107 Commits

Author SHA1 Message Date
aaron b936931668 Stage 3 worker v2.1 — saga-size limit + wedge detection + sudoers fixes
Production incident 2026-05-01: F14 re-cascade attempt surfaced three
compounding issues in cascade resilience.

stage3_worker.py changes:
- MAX_CHUNKS_PER_SAGA=10 — large documents split into multiple bulk
  commits, all sharing the same saga tag for Graphiti document linking.
  Original implementation sent all chunks as one saga; 17-19 chunk sagas
  deadlocked sidecar's Python-side coordination.
- recover_wedge() function — restarts aaronai-graphiti.service when
  consecutive_failures hits threshold. Mirrors Stage 2 pattern.
- run() loop adds consecutive_failures counter with threshold-2
  escalation. Resolves F28 + F29 from code review.
- Worker version bumped 2.0 -> 2.1.
- post_bulk() helper extracts shared HTTP POST + error handling.

Outside-repo changes (system config, separately documented):
- WatchdogSec=600 commented in stage2 + stage3 systemd unit files.
  Workers have no sd_notify support; per-request timeouts in code
  handle the actual failure modes.
- /etc/sudoers.d/aaron-aaronai created with NOPASSWD entries for
  systemctl restart ollama and restart aaronai-graphiti.service.
  Stage 2's existing recover_wedge() was silently broken since
  deployment due to this gap.

.gitignore — added rules for *.bak files, runtime artifacts
(watcher_heartbeat, dreamer_state.json, corpus_integrity_report.json,
watcher_state.json, watcher_status.json), Python cruft, virtual env,
.env, editor/OS files, and Aaron AI runtime data (conversations.db,
sessions.db, memory.md, settings.json).

Untracked 11 files that shouldn't have been committed in 465f2f7
(this morning): backup files and runtime artifacts.

Re-cascading Shop Class (414KB) and BirdAI-Experiments-Log.md (192KB)
through the patched worker after re-extracting full text from disk.
Cascade in progress at commit time.
2026-05-01 05:18:09 +00:00
aaron 465f2f725b Code review fixes: CV pinning, F1 (excluded_sources), F14 (50KB truncation), F37
- api.py: strip CV pinning workaround (parity violation, see architecture doc)
- dream.py: F1 — retrieve_graphiti() now accepts excluded_sources, over-fetches
  3x and filters in-process. Was silently dropping the parameter; would have
  confounded E3 with broken cross-stage exclusion in Graphiti arm.
- watcher.py + ingest.py: F14 — drop full_text[:50000] truncation. Was
  propagating through entire cascade. Postgres TEXT can hold up to 1GB.
- corpus_integrity.py: F37 — same truncation, third path now clean.

Backups: api.py.bak.*, dream.py.bak.*, watcher.py.bak.*, ingest.py.bak.*,
corpus_integrity.py.bak.* timestamped pre-fix.

