CHANGELOG.md

Changelog

0.4.0

Major Features

Breaking Changes:

• chat() method now returns full ChatResponse object instead of just the response string
  • Response includes: ok, response, intermediate_steps, chat_id, and optional references
  • Migration: Change response = await skald.chat(...) to response = await skald.chat(...); text = response["response"]
• delete_memo() method now returns {"ok": bool} instead of None

New Features:

Conversation Continuity:

• Added chat_id field to ChatRequest and ChatResponse for multi-turn conversations
• Use chat_id from previous response to maintain conversation context

Advanced RAG Configuration:

• Added comprehensive RAGConfig type with full pipeline control:
  • llm_provider: Choose between "openai", "anthropic", or "groq"
  • query_rewrite: Enable query reformulation for vague queries
  • vector_search: Configure top_k (1-200) and similarity_threshold (0.0-1.0)
  • reranking: Enable sophisticated reranking with configurable top_k (1-100)
  • references: Enable citation markers and source attribution

References/Citations System:

• Added References and MemoReference types
• Chat responses can include [[N]] citation markers with source mapping
• Streaming chat now emits "references" event type with JSON-encoded references
• Updated ChatStreamEvent to support "references" event type

System Prompt Customization:

• Added system_prompt field to ChatRequest for custom AI behavior

Removed:

• SearchMethod type (obsolete - search method is no longer configurable)

0.3.1

• Remove legacy documentation for document generation methods

0.3.0

• Added create_memo_from_file method for document uploads (PDF, DOC, DOCX, PPTX)
• Added check_memo_status method to track memo processing status
• create_memo now returns {"memo_uuid": str} instead of { "ok": True }

0.2.0

• Removed generate and streamed_generate methods
• Removed search_method from search, now uses semantic search and is no longer configurable
• chat method now returns the actual response content from the model rather than a JSON response.pytho