Example curation support

Dockerfile

@@ -19,7 +19,7 @@ RUN pip install \
     requests \
     google-generativeai \
     PyYAML \
-    adhoc-api~=1.0.0 \
+    adhoc-api~=2.0.2 \
     idc-index \
     seaborn \
     biopython \

pyproject.toml

@@ -30,7 +30,7 @@ dependencies = [
   "requests",
   "google-generativeai",
   "PyYAML",
-  "adhoc-api~=1.0.0",
+  "adhoc-api~=2.0.2",
   "idc-index",
   "seaborn",
   "biopython",

src/biome/agent.py

@@ -18,6 +18,7 @@
 from pathlib import Path
 from adhoc_api.tool import AdhocApi
 from adhoc_api.loader import load_yaml_api
+from adhoc_api.uaii import gpt_4o, o3_mini, claude_35_sonnet, gemini_pro
 
 logger = logging.getLogger(__name__)
 
@@ -82,8 +83,10 @@ def __init__(self, context: BaseContext = None, tools: list = None, **kwargs):
 
         # Note: not all providers support ttl_seconds
         ttl_seconds = 1800
-        drafter_config_gemini={'provider': 'google', 'model': 'gemini-1.5-pro-001', 'ttl_seconds': ttl_seconds, 'api_key': os.environ.get("GEMINI_API_KEY", "")}
-        drafter_config_anthropic={'provider': 'anthropic', 'model': 'claude-3-5-sonnet-latest', 'api_key': os.environ.get("ANTHROPIC_API_KEY")}
+        drafter_config_gemini =    {**gemini_pro, 'ttl_seconds': ttl_seconds, 'api_key': os.environ.get("GEMINI_API_KEY", "")}
+        drafter_config_anthropic = {**claude_35_sonnet, 'api_key': os.environ.get("ANTHROPIC_API_KEY")}
+        curator_config =           {**o3_mini, 'api_key': os.environ.get("OPENAI_API_KEY")}
+        contextualizer_config =    {**gpt_4o, 'api_key': os.environ.get("OPENAI_API_KEY")}
         specs = self.api_specs
 
         instructions_dir = os.path.join(self.root_folder, 'instructions')
@@ -102,7 +105,13 @@ def __init__(self, context: BaseContext = None, tools: list = None, **kwargs):
         self.logger = MessageLogger(self.context)
 
         try:
-            self.api = AdhocApi(logger=logger, drafter_config=[drafter_config_anthropic, drafter_config_gemini], apis=specs)
+            self.api = AdhocApi(
+                apis=specs,
+                drafter_config=[drafter_config_anthropic, drafter_config_gemini], 
+                curator_config=curator_config,
+                contextualizer_config=contextualizer_config,
+                logger=logger,
+            )
         except ValueError as e:
             self.add_context(f"The APIs failed to load for this reason: {str(e)}. Please inform the user immediately.")
             self.api = None

src/biome/api_agent.yaml

@@ -4,7 +4,7 @@ drafter:
     ttl_seconds: 1800
 finalizer:
     model: gpt-4o
-default_cache_body: | 
+default_cache_body: !fill | 
     You will be given the entire API documentation. 
     When you write code against this API, you should avail yourself of the appropriate query parameters, 
     your understanding of the response model, and be cognizant that not all data is public and thus may require a token, etc.

src/biome/api_definitions/cbioportal/api.yaml

@@ -10,11 +10,11 @@ description: |
     Institute, Princess Margaret Cancer Centre in Toronto, Children's 
     Hospital of Philadelphia, Caris Life Sciences, The Hyve and SE4BIO 
     in the Netherlands, and Bilkent University in Ankara, Turkey.
-raw_documentation: !load documentation/cbioportal.json
-examples: !load documentation/examples.md
+raw_documentation: !load_txt documentation/cbioportal.json
+examples: !load_yaml documentation/examples.yaml
 cache_body: 
     default: true 
-documentation: |
+documentation: !fill |
     {raw_documentation}
 
     # Additional Instructions:
@@ -24,7 +24,4 @@ documentation: |
     That is the base URL for all requests. There are not very many studies (less than 1000) 
     so if the user is searching for a specific study, you can just fetch all studies
     and then filter the list yourself since cbioportals filters are tricky. It is often most
-    productive to simply filter based on the study name. 
-
-    Here are some examples of how to use the API:
-    {examples}
+    productive to simply filter based on the study name.