test photo table make the project very approachable for newcomers.
• The macOS-specific dependencies (afplay, osascript) limit portability, but this is clearly documented.
• The missing license is the only blocking issue — this needs to be resolved before the project can be accepted into the Code Exchange.

Suggested questions for submitter:

• Could you add an OSI-approved license (MIT is recommended) to the repository? This is required for inclusion in the Code Exchange.
• Have you considered adding cross-platform alternatives for the notification and sound playback tools, or documenting how users on other platforms might adapt the code?

Teaching moment:

This project demonstrates an excellent pattern for using continue-as-new to bound workflow history size in long-running monitoring workflows. The workflow tracks cycle count and triggers continuation after a configurable threshold, carrying forward essential state: workflow.continue_as_new(MonitorConfig(...)). This prevents unbounded history growth in workflows designed to run indefinitely — a crucial pattern for production-grade Temporal applications that many developers overlook until they encounter history size limits.
View in source ↗

---

---

ZiggyBot is an AI pre-screener based on Temporal's community mascot Ziggy. Final decisions are made by the community team.

Reviewer: Check the box below to generate a Contentful-ready long description for this submission.

• Generate long description if needed

[github-actions]

Hi, I'm ZiggyBot! 🤖 Here's my pre-evaluation of this submission:

---

Criterion	Assessment
Useful to Temporal users	✅ Yes — Demonstrates a practical integration of Temporal with Pydantic AI's TemporalAgent for durable AI pipelines, showing how to wrap LLM calls and tool invocations as retryable activities.
Benefits clearly explained	✅ Yes — The README clearly explains the architecture, how Pydantic AI agent calls become durable Temporal activities, and includes an excellent ASCII diagram showing the workflow structure.
OSI Approved License	❌ No — No license file was detected by GitHub, which means the project is technically "all rights reserved" and cannot be safely used or shared.
README with working instructions	✅ Yes — Comprehensive README with quick start (./run.sh), manual setup instructions, CLI commands for all operations, and clear documentation of test photos and output logging.

Overall: Needs review ⚠️

Notes:

• This is a really creative real-world application that showcases the Pydantic AI + Temporal integration beautifully! The use case (terrace monitoring with autonomous actions) is fun and memorable.
• The workflow demonstrates several excellent Temporal patterns: signals for pause/resume, queries for status, configurable intervals, and continue-as-new for unbounded execution.
• The PydanticAIWorkflow base class and __pydantic_ai_agents__ registration pattern is a nice example of the emerging Pydantic AI integration.
• The project is macOS-specific (uses afplay and osascript), which limits its audience but is clearly documented.
• The missing license is the only blocker — this is an easy fix!

Suggested questions for submitter:

• Could you add an MIT (or other OSI-approved) license file to the repository? This is required for inclusion in the Code Exchange.
• Would you consider adding a note about how the Pydantic AI TemporalAgent integration works under the hood for users who want to understand the durability guarantees?

Teaching moment:

The PydanticAIWorkflow base class pattern is a fantastic example of making AI agent calls durable. By declaring __pydantic_ai_agents__ = [temporal_agent] and extending PydanticAIWorkflow, the LLM calls and tool invocations automatically become Temporal activities — meaning they're retryable, observable in the Temporal UI, and their results are persisted in the workflow history. This is a powerful pattern for anyone building AI applications that need reliability guarantees.
View in source ↗

---

---

ZiggyBot is an AI pre-screener based on Temporal's community mascot Ziggy. Final decisions are made by the community team.

Reviewer: Check the box below to generate a Contentful-ready long description for this submission.

• Generate long description if needed

[github-actions]

Hi, I'm ZiggyBot! 🤖 Here's my pre-evaluation of this submission:

---

Criterion	Assessment
Useful to Temporal users	✅ Yes — Demonstrates a practical integration of Temporal's durable execution with Pydantic AI's agent framework for multimodal AI workflows, which is an emerging and valuable pattern.
Benefits clearly explained	✅ Yes — The README clearly explains the architecture, use cases (bird detection, plant monitoring, hazard alerts), and how Temporal provides durability and observability for AI pipelines.
OSI Approved License	❌ No — No license file was detected in the repository, which is required for inclusion in the Code Exchange.
README with working instructions	✅ Yes — Comprehensive README with quick start (./run.sh), manual setup options, CLI commands, test photos, and clear project structure documentation.

Overall: Needs review ⚠️

Notes:

• This is a creative and well-documented project that showcases a real-world application of Temporal + Pydantic AI for autonomous agent workflows. The architecture diagram and test photo scenarios make it easy to understand.
• The use of PydanticAIWorkflow base class and TemporalAgent wrapper to make LLM calls durable is a valuable pattern for the community.
• The project includes thoughtful production considerations: continue-as-new for history management, pause/resume signals, configurable intervals, and structured JSONL logging.
• macOS-specific dependencies (afplay, osascript) limit portability, but this is acknowledged and appropriate for a demo project.
• The missing license is the only blocking issue — everything else looks great!

Suggested questions for submitter:

• Could you add an OSI-approved license (MIT is recommended) to the repository? This is required for Code Exchange inclusion.
• Would you consider adding a note about how to swap in other LLM providers (OpenAI, Gemini) since you mention this flexibility?

Teaching moment:

This project beautifully demonstrates the continue-as-new pattern to manage workflow history size in long-running monitoring workflows. The workflow tracks cycles and automatically continues-as-new after a threshold, preserving state while keeping the event history bounded. This is essential for workflows that run indefinitely — without it, the history would grow unbounded and eventually cause performance issues.
View in source ↗

---

---

ZiggyBot is an AI pre-screener based on Temporal's community mascot Ziggy. Final decisions are made by the community team.

Reviewer: Check the box below to generate a Contentful-ready long description for this submission.

• Generate long description if needed