Models

The framework ships five model adapters and accepts any class that satisfies the Model protocol. String resolution by prefix means most users never construct an adapter directly.

from loomflow import Agent
 
agent = Agent("...", model="claude-opus-4-7")     # → AnthropicModel
agent = Agent("...", model="gpt-4o")              # → OpenAIModel
agent = Agent("...", model="mistral-large")       # → LiteLLMModel
agent = Agent("...", model="bedrock/claude-3...")  # → LiteLLMModel
agent = Agent("...", model="ollama/llama3")       # → LiteLLMModel (local)
agent = Agent("...", model="echo")                # → EchoModel (zero-key)

Five shipped adapters

When to construct an instance directly

• Custom client config. Pass your own AsyncOpenAI / AsyncAnthropic:

  from openai import AsyncOpenAI
  from loomflow import Agent
  from loomflow.model.openai import OpenAIModel
   
  client = AsyncOpenAI(timeout=30.0, max_retries=0)
  agent = Agent("...", model=OpenAIModel("gpt-4o", client=client))

• Tuning provider-specific knobs (max_tokens, temperature defaults, metadata):

  from loomflow.model.anthropic import AnthropicModel
   
  agent = Agent("...", model=AnthropicModel(
      "claude-opus-4-7",
      api_key="...",
      max_tokens=8192,
      temperature=0.2,
  ))

Read more

Custom Model adapter

The Model protocol requires only stream(). complete() is optional — adapters may override it with a non-streaming call, and otherwise it falls back to consuming the stream internally.

from collections.abc import AsyncIterator
from typing import Protocol, runtime_checkable
from loomflow.core.types import Message, ModelChunk, ToolDef
 
@runtime_checkable
class Model(Protocol):
    @property
    def name(self) -> str: ...
 
    async def stream(
        self,
        messages: list[Message],
        *,
        tools: list[ToolDef] | None = None,
        output_schema: type | None = None,
    ) -> AsyncIterator[ModelChunk]:
        ...

stream() is an async generator yielding ModelChunks. Every adapter yields:

• ModelChunk(kind="text", text="...") for streaming tokens.
• ModelChunk(kind="tool_call", tool_call=...) for assistant tool calls.
• ModelChunk(kind="usage", usage=...) for input/output token counts.

The agent loop aggregates chunks; it doesn’t care about chunk boundaries.