symphony-core

Provider integrations and model-neutral streaming types for Symphony agents. Python import: core_ai.

Availability: this package is not currently published on PyPI. Use it from the repository's uv workspace and follow the source installation guide.

Minimal example

from core_ai import build_default_registry, default_model_id

registry = build_default_registry()
model_id = default_model_id(registry)
print(model_id)  # provider:model

The registry includes supported cloud providers for which credentials are configured. Ollama and generic local OpenAI-compatible providers are opt-in. See the provider/environment reference.

What the package provides

• Provider adapters for OpenAI, Anthropic, Gemini, Grok, OpenRouter, Vercel AI Gateway, Ollama, and compatible local endpoints.
• ModelRegistry and provider-prefixed model IDs (provider:model).
• Provider-independent Message and StreamEvent types, including streamed text and tool calls.
• A checked-in generated catalog of supported tool-calling models; ordinary installation/build does not refresh it.

For public interfaces and streaming behavior, see the symphony-core guide. To build an agent that uses these providers, continue to symphony-harness.

Maintainer: refresh the checked-in catalog

The generated model list is a snapshot; building the workspace never updates it. To deliberately refresh and commit the catalog:

cd core_ai
uv run python scripts/generate_models.py

The script reads models.dev. When that source is unavailable, it keeps the existing snapshot unless CORE_AI_GENERATE_STRICT=1 is set. See scripts/generate_models.py for maintainer options.