v0.3.1

Changelog

v0.3.1

• feat: new way to provide messages endpoints and sse/streamable endpoints in MCP servers implementations to help cases where these endpoints are dynamically updated and may vary (e.g providing user_ids in endpoints)
• fix: updating trace parameters instead of replacing them in clone method.

v0.3.0

Changelog

v0.3.0

• feat: new before_call and after_call callbacks to help users provide hooks before and after tool calls (used for HITL as well)
• feat: AgentTeam and AgentPipeline instances can now be converted into ASGI APIs
• feat: Enhanced Context objects to provide clear, resumable and easy way to provide/retrieve/resume Agent calls.
• feat: Enhanced Step objects to provide clear, resumable and easy way to provide/retrieve/resume Agent calls.
• feat: Tool calls and structured outputs in CerebrasGenerationProvider
• feat: new ModelKind type variable to help users to obtain provider-specific models for specific categories, like "standard", "nano", "mini", etc. Useful when using FailoverGenerationProvider since the model you might use in one might not be present in the other, but you want the same kind of model for each provider you are using.
• feat: new timeout_m, timeout_s to generation configuration
• fix: wrong pricing calculations in GoogleGenerationProvider and CerebrasGenerationProvider classes.
• fix: prices not showing up properly in Langfuse view.

v0.2.12

Changelog

v0.2.12

• fix: wrong llama-3.370b pricing on cerebras

v0.2.11

Changelog

v0.2.11

• refactor: changing cerebras standart to llama3.3-70b

v0.2.10

Changelog

v0.2.10

• fix: not propagating tracing_client when present in any of providers passed to add method.

v0.2.9

Changelog

v0.2.9

• feat: new __add__ magic method added to GenerationProvider class

v0.2.8

Changelog

v0.2.8

• fix: wrong pricing mapping when a type of ModelKind is provided to providers

v0.2.7

Changelog

v0.2.7

• feat: structured outputs for cerebras

v0.2.6

Changelog

v0.2.6

• feat: new way to use models of the same kind with an interface compatible with all providers (ModelKind TypeAlias)
• fix: wrong input/output token calculation in GoogleGenerationProvider
• feat: tool support in OpenAI and Cerebras

v0.2.5

Changelog

v0.2.5

• feat: new timeout_s and timeout_m parameter in GenerationConfig class to provide an easier way to declare timeouts.
• feat: new tool calling capabilities in cerebras provider