Enabling async agent creation and using async agent invocation

[uesleilima]

PR: Major Refactor – Native Async APIs, Dependency Upgrades, and Adapter Modernization

Overview

Implementation of Issue #45

The main driver for this change is to support native async/await throughout the API, greatly increasing flexibility, performance, and alignment with the latest patterns in the Langchain and FastAPI ecosystems.

---

Key Changes

1. Full Async Support for Agent Lifecycle and APIs

• Introduced async-first APIs for agent lifecycle (acreate_agent, ainvoke, etc.), FastAPI routers, and key integration points.
• Every FastAPI endpoint & router now natively supports async execution.
• Agent factories, adapters, and model bridges leverage async execution, making it easier to work with streaming, real-time LLMs, and scalable deployments.

2. Dependency and Model Adapter Modernization

• Upgraded dependencies to the latest major and minor versions for strong compatibility with modern LLM/client packages.
  • New minimums: langchain, openai, langchain-openai/antrhopic, fastapi etc.
• Refactored model adapters for LlamaCpp, and more, using new import paths and proper async support.

3. Consistent Prompt Handling and Factory APIs

• Updated react agent construction to use new parameter (from messages_modifier to prompt)

4. Streaming and Real-Time Features

• Streaming support has been improved, including async iterators and SSE/streaming endpoint compatibility, which is essential for modern chat and LLM multi-turn applications.

5. Dependency Pinning & Poetry Lock Refresh

• All dependencies in pyproject.toml and poetry.lock were reviewed and updated to ensure maximum compatibility with async frameworks and modern LLM model packages.
• Superfluous or outdated development packages have been cleaned up, contributing to lighter and more reliable CI/CD and deployment.

6. Example Factories, Test Suites, and Documentation Updates

• All examples and test agent factories have been updated for async signatures and new model/adapters.
• In-code and README docs now detail how to leverage async and streaming patterns in downstream/codegen integrations.

---

Motivation

• Adopt async-first patterns: Allows this backend and its users to work efficiently with concurrent, streaming, or long-running LLM tasks—a must for modern API usage and real-time AI applications.
• Modernize for the latest Langchain/OpenAI/Anthropic/LLM adapters: Ensures out-of-the-box compatibility with the fastest-moving open-source LLM landscape.
• Improve maintainability: Clearer separation of concerns, more explicit factory/adapter/subclass patterns, and tighter test/CI ensure future-proofing and easier onboarding for new contributors.

---

Migration/Breaking Changes

• All agent and API lifecycle hooks are now async. Downstream consumers must use await and async FastAPI endpoints when customizing.
• Prompt & agent factory signatures are now consistently async; see new examples for idiomatic usage.
• Sync/legacy interfaces are not supported in this version. Migration is required for downstream consumers.

---

Compatibility

• This PR is a major, breaking upgrade for consumers of the former synchronous-only codebase.
• Strongly recommended for new projects, or those planning to run with modern Langchain/FastAPI/LLM stacks and requiring scalable, async/streaming support.

---

Thank you for considering this major upgrade—this PR sets up the project and its users for scalable, concurrent, and forward-looking LLM API deployments!

[benjaminvdb]

Great job! It would be wonderful to see this merged ❤️

[caffeinism]

This is a much needed feature for me.

[samuelint]

@uesleilima please fix lint issues before I merge this feature.

[samuelint]

This PR is a duplicate of #52, which includes a few additional changes, so I’ll be merging that one instead.
@uesleilima — I believe @caffeinism built upon your work from this PR, so your contributions are definitely not lost 😊 Thanks a lot for your work!