Message Splitter
Separate conversation messages by speaker role, topic shift, and thread boundary. Long, monolithic AI conversations become navigable segments that can be individually searched, extracted from, and linked to related material across the corpus.
Core Capabilities
| Split |
Label |
Link |
| Automatic speaker role identification and topic shift detection within long conversations. |
Each segment receives auto-generated topic summaries and speaker attribution. |
Individual segments are searchable, extractable, and linkable across the corpus. |
The Problem We Solve
"My AI conversations are 200-message walls of text"
A single AI session might span five topics over hundreds of messages. Without splitting, it's an unsearchable monolith. The Message Splitter creates clean, navigable segments — each with a topic label, speaker attribution, and position in the thread.
"I can't search for the part where we discussed X"
Full-conversation search returns the entire conversation. Segment-level search returns just the relevant portion — the exact 15 messages where you discussed database schema design, not the 200-message session that happened to contain it.
"The conversation mixed three different topics"
AI conversations naturally drift between topics. The Message Splitter detects these shifts and creates separate segments so each topic can be independently searched, linked, and extracted without the others.
How It Works
- Ingest — Conversations arrive from the Import Queue in normalised format
- Identify — Speaker role detection classifies each message as user, AI, or system
- Detect — Topic shift analysis identifies points where the conversation changes subject
- Split — The conversation is divided into segments at topic boundaries
- Label — Each segment receives an auto-generated topic summary based on content analysis
- Index — Segments enter the Corpus Index as individually searchable, linkable items
What We Deliver
- Automatic speaker role identification — user, AI, system
- Topic shift detection within long conversations
- Thread boundary identification for multi-topic sessions
- Segment labelling with auto-generated topic summaries
- Individual segment linking for targeted search results
- Segment-level extraction — pull code, insights, or summaries from specific segments
- Segment timeline showing topic evolution within a conversation
- Manual split point adjustment for edge cases
Integration with Other Features
- Import Queue — Conversations flow from import to splitting automatically
- Code Extractor — Code blocks are identified within segments with full context
- Corpus Index — Segments appear as individual searchable items
- Topic Clusters — Segments from different conversations cluster together by topic
Related
→ Explore All Features