compact-framework

To summarize long Telegram threads into structured memory, this Skill distills conversations using an 80/20 rule to preserve critical decisions and tasks. It creates a searchable Warm Memory layer, preventing token explosion while retaining essential context for future reference.

The best way to manage RAM usage during chat distillation is using RAM-aware processing. This Skill automatically selects cloud or local AI based on available system memory, preventing resource exhaustion when processing extensive Telegram thread conversations.

Yes, ripgrep is required to search existing memories before distilling Telegram threads. The Skill operates a Search First phase using ripgrep to locate prior memories, preventing duplicate storage and ensuring context continuity across multi-tier storage layers.

Yes, you can use the optional Gemini API for cloud distillation of Telegram threads. When local RAM is insufficient, the RAM-aware processing phase selects the cloud AI alternative to distill conversations without causing system resource exhaustion.

Multi-tier storage handles archived Telegram thread memories by routing data across Hot, Warm, and Cold tiers. Distilled memories are stored under /root/.openclaw/memory, where specific signals trigger archival to the appropriate tier for long-term retrieval.

Summarizing long Telegram chats preserves critical decisions by applying the 80/20 distillation rule, which filters out conversational filler and retains only essential tasks, decisions, and metadata. This structured approach ensures searchable Warm Memory without token bloat.