Re-cascaded Shop Class as Soulcraft (only already-cascaded source affected
by F14, 414KB).
2026-05-01 02:26:37 +00:00
aaron 25e42c0231 corpus_integrity.py: write unreadables with retry_count=0 so OCR can retry when it ships 2026-04-30 22:03:48 +00:00
aaron 7822fb1cc1 corpus_integrity.py: write unreadable files to ingest_failures for UI visibility 2026-04-30 21:59:06 +00:00
aaron 74e2c34f43 corpus integrity: ingest_failures tracking in watcher, reconciliation script, corpus status/retry/reconcile endpoints 2026-04-30 21:54:39 +00:00
aaron f11cacd9c9 add experiment scripts and results; watcher.py latest changes 2026-04-30 18:06:03 +00:00
aaron 1cf26df450 api.py: return error_type=transcription_failed on Whisper crash, frontend retry logic can now distinguish from network failures 2026-04-30 17:45:47 +00:00
aaron 7cd765146a stage3_worker.py: log sidecar response body on non-200 2026-04-30 17:37:28 +00:00
aaron 58515ebec0 graphiti_service.py: add traceback logging, log file handler for all endpoints 2026-04-30 17:36:19 +00:00
aaron 91166367fa E3: add Graphiti retrieval branch to dream.py, E3 experiment script with blinding 2026-04-30 17:17:28 +00:00
aaron 2b3c2380a0 watcher.py: in-process ingest, embedder loaded once at startup, startup recovery, heartbeat, no duplicate logging 2026-04-30 16:42:44 +00:00
aaron 2fb50cce71 ingest.py: guard Stage 2 enqueue behind SKIP_STAGE2_ENQUEUE env var for migration runs 2026-04-30 16:20:11 +00:00
aaron c08f57a6f2 stage2/3 workers: remove duplicate StreamHandler, stdout captured by systemd 2026-04-30 16:12:51 +00:00
aaron cae7fb8775 dream.py v1.1: score-band exclusion for Early REM, DREAMER_VERSION constant, manifest versioning 2026-04-30 15:51:11 +00:00
aaron b53717af5b dream.py: enrich manifest with retrieval breadth metrics 2026-04-30 06:14:55 +00:00
aaron 2b9a1782c1 feat: stage2/3 pipeline, taxonomy-free cascade, E1.8/E4 experiments, corpus migration state 2026-04-30 04:04:31 +00:00
aaron 62b5b5453a fix: max_coroutines=2, saga support in sidecar; stage3 chunking; TIMEOUT_MAX 0 persistent in falkordb compose 2026-04-30 04:01:02 +00:00
aaron 95d022ec64 fix: FalkorDriver database=aaron, build indices on correct graph 2026-04-29 21:34:20 +00:00
aaron d91a5675ff capture: public SSE endpoint for transcription completion events 2026-04-29 18:00:54 +00:00
aaron c42d898504 emit capture_saved SSE event when async transcription completes 2026-04-29 17:58:01 +00:00
aaron a05fcec882 async voice transcription — return immediately, whisper runs in background 2026-04-29 17:48:22 +00:00
aaron eb7cf3be10 upgrade whisper small -> large-v3, bump cpu_threads to 8 2026-04-29 17:35:03 +00:00
aaron 3f6c435be4 add client_time to chat context — user-supplied, not logged 2026-04-29 17:26:03 +00:00
aaron 21557790d9 capture: return error_type on transcription failure instead of HTTP 500 2026-04-29 17:04:56 +00:00
aaron 794e0aeddd update whisper prompt: add BirdAI stack terms, remove stale ChromaDB 2026-04-29 16:47:30 +00:00
aaron d271e17929 add sourcing constraint to system prompt, close hallucination gap 2026-04-29 16:37:39 +00:00
aaron 5d83fb7601 fix: load_dotenv override=True, option b source exclusion 2026-04-29 16:32:09 +00:00
aaron 83d4f60d0d option b: cross-night source exclusion in dream pipeline 2026-04-29 16:19:52 +00:00
aaron b6fe350ab2 experiments: add consistency test and briefing generator results + scripts 2026-04-28 02:47:41 +00:00
aaron 037d747573 chore: archive deprecated chromadb and migration scripts 2026-04-28 00:15:46 +00:00
aaron d5b5c2ec14 Graphiti sidecar service + SentenceTransformer embedder — self-hosted, no OpenAI dependency 2026-04-27 18:21:22 +00:00
aaron 4ee2567400 Add SentenceTransformer embedder for Graphiti — self-hosted, no OpenAI dependency 2026-04-27 18:18:37 +00:00
aaron a1f732fc9e Dreamer: manifest writer, Late REM v1.2 (remove coherence pull) 2026-04-27 16:54:18 +00:00
aaron 03b3f012c3 Dreamer: prompt versioning, Early REM v1.1, prompt signature in headers 2026-04-27 16:50:21 +00:00
aaron 6776637178 Remove hardcoded PG password fallbacks — require PG_DSN env var in all scripts 2026-04-27 05:16:37 +00:00
aaron a1f5c1049a Fix dreamer status display, watcher excludes Media/, remove NVM debt item 2026-04-27 05:08:01 +00:00
aaron d3239aba17 Image capture — extend /api/capture for image+voice, Claude vision description, Media/ WebDAV, watcher excludes Media/ 2026-04-27 04:28:31 +00:00
aaron ef2fddc47f Redesign dreamer — interdependent pipeline, NREM→Early REM→Late REM→Synthesis 2026-04-26 23:41:24 -04:00
aaron 7af246ac01 APScheduler — replace systemd timers, in-process dream and ingest scheduling 2026-04-27 03:04:33 +00:00
aaron 9b312d936f Add SSE endpoint and dream notify — /api/events and /api/events/notify 2026-04-27 02:20:50 +00:00
aaron 9088b5643d Add /api/dreamer/status and /api/dreamer/run endpoints 2026-04-27 01:27:09 +00:00
aaron a07de922df Add /api/capture and /api/captures endpoints — auth-free, WebDAV delivery to Journal/Captures/ 2026-04-26 22:39:55 +00:00
aaron 8c8fba11b8 Add nightly conversation indexing — Aaron AI conversations into pgvector at 2:30AM 2026-04-26 21:28:40 +00:00
aaron f78b83042b Migrate to pgvector — remove ChromaDB from api.py, ingest scripts, dream.py 2026-04-26 21:16:04 +00:00
aaron d2eed98906 Pre-pgvector migration checkpoint — upsert, allow_replace_deleted, maintenance timer 2026-04-26 20:19:49 +00:00
aaron 72c8d1ef2c Recalibrate mode similarity ranges for current corpus 2026-04-26 15:57:41 -04:00
aaron 08b95f1c0c Update ingest_claude.py — handle .json format, ingest memories, handle both array and jsonl 2026-04-26 19:13:17 +00:00
aaron 679222ad5f Add dream.py — WebDAV delivery, four validated modes, calibrated similarity ranges 2026-04-26 15:04:25 -04:00
aaron fd249c5e96 Add dream.py — four mode dreamer with validated prompts and calibrated similarity ranges 2026-04-26 14:45:28 -04:00
aaron 619a9295ce Add Claude conversation export ingestion script 2026-04-26 13:10:02 -04:00