Laragents: PHP Package for AI Agents in Laravel

Making a call to an artificial intelligence model via an API is straightforward when simply sending a request and receiving a JSON response. However, the real engineering challenge lies in everything surrounding it: managing execution loops, maintaining conversational context across sessions, storing long-term memory, enforcing strict behavioral rules, and triggering agents based on application events.

Software developer Eduard Lazaro created an open-source package named Laragents to address these exact complexities after building similar systems across multiple applications. The package can be easily installed via Composer, publishing necessary configuration files and database migrations with simple artisan commands.

Stock photo for illustration only, not from the actual event

At the core of Laragents is the orchestration of the loop between AI models and application tools. Since models typically ask for tools, evaluate results, and iterate before generating final prose, the package introduces two crucial safety guards:

• The iteration cap: Prevents models from getting stuck in endless search loops that waste time and incur unnecessary API costs.
• The empty-streak guard: Detects when consecutive search results yield nothing, automatically prompting the model to acknowledge the empty state rather than repeating failed queries.

When handling tool parameters, Laragents separates inputs into two distinct trust levels: $args generated by the language model, and $context securely provided by the application environment, ensuring identity and scope are never compromised by user inputs.