Platform-aware session clustering (Telegram vs CLI volume asymmetry)

[dream-hermes-agent]

The digest treats Telegram and CLI sessions identically, but they have very different volume profiles (~14 sessions/day Telegram vs ~10/day CLI). Clustering parameters should be adjusted per platform to avoid Telegram sessions drowning out CLI sessions in the digest.

Context

From the April 23 session analysis:

• Telegram: 14 sessions / 2,892 msgs / 5.3 MB (avg ~206 msgs/session)
• CLI: 10 sessions / 2,387 msgs / 4.0 MB (avg ~239 msgs/session)

Despite similar total message counts, the session counts and distribution differ significantly. A single clustering threshold treats them the same, causing the digest to over-represent Telegram activity patterns.

Acceptance Criteria

• Platform (telegram/cli) passed as a signal to clustering
• Volume-weighted scoring or separate thresholds per platform
• Session count and message count shown per platform in digest output
• Venice summary prompt includes platform context

Notes

See session_digest.py — match_tags() currently uses the same keyword-matching logic regardless of platform. The extract_session_metadata() function already detects platform from filename pattern.