Model Setup¶

Promptise supports any LLM provider that LangChain integrates with. Models are specified as a string in the format "provider:model-name".

Provider String Format¶

agent = await build_agent(
    model="openai:gpt-5-mini",   # provider:model-name
    servers=...,
)

Supported Providers¶

Provider	Format	Example	Env Variable
OpenAI	`openai:model`	`openai:gpt-5-mini`	`OPENAI_API_KEY`
Anthropic	`anthropic:model`	`anthropic:claude-sonnet-4`	`ANTHROPIC_API_KEY`
Google	`google:model`	`google:gemini-2.5-pro`	`GOOGLE_API_KEY`
Ollama	`ollama:model`	`ollama:llama3`	(local, no key needed)

Set the appropriate environment variable for your provider:

# OpenAI
export OPENAI_API_KEY=sk-...

# Anthropic
export ANTHROPIC_API_KEY=sk-ant-...

# Google
export GOOGLE_API_KEY=AIza...

The `ModelLike` Type¶

The model parameter accepts three types:

1. Provider String (recommended)¶

The simplest option. Promptise uses LangChain's init_chat_model to resolve the string:

agent = await build_agent(
    model="openai:gpt-5-mini",
    servers=my_servers,
)

2. LangChain Chat Model Instance¶

Pass a pre-configured BaseChatModel for full control over parameters like temperature, max tokens, and base URL:

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    model="gpt-5-mini",
    temperature=0.2,
    max_tokens=4096,
)

agent = await build_agent(
    model=llm,
    servers=my_servers,
)

3. LangChain Runnable¶

Any LangChain Runnable that accepts chat messages and returns a response. Useful for custom chains or model wrappers:

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(model="gpt-5-mini").with_retry(max_retries=3)

agent = await build_agent(
    model=llm,
    servers=my_servers,
)

Using Ollama (Local Models)¶

Run models locally with Ollama -- no API key required:

# Install and start Ollama, then pull a model
ollama pull llama3

agent = await build_agent(
    model="ollama:llama3",
    servers=my_servers,
)

Local model limitations

Local models vary in their ability to use tools reliably. For production agent systems with tool calling, cloud providers (OpenAI, Anthropic) provide the most consistent results.

Default Model¶

All documentation examples use openai:gpt-5-mini as the default. It offers a good balance of capability, speed, and cost for development and testing.