feat(llm): virtualize provider interface behind a normalized Adapter - #1683

Merged

dsapandora merged 25 commits into

feat/llm-provider-adapter

Jul 31, 2026

Commits on Jul 31, 2026

refactor(llm): extract stream content parsing into a named seam
dsapandora
committed
feat(llm): add normalized provider Adapter interface
dsapandora
committed
refactor(llm): move stream content parser to the adapter module
dsapandora
committed
feat(llm): add LangChainAdapter
dsapandora
committed
feat(llm): add AnthropicAdapter
dsapandora
committed
feat(llm): add OpenAIAdapter
dsapandora
committed
feat(llm): add drive_adapter Event→callback shim
dsapandora
committed
feat(llm): drive the LangChain streaming path through the adapter
dsapandora
committed
fix(llm): flatten non-streaming _chat content to a string
dsapandora
committed
feat(llm): route the native Anthropic path through an adapter
dsapandora
committed
feat(llm): route the OpenAI Responses path through an adapter
dsapandora
committed
refactor(llm): unify the non-streaming path through the adapter
dsapandora
committed
feat(llm_anthropic): per-node extended-thinking control with context gate
dsapandora
committed
test(agent_crewai): drive NativeAnthropicAdapter for native stop-threading
dsapandora
committed
fix(sync-models): make the extendedThinking toggle capabilities-aware
dsapandora
committed
refactor(llm): address CodeRabbit — dead adapters, store=False, Responses fallback
dsapandora
committed
fix(sync-models): scope the extendedThinking toggle to nodes that define the field
dsapandora
committed
docs(llm_anthropic): soften the extendedThinking field description
dsapandora
committed
fix(llm): keep the non-streaming path on invoke via adapter.collect()
dsapandora
committed
fix(llm): address review — parse_bool, opus-5 toggle, custom, Responses tests
dsapandora
committed
fix(llm): close Responses stream + honor Haiku 4.5 extended thinking
dsapandora
committed
refactor(llm): reuse LangChainAdapter.collect() in the Responses fallback
dsapandora
committed
refactor(llm): expose collect() reasoning so the fallback keeps its lane
dsapandora
committed
refactor(llm): read the block vocabulary from flatten_content_blocks
dsapandora
committed
docs(llm_anthropic): describe the extended-thinking toggle as opt-in
dsapandora
committed