docs: Fase C — R3 (Ollama), R4 (cron jobs), R5 (agent name), R6 (Qdrant)

R3: .env.example — Embedding backend reestruturado com Option A (Ollama,
recomendado para produção) e Option B (OpenRouter). install.md passo 4
atualizado para mencionar Ollama como opção local.

R4: install.md passo 8 — adicionados 3 cron jobs stack-base na tabela
e nos exemplos CLI: holographic-memory-backup (backup semanal do
memory_store.db), wiki-raw-ingest-monitor (detecção de drift em raw/),
maas-heartbeat (healthcheck Qdrant/Redis/ARQ).

R5: .env.example + install.md passo 6 — adicionado HERMES_AGENT_NAME.
state.py write_entry agora loga warning quando AGENT_NAME não está
definido.

R6: install.md Troubleshooting — nota sobre coexistência de coleções
Qdrant (knowledge_base + outras coleções de agentes externos).

.env.example

@@ -4,10 +4,6 @@
 
 # ── Required ──────────────────────────────────────────
 
-# OpenRouter API key — REQUIRED only when EMBEDDING_API_BASE points to OpenRouter.
-# For local providers (Ollama, vLLM, llama.cpp) leave unset or comment out.
-OPENROUTER_API_KEY=sk-or-...
-
 # Redis password (generate with: openssl rand -hex 16)
 REDIS_PASSWORD=
 
@@ -49,12 +45,14 @@ MAA_ENV_PATH=/home/your-user/.env
 
 # ── Embedding Backend ─────────────────────────────────
 
-# Default: OpenRouter with Qwen3-Embedding-8B.
-# Recommended model: multilingual (excellent for non-English content),
-# high-quality 4096d embeddings, fast inference, affordable pricing.
-# To use a local model (Ollama, vLLM, llama.cpp), set both vars:
-#   EMBEDDING_API_BASE=http://host.docker.internal:11434/v1
-#   EMBEDDING_MODEL=nomic-embed-text
+# Choose ONE backend. If both are set, Ollama takes priority (local, zero cost).
+#
+# Option A: Ollama (recommended for production, local, zero latency)
+OLLAMA_BASE_URL=http://localhost:11434
+OLLAMA_EMBEDDING_MODEL=nomic-embed-text
+#
+# Option B: OpenRouter (cloud, pay-per-use, multilingual Qwen3-Embedding-8B)
+# OPENROUTER_API_KEY=sk-or-...
 # EMBEDDING_API_BASE=https://openrouter.ai/api/v1
 # EMBEDDING_MODEL=qwen/qwen3-embedding-8b
 
@@ -65,6 +63,11 @@ MAA_ENV_PATH=/home/your-user/.env
 
 # ── Strongly Recommended ──────────────────────────────
 
+# Agent identity — distinguishes this agent in fabric entries and cross-agent
+# handoffs. Without this, fabric entries use "agent" as a fallback and
+# multi-agent deployments cannot tell agents apart.
+HERMES_AGENT_NAME=hermes
+
 # LLM extraction token limit — 1024 is too small, causes fabric truncation
 ICARUS_EXTRACTION_MAX_TOKENS=4096
 

icarus/state.py

@@ -313,6 +313,11 @@ def write_entry(entry_type, content, summary, tier="hot", tags="", platform="cli
     ts = now.strftime("%Y-%m-%dT%H%MZ")
     ts_iso = now.strftime("%Y-%m-%dT%H:%M:%SZ")
     agent = AGENT_NAME or "agent"
+    if not AGENT_NAME:
+        logger.warning(
+            "icarus: HERMES_AGENT_NAME not set — fabric entries will use agent=\"agent\". "
+            "Set HERMES_AGENT_NAME=<name> in .env for multi-agent deployments."
+        )
     suffix = secrets.token_hex(2)
     # derive a short slug from the summary for human-readable filenames
     slug = re.sub(r"[^a-z0-9]+", "-", summary.lower().strip())[:40].strip("-")

setup/install.md

@@ -165,6 +165,11 @@ Execution Agent protocol — insert it after the referenced section.
 **`SOUL.md`** — add Ground Truth level 2 (injected memory) and context
 injection convention as documented in `modifications/soul-rulebook.md`.
 
+**`~/.hermes/.env`** — set `HERMES_AGENT_NAME=hermes` (or any unique name).
+This distinguishes your agent in fabric entries and enables multi-agent
+handoff. Without it, all entries use the fallback `agent: "agent"` and
+cross-agent features are disabled.
+
 These modifications ensure the agent treats injected memory as more
 authoritative than training knowledge, and knows where to find
 persisted information without re-discovering it.
@@ -215,6 +220,9 @@ that keep the memory stack healthy. Copy them to a location of your choice
 | `pre_validator.py` | On-demand | Semantic linter — queries knowledge_base before I/O actions |
 | `reflection_trigger.py` | Every 5 min | Triggers micro_reflection when ARQ worker is idle |
 | `bulk_wiki_ingest.py` | One-shot | Initial bulk ingestion of existing wiki content |
+| `holographic-memory-backup.py` | Weekly (Mon 4am) | Dump and compress `memory_store.db` to backup directory |
+| `wiki-raw-ingest-monitor.py` | Twice/week (Mon/Thu 3am) | Detects new or drifted files in `raw/` vs FTS5 index |
+| `maas-heartbeat.py` | Every 6 hours | Health-check ping against Qdrant, Redis, and ARQ queue depth |
 
 **Using Hermes cron (recommended):**
 
@@ -246,6 +254,27 @@ hermes cron create \
   --script /path/to/scripts/semantic_dedup.py \
   --no-agent \
   --deliver local
+
+hermes cron create \
+  --name "holographic-memory-backup" \
+  --schedule "0 4 * * 1" \
+  --script /path/to/scripts/holographic-memory-backup.py \
+  --no-agent \
+  --deliver local
+
+hermes cron create \
+  --name "wiki-raw-ingest-monitor" \
+  --schedule "0 3 * * 1,4" \
+  --script /path/to/scripts/wiki-raw-ingest-monitor.py \
+  --no-agent \
+  --deliver local
+
+hermes cron create \
+  --name "maas-heartbeat" \
+  --schedule "0 */6 * * *" \
+  --script /path/to/scripts/maas-heartbeat.py \
+  --no-agent \
+  --deliver local
 ```
 
 **Before enabling decay scanner:** run `backfill_decay_metadata.py` once to
@@ -307,3 +336,11 @@ Check: OpenRouter API key is set, `context_enhancer.py` can import, gateway rest
 
 ### Decay scanner produces "0 archived" every week
 Most likely: point payloads missing `last_accessed_at` or `importance_score` metadata. Run backfill before enabling decay.
+
+### Multiple collections in Qdrant dashboard
+
+The Memory OS uses the `knowledge_base` collection exclusively. Other
+collections you may see (e.g., from other Hermes agent plugins or standalone
+agents) are safe to coexist — Qdrant isolates each collection at the storage
+and query level. Do NOT delete collections you did not create — they may
+belong to other agents sharing the same Qdrant instance.