Llm4s

A builder for an Anthropic client: apiKey and model are required (#1489).

A builder for an Azure OpenAI client: endpoint, apiKey and model (the deployment) are required (#1489).

Wraps a JLlmClient in a JAgent ready to accept natural-language queries, with no tools.

Wraps a JLlmClient in a JAgent that can call tools. Tools belong to the agent: an agent whose tools cannot be offered together (a name clash, an invalid schema) fails every run with the reason.

createAgent with tools, whose model calls stream when streaming is set: then a JAgent.stream of it carries the answer's text deltas (AgentEvents.TextDelta()). Without it, the agent calls the model's complete, and a stream carries no deltas.

Wraps a JLlmClient in a JAgent that can call tools, defined from Java with JTool.builder or JTool.fromAnnotated (#1484). A null argument, a tool listed twice, two tools with one name, or tools the agent cannot offer together is a failed result, not an agent that fails every run.

JAgent agent = Llm4s.createAgentWithTools(client, List.of(weather)).get();

Creates a JLlmClient from an explicit ProviderConfig. Useful when the caller constructs the config programmatically rather than relying on environment variables.

createClient whose client reports every call (request outcome, duration, tokens and cost) to metrics, for example a MicrometerMetrics from llm4s-observability-micrometer over the application's MeterRegistry, or a PrometheusMetrics from llm4s-observability-prometheus. The Spring Boot starter uses it to attach the application's MetricsCollector bean. A null metrics records nothing; a null config is a failed result.

Creates a JLlmClient from the default provider: the section of application.conf that llm4s.providers.provider names. A provider's API key comes from its section's apiKey, else from llm4s.credentials (bound to the vendor's usual environment variable, e.g. OPENAI_API_KEY).

Creates a JEmbeddingClient for the embedding model configured as chat providers are, in the application's application.conf: llm4s.embeddings.model, provider/model (for example openai/text-embedding-3-small), which EMBEDDING_MODEL sets; the provider's key comes from its llm4s.embeddings.<provider> block or its vendor variable, such as OPENAI_API_KEY.

Does not throw and sends no request: no model configured, an embedding provider that is not on the classpath or a missing key is a failed LlmResult of kind CONFIGURATION, whose message says what to set. A model its provider module does not list, such as an Ollama model you pulled yourself, is created: its dimensions() is empty, and its vectors are as long as the provider makes them.

Creates a JEmbeddingClient for model, provider/model (for example openai/text-embedding-3-large or ollama/bge-m3), rather than the model llm4s.embeddings.model names - which need not be set. Everything else comes from configuration as for createDefaultEmbeddingClient: the provider's key from its llm4s.embeddings.<provider> block or its vendor variable, such as OPENAI_API_KEY, and its base URL and timeouts from that block.

Does not throw and sends no request: a null model is a failed result of kind VALIDATION; a model not in the form provider/model, an embedding provider that is not on the classpath, or a missing key is one of kind CONFIGURATION, whose message says what to set. A model its provider module does not list is created, as for createDefaultEmbeddingClient.

A builder for a client of the Google Gemini API: apiKey and model are required (#1489).

A builder for an Ollama client: model is required, and the server is http://localhost:11434 unless set (#1489).

A builder for an OpenAI client: apiKey and model are required (#1489).

A builder for a client of any server that speaks OpenAI's chat-completions API - vLLM, LM Studio, LiteLLM, a gateway: baseUrl and model are required, the API key optional (#1489).

Wraps an Agent built with Agent.builder, for what the factories above do not set: its middleware (guardrails, approvals), handoffs, system prompt or runtime.

Wraps a Scala LLMClient you built yourself - a ReliableClient around a provider's client, your own implementation, or a test double - in a JLlmClient. Closing the wrapper closes client.

Wraps a Scala EmbeddingClient you built yourself - over your own EmbeddingProvider, or a test double - in a JEmbeddingClient that asks for model and reports its dimensions.