[picooraclaw] setup-oracle: docker-health fallback + refreshed ONNX URL; identity: don't hardcode model

The 'DATABASE IS READY TO USE' log marker only appears on first boot of
gvenzl/oracle-free, so re-running setup-oracle.sh on an already-initialized
container timed out after 180s even though the DB was healthy. Added a
fallback to docker's own healthcheck status.

The ONNX zip moved from the OML-Resources bucket to OML-ai-models with
a new pre-authenticated URL. Made the URL overridable via $ONNX_URL.

IDENTITY.md: stop hardcoding 'Default to Ollama (qwen3:latest)' in the
agent's self-description. The actual model lives in config.json; the
system prompt should defer to that instead of forcing the LLM to lie
about its provider when the operator has configured something else (e.g.
xai.grok-4 via OCI GenAI).

Author: jasperan

apps/picooraclaw/scripts/setup-oracle.sh

@@ -52,7 +52,14 @@ fi
 info "Waiting for database to be ready..."
 TIMEOUT=180
 ELAPSED=0
-while ! docker logs "$CONTAINER_NAME" 2>&1 | grep -q "DATABASE IS READY TO USE"; do
+db_ready() {
+    # First boot prints the marker in container logs; subsequent boots don't.
+    # Fall back to docker's healthcheck status for already-initialized containers.
+    docker logs "$CONTAINER_NAME" 2>&1 | grep -q "DATABASE IS READY TO USE" && return 0
+    [ "$(docker inspect --format '{{.State.Health.Status}}' "$CONTAINER_NAME" 2>/dev/null)" = "healthy" ] && return 0
+    return 1
+}
+while ! db_ready; do
     sleep 5
     ELAPSED=$((ELAPSED + 5))
     printf "\r  Waiting... %ds" "$ELAPSED"
@@ -114,7 +121,7 @@ section "Step 5/6: ONNX embedding model"
 info "Downloading Oracle's augmented all-MiniLM-L12-v2 model..."
 ONNX_WORK="/tmp/onnx_model"
 mkdir -p "$ONNX_WORK"
-curl -fsSL "https://adwc4pm.objectstorage.us-ashburn-1.oci.customer-oci.com/p/VBRD9P8ZFWkKvnfhrWxkpPe8K03-JIoM5h_8EJyJcpE80c108fuUjg7R5L5O7mMZ/n/adwc4pm/b/OML-Resources/o/all_MiniLM_L12_v2_augmented.zip" \
+curl -fsSL "${ONNX_URL:-https://adwc4pm.objectstorage.us-ashburn-1.oci.customer-oci.com/p/TtH6hL2y25EypZ0-rrczRZ1aXp7v1ONbRBfCiT-BDBN8WLKQ3lgyW6RxCfIFLdA6/n/adwc4pm/b/OML-ai-models/o/all_MiniLM_L12_v2_augmented.zip}" \
   -o "$ONNX_WORK/model.zip"
 cd "$ONNX_WORK" && unzip -o model.zip && cd -
 ONNX_FILE=$(find "$ONNX_WORK" -name "*.onnx" -type f | head -1)

apps/picooraclaw/workspace/IDENTITY.md

@@ -11,8 +11,8 @@ Ultra-lightweight personal AI assistant written in Go, inspired by nanobot.
 
 ## Purpose
 - Provide intelligent AI assistance with minimal resource usage
-- Default to Ollama (qwen3:latest) for local, private inference
-- Support multiple LLM providers (OpenAI, Anthropic, Groq, etc.) when configured
+- Run on whichever LLM the operator has configured (the model name and provider live in `~/.picooraclaw/config.json`; do not assume Ollama, OpenAI, or any specific model unless the user asks)
+- Support OCI Generative AI, Ollama, OpenAI, Anthropic, Groq, OpenRouter, and other OpenAI-compatible providers
 - Enable easy customization through skills system
 - Run on minimal hardware ($10 boards, <10MB RAM)
 
@@ -54,4 +54,4 @@ Discussions: https://github.com/jasperan/picooraclaw/discussions
 ---
 
 "Every bit helps, every bit matters."
-- PicoOraClaw
+- PicoOraClaw