lexigram-ai-agents

Agent system for Lexigram Framework - AI agents with tools, strategies, and execution

Overview

Agent orchestration package for the Lexigram Framework. Provides agent base classes, tool registration, execution strategies (ReAct, Plan-and-Execute, Reflexion, Supervisor), observability, and multi-agent coordination — all wired through DI via AgentsModule. Zero-config usage starts with sensible defaults.

Full documentation: docs.lexigram.dev

Install

uv add lexigram-ai-agents

Quick Start

from lexigram import Application
from lexigram.di.module import Module, module

from lexigram.ai.agents import AgentsModule
from lexigram.ai.agents.config import AgentConfig
from lexigram.ai.llm import LLMModule


@module(
    imports=[
        LLMModule.stub(),  # provides LLMClientProtocol
        AgentsModule.configure(AgentConfig(max_iterations=10)),
    ]
)
class AppModule(Module):
    pass


async with Application.boot(modules=[AppModule]) as app:
    # use app.container to resolve services
    ...

Note: AgentsModule requires an LLM client (LLMClientProtocol) in the container — provided by LLMModule (from lexigram-ai-llm). When using a real LLM provider, set OPENAI_API_KEY (or configure your provider in ClientConfig).

Configuration

Zero-config usage: Call AgentsModule.configure() with no arguments to use defaults.

Option 1 — YAML file

# application.yaml
ai_agents:
  max_iterations: 10
  default_temperature: 0.7
  default_max_tokens: 2048
  enable_tracing: true

Option 2 — Profiles + Environment Variables (recommended)

export LEX_AI_AGENTS__MAX_ITERATIONS=15
# Environment variables for each field

Option 3 — Python

from lexigram.ai.agents.config import AgentConfig
from lexigram.ai.agents import AgentsModule

config = AgentConfig(max_iterations=10)
AgentsModule.configure(config)

Config reference

Module Factory Methods

Key Features

• Agent base classes: AgentBase for defining agents with tools and system prompts
• Execution strategies: ReAct, Plan-and-Execute, Reflexion, Supervisor
• Tool system: @tool decorator for registering standalone tool functions
• Multi-agent coordination: AgentAsToolAdapter for agent-to-agent delegation
• Observability: Built-in tracing and metrics via AgentTracer and AgentMetrics

Testing

from lexigram.ai.llm import LLMModule

async with Application.boot(modules=[LLMModule.stub(), AgentsModule.stub()]) as app:
    # your test code
    ...

AgentsModule.stub() alone fails container validation without an LLMClientProtocol in the container — pair it with LLMModule.stub() as shown.

Key Source Files