Repository Issue Activity (beta)

#1143 [Bug] update_peer_card silently truncates an over-cap card, so a card at the cap can never take a new entry

#1153 [Bug] LLM retry loop retries non-retryable errors and defers fallback to the final attempt

#1075 [Bug] Server and deriver cannot start on native Windows (fcntl import, add_signal_handler)

#729 Dreamer/deriver accumulates near-duplicate observations despite dedup guard

#940 [Bug] Re-adding an active session peer resets joined_at and hides perspective search results

#1107 [Feature] Expose workspace-level chat (PR #931) as an MCP tool

#967 Document (memory) retrieval should use hybrid search (semantic + full-text), not pure embedding

#1142 [Feature] Add Milvus as an optional VectorStore backend

#734 [Bug] SDK dialectic/chat calls need clearer timeout/shutdown ergonomics for threaded embedders

#1063 Make API Uvicorn worker count configurable (API_WORKERS)

#963 Transport failures permanently burn queue items (processed-with-error, never retried)

#728 Deriver silently marks representation work units `processed` when the embedding save fails on a transient 429 (no retry, no fail-loud → silent memory loss)

#771 [Bug] Deriver generates zero observations on chain-of-thought models (e.g. MiniMax-M3 / api.minimax.io)

#817 [Bug]: Deriver misattributes AI agent speech to the human user in multi-peer sessions

#1087 [Quality] Two-peer sessions do not promote cross-speaker target facts to the observed peer

#1105 [Quality] User peer card stays empty under Hermes auto-config despite agent-side observations being full of user-attributed facts

#1101 Tool calls in one assistant turn run serially, so a turn costs the sum of its reads

#1099 [Bug] Five scope-route tests fail on a clean checkout when the local .env sets DREAM_ENABLED=false

#1097 [Bug] pgvector-only reconciliation never processes documents

#956 No message-level DELETE — removing a single message requires deleting the whole session

#1078 [Bug] Agent calls `search_message` but Honcho only knows `search_messages`

#1082 conclusions query ranks recent/generic conclusions over query-relevant ones

#1061 OpenAI tool-call assistant `content: null` is normalized to an empty string during replay

#568 Misleading 422 from /search hides the real embedding error

#1048 [Bug] Filtered pgvector HNSW conclusion queries silently under-return with iterative scans off