model routing: cost/latency ranking with ranked fallback list (#849)

config/plano_config_schema.yaml

@@ -9,6 +9,7 @@ properties:
       - 0.1-beta
       - 0.2.0
       - v0.3.0
+      - v0.4.0
 
   agents:
     type: array
@@ -470,6 +471,106 @@ properties:
         additionalProperties: false
         required:
           - jailbreak
+  routing_preferences:
+    type: array
+    items:
+      type: object
+      properties:
+        name:
+          type: string
+        description:
+          type: string
+        models:
+          type: array
+          items:
+            type: string
+          minItems: 1
+        selection_policy:
+          type: object
+          properties:
+            prefer:
+              type: string
+              enum:
+                - cheapest
+                - fastest
+                - none
+          additionalProperties: false
+          required:
+            - prefer
+      additionalProperties: false
+      required:
+        - name
+        - description
+        - models
+        - selection_policy
+
+  model_metrics_sources:
+    type: array
+    items:
+      oneOf:
+        - type: object
+          properties:
+            type:
+              type: string
+              const: cost_metrics
+            url:
+              type: string
+            refresh_interval:
+              type: integer
+              minimum: 1
+            auth:
+              type: object
+              properties:
+                type:
+                  type: string
+                  enum:
+                    - bearer
+                token:
+                  type: string
+              required:
+                - type
+                - token
+              additionalProperties: false
+          required:
+            - type
+            - url
+          additionalProperties: false
+        - type: object
+          properties:
+            type:
+              type: string
+              const: prometheus_metrics
+            url:
+              type: string
+            query:
+              type: string
+            refresh_interval:
+              type: integer
+              minimum: 1
+              description: "Refresh interval in seconds"
+          required:
+            - type
+            - url
+            - query
+          additionalProperties: false
+        - type: object
+          properties:
+            type:
+              type: string
+              const: digitalocean_pricing
+            refresh_interval:
+              type: integer
+              minimum: 1
+              description: "Refresh interval in seconds"
+            model_aliases:
+              type: object
+              description: "Map DO catalog keys (lowercase(creator)/model_id) to Plano model names used in routing_preferences. Example: 'openai/openai-gpt-oss-120b: openai/gpt-4o'"
+              additionalProperties:
+                type: string
+          required:
+            - type
+          additionalProperties: false
+
 additionalProperties: false
 required:
   - version

config/validate_plano_config.sh

@@ -1,6 +1,16 @@
 #!/bin/bash
 
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+REPO_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
+CLI_DIR="$REPO_ROOT/cli"
+
+# Use uv run if available and cli/ has a pyproject.toml, otherwise fall back to bare python
+if command -v uv &> /dev/null && [ -f "$CLI_DIR/pyproject.toml" ]; then
+  PYTHON_CMD="uv run --directory $CLI_DIR python"
+else
+  PYTHON_CMD="python"
+fi
+
 failed_files=()
 
 for file in $(find . -name config.yaml -o -name plano_config_full_reference.yaml); do
@@ -14,7 +24,7 @@ for file in $(find . -name config.yaml -o -name plano_config_full_reference.yaml
   ENVOY_CONFIG_TEMPLATE_FILE="envoy.template.yaml" \
   PLANO_CONFIG_FILE_RENDERED="$rendered_file" \
   ENVOY_CONFIG_FILE_RENDERED="/dev/null" \
-  python -m planoai.config_generator 2>&1 > /dev/null
+  $PYTHON_CMD -m planoai.config_generator 2>&1 > /dev/null
 
   if [ $? -ne 0 ]; then
     echo "Validation failed for $file"