Conductor
freeLaravelConductor is the orchestration layer for building AI features in Laravel. It gives you a clean, expressive API for working with agents, multi-step workflows, RAG pipelines, and tool calling, all without having to wire up the boilerplate yourself.
You bring your own API keys. Conductor handles the rest. Under the hood, all LLM calls go through Prism, so any provider Prism supports (Anthropic, OpenAI, Ollama, Mistral, Groq, Gemini, and more) works out of the box.
Agents can be defined inline with a fluent builder or as standalone classes when you need something reusable. They support streaming, structured output via JSON schemas, conversation memory across requests, automatic tool calling, and fallback providers if your primary goes down. Token usage, cost, and duration are tracked automatically on every call.
Workflows let you chain multiple steps together with dependency resolution, conditional execution, parallel step groups, configurable retries with backoff, and human approval gates that pause execution and persist state until someone signs off. This makes it straightforward to build things like content pipelines, review flows, or any multi-agent process where steps need to happen in a specific order.
The RAG system covers document ingestion, chunking with configurable size and overlap, embedding generation, and similarity-based retrieval. It plugs directly into agents so you can build context-aware bots without managing the retrieval layer separately. Supports an in-memory store for development and pgvector for production.
Monitoring is built in from the start. Every agent call gets logged with token counts, cost breakdowns, and timing. You can set spending caps per request, per workflow, or per hour, and Conductor will throw an exception and fire an event when a budget is hit. If you run Laravel Pulse, it registers a dashboard card automatically.
Testing gets first-class treatment too. Conductor::fake() works just like Laravel's other fakes. You can stub responses per agent, use sequences for agents called multiple times, and assert on agent calls, tool usage, token counts, and workflow completion.
Requires PHP 8.2+ and Laravel 11 or 12.