Phase 1 of the Scalable Approach for Adding New IDE Session Types RFC. Adding an IDE session type (Cursor, Windsurf, ...) currently requires changing 15+ files across migrations, proto, server, SDK, and UI.

Replace the fixed session_count_* columns on the ephemeral workspace_agent_stats table with a normalized child table keyed by app name, and report session counts from the agent as a dynamic proto map, so new IDEs flow end-to-end with zero schema, proto, or server changes.

Scope

• New workspace_agent_session_counts (workspace_agent_stats_id, app_name, count) child table with ON DELETE CASCADE; clean-cut migration copies the ~1 day ephemeral buffer, drops the four columns, and trims the insights covering index
• Proto Stats.session_counts map (field 13, Agent API v2.11); fields 8–11 stay deprecated and the server converts them, so old agents keep reporting without changes
• Agent: dynamic per-app counters; extractMagicSessionType normalizes instead of classifying against a fixed set; JetBrains process detection and SSH metric labels grouped by family
• New coderd/idemetadata leaf package: Normalize() (strip null bytes, truncate to 64 runes, lowercase, fold - to _ so reconnecting-pty stays canonical as reconnecting_pty, empty becomes unknown) plus the family map used for grouping
• Bounded input: a stats report is capped at 64 distinct normalized names, overflow summed into unknown; workspace usage API and coder ssh --usage-app accept arbitrary app names
• Read paths: latest-row queries aggregate the child table via LEFT JOIN LATERAL; rollup and template insights use plain hash joins (benchmarks in the PR description)
• External surfaces stay byte-identical: deployment stats API, Prometheus gauge names, and telemetry snapshots are unchanged until phase 2 (Expose per-app session counts in deployment stats, Prometheus, and the UI #27411)

Done when

• feat: normalize workspace agent session counts into a child table #27179 is merged
• Deployment stats, Prometheus, telemetry, and insights output are identical before and after the migration
• Old agents (fixed proto fields) and new agents (map) both report correctly against the new server

All phases

1. This issue (in review: feat: normalize workspace agent session counts into a child table #27179)
2. Expose per-app session counts in deployment stats, Prometheus, and the UI #27411 — expose per-app counts in deployment stats, Prometheus, UI
3. Migrate connection_logs.type from ENUM to TEXT #27412 — connection_logs.type ENUM → TEXT
4. Merge template_usage_stats *_mins into app_usage_mins JSONB #27413 — template_usage_stats *_mins → app_usage_mins JSONB
5. Report the actual IDE (Cursor, Windsurf, ...) for session tracking vscode-coder#1044 — report the actual IDE from the extension (and Send the specific JetBrains IDE name for session tracking jetbrains-coder#599 for JetBrains)

🤖 Created by Coder Agents on behalf of @EhabY.