v0.21.0

SymbolicAI v0.21.0 – Release Notes

What's Changed

• Local llama.cpp chat back end now targets the new server API, wiring in the updated tokenize/detokenize routes, propagating configured model/tool payloads, surfacing structured tool-call metadata, and exposing reasoning traces for downstream consumers.
• Embedding requests are aligned with the latest /v1/embeddings contract, adding explicit normalization control and reflecting the server’s batch output shape in the docs.
• Fallback execution now re-raises the originating exception on failure and treats limit as optional, preventing silent error swallowing in edge cases.

Full Changelog: v0.20.2...v0.21.0