Weave Code
Agents Laravel Package
llm-agents/agents
A Laravel package for building LLM-powered agents with tool calling, prompt management, and structured multi-step workflows. Integrates AI agents into PHP apps through a clean, framework-native API for automating tasks and composing agent pipelines.
llm-agents/agents is a Laravel package for building and orchestrating LLM-powered agents inside your application. It provides a clean, fluent API to define agents, connect tools, and run multi-step conversations without boilerplate.
- Agent definitions via fluent, class-based configuration
- Tool calling that lets agents invoke your PHP methods and services
- Conversation memory for stateful, multi-turn interactions
- Laravel-native integration with config, facades, and the service container
- Provider-agnostic support for multiple LLM backends
Frequently asked questions about Agents
- How do I install llm-agents/agents in a Laravel application?
- Install it with Composer using composer require llm-agents/agents, then pin the exact version in your composer.json. The package is a PHP SDK rather than a guaranteed first-class Laravel package, so expect to register a thin service provider and add a configuration file for provider credentials. Validate the dependency tree against your current Laravel constraints before adopting it.
- Which PHP and Laravel versions does llm-agents/agents support?
- Check the package's composer.json constraints and its test matrix rather than assuming support, because the last release was 2024-09-15 and adoption signals are weak. Run a Composer dependency analysis and a security scan against the exact PHP and Laravel versions you use. If the constraints conflict with your stack, evaluate a fork only if the use case is strategically important and someone owns it.
- Can I run LLM agents asynchronously using Laravel queues?
- Agent runs are iterative and potentially long-running, so enqueue them as Laravel queue jobs and let HTTP controllers only dispatch work and return a run status. The package is described as having queue and event integrations, but confirm timeout, retry, and cancellation behavior before relying on it. Use dedicated queues and budget worker capacity so agent traffic does not starve other jobs.
- Which LLM providers does llm-agents/agents work with?
- The package is described as an LLM provider abstraction with support for OpenAI, Anthropic, and other providers. Verify the current provider list and supported tool-calling patterns in the repository, and confirm they match the providers your organization has approved. Provider API changes are an ongoing risk, so keep provider configuration in application-owned code rather than leaking package types into domain services.
- Is llm-agents/agents actively maintained and production ready?
- Treat it cautiously: it has roughly 169 stars, zero dependents, a last release dated 2024-09-15, and a high technical risk rating. That does not make it unusable, but it does mean you should isolate it behind an internal adapter and run a time-boxed proof of concept instead of making it a platform dependency. Assign clear ownership for security patches and dependency updates.
- How do I test agents built with llm-agents/agents in Laravel?
- The blurb advertises Laravel-native testing integrations, but confirm what the package's own test suite covers before depending on it. In your tests, fake provider responses so agent runs are deterministic, then assert the expected tool calls, prompts, and outputs. Keeping prompts and tool definitions in application-owned structures makes those tests portable if you later swap the implementation.
- How do I define tools and tool calling for an agent?
- Agents are defined through a fluent PHP API where you register tools, prompts, and conversation flow. Keep tool definitions in your own application code and validate every tool input, because the model decides when and how to call them. Also enforce an allowlist and a maximum iteration count so a misbehaving model cannot chain tool calls indefinitely.
- How do I prevent runaway agent loops and unexpected LLM cost overruns?
- Enforce max iterations, token budgets, per-run timeouts, and queue-level concurrency limits around every agent run. Check whether the package exposes token usage, cost, and latency data so you can log and alert on them; if it does not, capture that in your adapter. A parallel pilot against direct provider calls helps you compare cost and reliability before scaling.
- What are the alternatives to llm-agents/agents for Laravel projects?
- Common alternatives are calling the official OpenAI or Anthropic SDKs directly behind your own AgentRunner interface, or adopting another orchestration layer. Because this package has weak adoption and a stale release, a thin adapter with one implementation backed by llm-agents/agents keeps migration cheap. You can then swap in direct provider calls without rewriting product code.
- How should I handle API keys and prompt injection risk with LLM agents?
- Store provider API keys in environment variables or a secret manager through Laravel configuration, never in code or version control. Treat all model output as untrusted input: restrict which tools can run, scope credentials per tool, and guard against prompt injection that tries to trigger unintended actions. Log agent runs through Laravel events and logging for auditability, and avoid sending sensitive data to providers you have not approved.
Weaver
How can I help you explore Laravel packages today?