A map of what we kept returning to.
1,670 meaning-only Telegram posts, embedded and grouped into 18 semantic islands. The gray space is part of the result.
- Posts
- 1,670
- Islands
- 18
- Bridges
- 498
- Snapshot
- 16 Jul 2026
Discovered islandBridge / outlier
Island index
What grew large in this archive.
Island size is attention, not importance. A large territory means the archive returned to that semantic neighborhood often; it does not make the topic more true.
01AI-native work26615.9%
02Claude Code1187.1%
03OpenAI platform754.5%
04Memory + context744.4%
05Gemini + media744.4%
06Multi-agent systems613.7%
07Coding-agent CLI543.2%
08Evals523.1%
09Prompting472.8%
10Agent skills472.8%
11Codex workflows432.6%
12Learning432.6%
13Cursor + IDE392.3%
14RAG + retrieval392.3%
15Runtime + sandbox392.3%
16Browser agents372.2%
17Claude models321.9%
18MCP + protocols321.9%
The terrain, not the timeline.
This snapshot is an embedded view of the public qwGAI Telegram archive. Islands are formed by meaning, not chronology or a prewritten taxonomy. Bridges are posts the density model could not assign cleanly.
- Source
- @qwgai public channel
- Corpus
- Semantic bodies only; links and Telegram metadata excluded
- Cutoff
- 16 Jul 2026
- Method
text-embedding-3-large· 1,024 dimensions · UMAP 15D → HDBSCAN · UMAP 2D for display- Embedding pass
- $0.1003 estimated API input cost
- Truth note
- The 2D shape is navigational, not distance measured in the world.
70.2% assigned29.8% bridges0.856 projection trustworthiness0.747 mean seed stability
Next snapshot: movement.
Repeat the same pipeline, align the map to stable anchors, then compare share, drift, births, splits, and merges.
+future run