Track Openai Assistant API costs

[christos-bsq]

Describe the feature or potential improvement

Idea

Monitor costs for tool usage within Langfuse,

Case: OpenAI's Assistants API

Besides, I/O tokens, OpenAI can also charge you for tool usage via the Assistants API:

Tool	Cost
Code Intepreter	$0.03 / session
File Search	$0.10 / GB of vector-storage per day (1 GB free)

source: https://openai.com/api/pricing/

Langfuse changes

• Support "tool usage cost monitoring" (besides token cost).
  • you can have "tool" and "cost per tool" per trace/span/generation

Then specifically for Assistants API, use the OpenAI response to calculate cost (e.g. if OpenAI's response has Code Interpreter in the tool usage, add the flat $0.03 as cost)

Other cases

A developer might also use other tools and APIs (e.g. a Google Search API) when building agents.

Using Langfuse as the central hub for cost monitoring would be nice.

Additional information

No response

[richardchen874-sys]

Assistant-style APIs can hide cost across threads, tool calls, retrieval, and retries. I would attach cost to both the provider call and the logical assistant run, otherwise the dashboard can be technically correct but not very actionable. I am testing an OpenAI-compatible multi-model API layer, and assistant/workflow-level cost tracking is one of the areas I am working on.