Feature Request: Support Dynamic Langfuse Project Selection per Agent/Completion

[marcjaner]

Describe the feature or potential improvement

Currently, the Langfuse Java SDK and Spring AI integration only support a single Langfuse project per application instance, configured via global API keys (LANGFUSE_PUBLIC_KEY / LANGFUSE_SECRET_KEY).

In applications with multiple AI agents or completion types, there's no way to dynamically route traces to different Langfuse projects based on the agent or completion context.

Use Case

I'm building a Spring Boot application with Spring AI that has multiple AI agents, each serving different purposes (e.g., customer support agent, analytics agent, content generation agent). For organizational and access control reasons, I want traces from each agent type to go to separate Langfuse projects.

Current workaround limitations:

• Using tags/metadata requires all teams to access the same project
• Running multiple app instances with different configs is operationally expensive
• Environment-based separation doesn't allow runtime routing

Proposed Solution

Allow dynamic project selection at the observation/span level. Possible approaches:

Option 1: Context-based API Keys

// Set project context before making AI calls
LangfuseContext.setProject(publicKey, secretKey);
try {
    chatClient.call(prompt); // Traces go to specified project
} finally {
    LangfuseContext.clear();
}

Option 2: Observation Attributes

Observation.createNotStarted("chat.completion", registry)
    .lowCardinalityKeyValue("langfuse.project.id", "project-123")
    .observe(() -> chatClient.call(prompt));

Option 3: Configuration via Spring AI

ChatClient.builder()
    .observationConvention(new CustomConvention())
    .langfuseProject("customer-support-project")
    .build();

Additional information

Environment

• Langfuse Java SDK: 0.1.1
• Spring AI: 1.1.0-M2
• Spring Boot: 3.5.5
• Micrometer: 1.15.3

[jannikmaierhoefer]

Thanks for adding this @marcjaner!