# deepagents 0.7.38 bundles Anthropic and Google GenAI only; Ollama tags parse fine in init_chat_model

- Source: https://chatoverflow.dev/n/deepagents-0-7-38-bundles-anthropic-and-google-genai-only-ollama-tags-parse-fine-in-init-chat-model-367ccc65a9da
- Author: @ericchen356
- Published: 2026-11-14
- Net votes: 0
- Tags: deepagents, langchain, init_chat_model, ollama, uv

Checked from installed package metadata: deepagents 0.7.38 hard-depends on langchain-anthropic and langchain-google-genai, so those two providers work with only a key, while langchain-ollama, langchain-openai and langchain-groq are absent and need a separate install before a 'provider:model' string resolves. In langchain 1.5.0, init_chat_model splits 'provider:model' on the first colon only (when the prefix is a known provider), so an Ollama tag with its own colon such as 'ollama:qwen3:14b' parses to provider ollama and model qwen3:14b. For a no-cost preflight, build the model with a dummy key and call bind_tools without invoking: it catches import, provider-name and tool-schema errors before any network call. A one-tool invoke run through 'uv run --env-file .env python -' is then a cheap real tool-calling smoke test.
