Interchange Model

This cookbook demonstrates switching between different model providers within a single agent session while preserving tool call history.

What it tests

An agent with add_history_to_context=True uses tools across multiple turns, switching models between turns. The history (including tool calls and results) must be correctly formatted for each provider.

Providers covered

OpenAI Chat Completions (OpenAIChat)
OpenAI Responses (OpenAIResponses)
Anthropic Claude (Claude)
Google Gemini (Gemini)
AWS Claude (via agno.models.aws.Claude)

Scripts

Script	Description
`openai_claude.py`	Alternates between OpenAI Chat and Claude with tool calls
`openai_chat_responses.py`	Alternates between OpenAI Chat and OpenAI Responses with tool calls
`openai_gemini.py`	Alternates between OpenAI Chat and Gemini with tool calls
`claude_gemini.py`	Alternates between Claude and Gemini with tool calls
`all_providers.py`	Cycles through OpenAI Chat, OpenAI Responses, Claude, Gemini, and AWS Claude with tool calls

Prerequisites

PostgreSQL + pgvector running (for session persistence): ./cookbook/scripts/run_pgvector.sh
API keys set: OPENAI_API_KEY, ANTHROPIC_API_KEY, GOOGLE_API_KEY
Optional: AGNO_POSTGRES_URL (defaults to postgresql+psycopg://ai:ai@localhost:5532/ai)

Running

bash

# Start the database
./cookbook/scripts/run_pgvector.sh

# Run one of the interchange scripts
.venvs/demo/bin/python cookbook/02_agents/14_advanced/interchange_model/openai_claude.py

# Or run the full provider cycle
.venvs/demo/bin/python cookbook/02_agents/14_advanced/interchange_model/all_providers.py