apollo-telegram-ingest

# Apollo Telegram Ingest Captures flagged Telegram conversation sessions as structured, searchable summary inputs inside `Inputs/Telegram/`. These records summarize the session's context, capture key takeaways or next steps, and store the metadata required to look up the complete, raw transcripts in the local session database (`state.db`). ## Synced Path - Initial Creation / Folder: `/home/justin.guest/Developer/obsidian-vault/Inputs/Telegram/YYYY-MM-DD - Telegram - [Title].md` (e.g., `2026-06-11 - Telegram - Background Ingestion of Flagged Telegram Sessions.md`) - Category: `[[Telegram]]` --- ## Triggering Ingestion Sessions are flagged for ingestion in two ways during Telegram chats: 1. **Reaction:** Justin reacts to any bot response using the `🧠` emoji. 2. **Text Message:** Justin includes the `🧠` emoji in any of his messages during the session. --- ## State & Schema Integration Ingestion runs in the background to avoid polluting active session context. - **Database Path:** `/home/justin.guest/.hermes/state.db` - **Schema Columns (Added in `sessions` table):** * `brain_flagged INTEGER DEFAULT 0` — set to `1` when flagged. * `ingested INTEGER DEFAULT 0` — set to `1` once processed by the background cron job. ### Platform Hooks (Reaction Capture) A `MessageReactionHandler` is configured in `gateway/platforms/telegram.py` to listen for incoming reaction updates. When a `🧠` reaction is detected: 1. It looks up the message ID in the `messages` table under `platform_message_id`. 2. It resolves the associated `session_id`. 3. It updates `sessions` to set `brain_flagged = 1` directly in `state.db`. --- ## Background Plumbing (Cron & Poller) A background poller script `fetch_telegram_brains.py` runs on a schedule (every 30 minutes) as part of a scheduled cron job: - **Scan Phase:** Queries `state.db` for sessions that are either marked `brain_flagged = 1` or contain a user message with the `"🧠"` string, where `ingested = 0`. - **Extraction Phase:** Formats the session metadata and raw message history, printing it as JSON to be digested by the background agent. - **Mark Phase:** Runs with `--mark-processed <SESSION_ID>` to set `ingested = 1` in `state.db`. --- ## Ingest Note Structure Every Telegram session log note must use this frontmatter format: ```yaml --- id: 'YYYYMMDDHHmmss' # Numerical string based on session start time daily_note: "[[Daily Notes/YYYY-MM-DD Weekday|YYYY-MM-DD Weekday]]" # Symmetrical daily note link category: "[[Telegram]]" # Quoted category link session_id: "YYYYMMDD_HHMMSS_xxxxxxxx" # The unique session identifier in state.db source_db: "/home/justin.guest/.hermes/state.db" # Database where session resides --- ``` Below the frontmatter, the body is structured as follows: ```markdown # 📥 Telegram Session Capture: [Title] - **Session ID:** `session_id` - **Date:** YYYY-MM-DD - **Database:** `/home/justin.guest/.hermes/state.db` ## Summary - **Topic A:** [Provide a concise, high-quality summary of the session: what was discussed, decisions made, and any open questions.] - **Topic B:** [Concise statement, attributing decisions/takeaways accurately] ```