Sagents provides a robust Elixir framework for building interactive AI agents, integrating LLM capabilities with Elixir's OTP for production-ready applications. It targets Elixir developers needing real-time user interaction, human oversight, and scalable agent orchestration, offering a powerful alternative to simpler LangChain integrations for complex, interactive systems.

How It Works

The system leverages Elixir's OTP GenServer and supervision trees for fault-tolerant agent processes. Agents are constructed as explicit, composable Elixir pipelines (execution modes) built from reusable steps, integrating LangChain for LLM calls. A sophisticated middleware system allows for modular extension of agent capabilities, while features like Human-In-The-Loop (HITL) and SubAgents enhance control and delegation. Phoenix integration provides real-time UI updates and smart resource management.

Quick Start & Requirements

• Installation: Add {:sagents, "~> 0.9.0"} to mix.exs.
• Prerequisites: Elixir/OTP, LLM API keys (Anthropic, OpenAI, Google) configured as environment variables, Phoenix.PubSub, Phoenix.Presence. Optional: Horde for distributed deployments.
• Setup: Integrate Sagents.Supervisor into your application's supervision tree. Configure LLM providers and models.
• Resources: See agents_demo for an interactive example and sagents_live_debugger for debugging tools. LangChain documentation provides detailed LLM configuration.

Highlighted Details

• Human-In-The-Loop (HITL): Granular control over sensitive operations with customizable approval workflows for individual tool calls.
• Composable Execution Modes: Define agent run loops as explicit Elixir pipelines, mixing built-in or custom steps.
• SubAgents: Enable delegation of complex tasks to specialized child agents for efficient context management and parallel execution.
• Extensible Middleware: A plugin architecture for adding features like task management (TodoList), virtual file systems (FileSystem), summarization, and observability.
• Real-time Integration: Seamless integration with Phoenix LiveView via PubSub for streaming events and Presence for intelligent resource management.
• Cluster-Aware Distribution: Optional Horde integration for distributed agent execution across a cluster with state migration.
• Structured Completion: until_tool mode enforces structured output by looping until a specific tool is called.
• Virtual Filesystem: Provides isolated, in-memory file operations with optional persistence.

Maintenance & Community

No specific details on maintainers, community channels (e.g., Discord, Slack), or roadmap were found in the provided README.

Licensing & Compatibility

• License: Apache-2.0.
• Compatibility: Permissive license suitable for commercial use and integration into closed-source applications.

Limitations & Caveats

Requires significant Elixir/OTP development expertise. Relies on external LLM providers, necessitating API key management and incurring usage costs. Distributed deployments require understanding and configuring Horde.