Skip to main content

Module conversation

Module conversation 

Source
Expand description

Core conversation types and processing.

Defines recall-echo’s own conversation types — these are the universal internal format. All input adapters (JSONL transcripts, pulse-null Messages) produce these types, which then flow into the archive pipeline.

Structs§

Conversation
A parsed conversation — metadata + entries. Produced by input adapters (JSONL, pulse-null), consumed by archive pipeline.

Enums§

ConversationEntry
A parsed conversation entry — the universal internal format. All input adapters produce these.

Functions§

calculate_duration
Calculate duration string from two ISO timestamps.
condense_for_summary
Condense a conversation into a text block suitable for LLM summarization. Keeps it short to minimize token usage.
conversation_to_markdown
Convert conversation entries into a markdown document for archival.
date_from_timestamp
Extract just the date portion from an ISO timestamp.
extract_summary
Algorithmic summary extraction — first user message, truncated.
extract_topics
Algorithmic topic extraction from conversation entries. Uses keyword frequency with stop-word filtering and tool-target boosting.
strip_channel_prefix
Strip [Channel: …] and “User message:” prefixes from text.
truncate
Truncate a string, appending a notice if it was cut.
utc_now
Get current UTC timestamp in ISO 8601 format.