Weave Code
Llphant Laravel Package
theodo-group/llphant
LLPhant is a PHP generative AI framework for building LLM-powered apps with OpenAI, Anthropic, Mistral, and Ollama. It provides chat, embeddings, vector stores, RAG, agents, and document loaders, with straightforward integration into Laravel and plain PHP projects.
LLPhant - A comprehensive PHP Generative AI Framework using OpenAI GPT 4. Inspired by Langchain
Frequently asked questions about Llphant
- How do I install LLPhant in a Laravel project?
- Install it with Composer using composer require theodo-group/llphant, pinning to ^1.0.7 if you want the latest patch. It runs on PHP 8.x and fits behind Laravel's service container and config publishing, so you can wire provider credentials through your .env file. No schema changes, migrations, or new infrastructure are required. Resolve it in a sandbox branch first to confirm PHP and dependency constraints against your existing lockfile.
- Which LLM providers does LLPhant support out of the box?
- LLPhant ships clients for OpenAI, Anthropic, Mistral, and Ollama, plus OpenAI-compatible endpoints. Provider choice is configuration rather than architecture, which is why new backends arrive as a config class instead of a forked client. If you do not need a given provider, its config is inert and adds no runtime cost. Abstract the library behind a thin interface of your own so swapping providers never touches call sites.
- What is HeabsyConfig in LLPhant 1.0.7, and do I need it?
- HeabsyConfig is a provider configuration class for the OpenAI-compatible Heabsy API, added in release 1.0.7 via contributor PR #520. It only matters if Heabsy is a required backend for one of your roadmap items; otherwise the release is informational for you. The architecturally interesting part is that a new provider landed as a config class rather than a breaking refactor, which confirms the provider abstraction is a stable extension point. Adopt the patch to stay current, but feature-flag the provider until you have exercised it.
- Is upgrading LLPhant from 1.0.6 to 1.0.7 a breaking change?
- No. The delta is a single additive config class and touches nothing existing, so existing OpenAI, Anthropic, Mistral, and Ollama paths cannot regress by inspection. There are no schema changes, no data migration, and no config rewrites. Run your existing test suite in a sandbox branch to confirm dependency resolution and that the new config is discoverable through the library's provider-selection mechanism.
- Is LLPhant a Laravel package or a framework-agnostic PHP library?
- LLPhant is a PHP LLM-orchestration library with clients, embeddings, vector stores, RAG pipelines, and agents, inspired by Langchain. It is not Laravel-first, but it slots cleanly behind Laravel service classes and config rather than competing with Laravel primitives. In practice that means queue workers such as Horizon for ingestion and embedding jobs, Guzzle for HTTP, and Eloquent or a dedicated vector store for retrieval state. Straightforward integration is possible in both Laravel and plain PHP projects.
- How do I build a RAG pipeline with LLPhant in Laravel?
- LLPhant gives you document loaders, embeddings, vector stores, and RAG pipelines as composable pieces. In Laravel, push ingestion and embedding work onto queued jobs so chat requests stay fast, and keep retrieval state in a dedicated vector store or Eloquent. Wrap the pipeline behind your own service class so the orchestration layer stays swappable. Verify the retrieval path with a contract test that asserts the returned context shape, not just that the call succeeded.
- Can I point LLPhant at a custom or self-hosted OpenAI-compatible endpoint?
- The provider abstraction is a confirmed extension point, and HeabsyConfig is a working example of registering an OpenAI-compatible backend through configuration alone. Whether base-URL override and standard auth injection are exposed for your endpoint should be verified against the config class before you commit to it. If you already maintain a hand-rolled OpenAI-compatible config, compare behavior first and only replace it with the upstream version when the request and response handling matches. This avoids accumulating two parallel configuration paths.
- How should I test LLPhant provider integrations in CI?
- Treat OpenAI compatibility as a dialect rather than a contract and write contract tests that assert response schema and token accounting, not just HTTP 200. Exercise the provider with a non-production key behind a feature flag, and canary a single low-risk workload before broader rollout. Keep those tests isolated from your unit suite so provider latency and drift do not make CI flaky. Running your existing suite against the patch release in a sandbox branch covers the regression check cheaply.
- Should I pin the LLPhant version in composer.json?
- Yes. Patch-level releases arriving with modest ceremony, such as 1.0.6 to 1.0.7, suggest a fast-moving upstream, so pinning is advisable. Constrain to ^1.0.7 or hold at your current pin if you do not need the new provider, and document an explicit upgrade policy. Because the library is a dependency posture question rather than just a code question, review each release for provider and supply-chain implications, not only changelog features.
- What are the risks of adding an OpenAI-compatible provider like Heabsy to production?
- Technical risk for the release itself is low because the change is additive, but provider risk is separate and unverified. Heabsy is a low-visibility endpoint, and its operational maturity, SLA, data-residency posture, and DPA coverage cannot be confirmed from release notes alone. OpenAI compatibility does not guarantee identical response schemas or token accounting, so cover both with contract tests on your side. Clear it through procurement and security before any production traffic reaches it.
Weaver
How can I help you explore Laravel packages today?