llm¶

Multi-provider LLM abstraction via OmniLLM.

Model Factory¶

import "github.com/plexusone/agentkit/llm"

cfg := config.LoadConfig()

factory := llm.NewModelFactory(cfg)
model, err := factory.CreateModel(ctx)
if err != nil {
    log.Fatal(err)
}
defer factory.Close()

Supported Providers¶

Provider	Environment Variables
Gemini	`GEMINI_API_KEY`, `LLM_MODEL`
Claude	`CLAUDE_API_KEY`, `LLM_MODEL`
OpenAI	`OPENAI_API_KEY`, `LLM_MODEL`
xAI	`XAI_API_KEY`, `LLM_MODEL`
Ollama	`OLLAMA_URL`, `LLM_MODEL`

Provider Selection¶

Set via environment:

export LLM_PROVIDER=gemini
export GEMINI_API_KEY=your-key
export LLM_MODEL=gemini-2.0-flash-exp

Or in code:

cfg := config.LoadConfig()
cfg.LLMProvider = "claude"
cfg.LLMModel = "claude-3-5-sonnet-20241022"

factory := llm.NewModelFactory(cfg)

Model Interface¶

The factory returns an OmniLLM model that implements a standard interface:

model, _ := factory.CreateModel(ctx)

// Generate text
response, err := model.Generate(ctx, prompt)

// Chat completion
messages := []Message{
    {Role: "user", Content: "Hello"},
}
response, err := model.Chat(ctx, messages)

Observability¶

Enable LLM observability with OmniObserve:

export OBSERVABILITY_ENABLED=true
export OBSERVABILITY_PROVIDER=opik  # or langfuse, phoenix

cfg := config.LoadConfig()
// Observability is automatically configured
factory := llm.NewModelFactory(cfg)

Adapters¶

The llm/adapters package provides adapters for specific frameworks:

import "github.com/plexusone/agentkit/llm/adapters"

// Eino adapter
einoModel := adapters.NewEinoAdapter(model)

// ADK adapter
adkModel := adapters.NewADKAdapter(model)

Usage with BaseAgent¶

The BaseAgent automatically creates and manages the LLM:

ba, _ := agent.NewBaseAgent(cfg, "my-agent", 30)

// LLM is available via base agent
response, err := ba.Generate(ctx, prompt)