[Improvement] Putting all the future data into one object

examples/portfolio/agents/advisor_agent.py

@@ -3,21 +3,18 @@
 # Final stage. Turns the computed portfolio metrics and risk figures into a
 # short, plain-English briefing using a small, cheap model on AWS Bedrock
 # (Converse API), called via ventis.llm.bedrock so token/cost telemetry gets
-# recorded onto this execution's future:<future_id>:metrics hash. Configure
+# recorded onto this execution's future:<future_id> hash. Configure
 # with env vars:
 #   BEDROCK_MODEL_ID  (default: meta.llama3-8b-instruct-v1:0)
 #   AWS_REGION        (default: us-east-1)
 #
 # If the LLM is unavailable (returns an empty string), it falls back to a
 # deterministic templated summary so the pipeline still returns.
 #
-# Resource profile: cheap CPU; the LLM cost sits in LLMAgent, not here.
+# Resource profile: cheap CPU; the LLM cost sits in the Bedrock call, not here.
 
-import sys
 import os
 
-sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "stubs"))
-from llm_agent import LLMAgent
 try:
     from ventis.llm.bedrock import call_bedrock
 except ImportError:
@@ -27,16 +24,14 @@
 class AdvisorAgent(object):
     def __init__(self):
         self.tools = [self.summarize]
-        self.llm = LLMAgent()
+        self.model_id = os.environ.get(
+            "BEDROCK_MODEL_ID", "meta.llama3-8b-instruct-v1:0"
+        )
+        self.region = os.environ.get("AWS_REGION", "us-east-1")
 
     def summarize(self, holdings: dict, metrics: dict, risk: dict) -> str:
         """Write a short plain-English briefing on the portfolio."""
         prompt = self._build_prompt(holdings, metrics, risk)
-        text = self.llm.complete(
-            prompt=prompt, max_tokens=400, temperature=0.2
-        ).value()
-        if not text:
-            print("AdvisorAgent: LLM returned no output; using templated summary.")
         try:
             response = call_bedrock(
                 model_id=self.model_id,
@@ -48,7 +43,6 @@ def summarize(self, holdings: dict, metrics: dict, risk: dict) -> str:
         except Exception as e:
             print(f"AdvisorAgent: Bedrock call failed ({e}); using templated summary.")
             return self._fallback_summary(metrics, risk)
-        return text
 
     def _build_prompt(self, holdings: dict, metrics: dict, risk: dict) -> str:
         lines = ["You are a portfolio analyst. Given the figures below, write a "

examples/portfolio/agents/intent_agent.py

@@ -7,20 +7,9 @@
 #      ->  {"holdings": {"AAPL": 0.4, "MSFT": 0.35, "NVDA": 0.25},
 #           "lookback_days": 180}
 #
-<<<<<<< HEAD
-# The actual model call lives in the shared LLMAgent (remote, resolved via
-# .value()) — this agent only builds the prompt and parses the result, so no
-# Bedrock boilerplate lives here. If the LLM is unavailable or returns
-# unparseable output, parse() raises: there is no fallback, the request fails
-# loudly rather than guessing at the holdings. Weights are renormalized to 1.0.
-#
-# Resource profile: cheap CPU; the LLM cost sits in LLMAgent, not here.
-
-import sys
-=======
 # Calls AWS Bedrock (Converse API) via ventis.llm.bedrock -- same pattern as
 # AdvisorAgent -- so token/cost telemetry gets recorded onto this execution's
-# future:<future_id>:metrics hash. Configure with env vars:
+# future:<future_id> hash. Configure with env vars:
 #   BEDROCK_MODEL_ID  (default: meta.llama3-8b-instruct-v1:0)
 #   AWS_REGION        (default: us-east-1)
 #
@@ -31,36 +20,21 @@
 # Resource profile: cheap CPU, single call per request, on the critical path
 # before the fan-out.
 
->>>>>>> remotes/origin/telemetry-signals
 import os
 import re
 import json
 
-<<<<<<< HEAD
-sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "stubs"))
-from llm_agent import LLMAgent
-=======
 try:
     from ventis.llm.bedrock import call_bedrock
 except ImportError:
     from bedrock import call_bedrock
->>>>>>> remotes/origin/telemetry-signals
 
 DEFAULT_LOOKBACK_DAYS = 365
 
 
 class IntentAgent(object):
     def __init__(self):
         self.tools = [self.parse]
-<<<<<<< HEAD
-        self.llm = LLMAgent()
-
-    def parse(self, query: str) -> dict:
-        """Parse a natural-language portfolio request into holdings + lookback."""
-        text = self.llm.complete(
-            prompt=self._build_prompt(query), max_tokens=300, temperature=0.0
-        ).value()
-=======
         self.model_id = os.environ.get(
             "BEDROCK_MODEL_ID", "meta.llama3-8b-instruct-v1:0"
         )
@@ -75,7 +49,6 @@ def parse(self, query: str) -> dict:
             region=self.region,
         )
         text = response["output"]["message"]["content"][0]["text"]
->>>>>>> remotes/origin/telemetry-signals
         if not text:
             raise ValueError("IntentAgent: LLM returned no output for the request.")
 
@@ -148,15 +121,7 @@ def _sanitize(self, parsed: dict) -> dict:
 
 
 if __name__ == "__main__":
-<<<<<<< HEAD
-    # Assumes the LLMAgent stub (Future-returning) is on the path, as it is
-    # inside the deployed pipeline.
-    agent = IntentAgent()
-    print(agent.parse(
-        "Analyze 40% Apple, 35% Microsoft and 25% Nvidia over the last 6 months"
-=======
     agent = IntentAgent()
     print(agent.parse(
         query="Analyze 40% Apple, 35% Microsoft and 25% Nvidia over the last 6 months"
->>>>>>> remotes/origin/telemetry-signals
     ))

examples/portfolio/config/global_controller.yaml

@@ -6,26 +6,6 @@
 # reflect each stage's real cost so the scheduler has placement decisions to make.
 
 agents:
-<<<<<<< HEAD
-  # Shared inference node. Owns all Bedrock plumbing; IntentAgent and
-  # AdvisorAgent delegate their model calls here. LLM-bound — scale replicas
-  # to match inference demand.
-  - name: LLMAgent
-    host: localhost
-    port: 8075
-    redis_port: 6379
-    replicas: 1
-    resources:
-      cpu: 1
-      memory: 512
-    entrypoint: agents/llm_agent.py
-
-  # Stage 0: parse the free-text request into structured holdings + lookback
-  # window (calls LLMAgent). Cheap CPU, one call per request, on the critical
-  # path before the fan-out.
-  - name: IntentAgent
-    host: localhost
-    port: 8076
   # Stage 0: parse the free-text request into structured holdings + lookback
   # window (calls Bedrock directly via ventis.llm.bedrock). Cheap CPU, one
   # call per request, on the critical path before the fan-out.
@@ -36,8 +16,6 @@ agents:
       cpu: 1
       memory: 256
     entrypoint: agents/intent_agent.py
-
-  # Stage 0: price history fetch. Network/IO-bound, cheap CPU. Called by
     provider: EC2
     instance_type: t3.micro
 

examples/portfolio/config/policy.yaml

@@ -14,7 +14,6 @@ rules:
   - match: {}
     access:
       - Workflow
-      - LLMAgent
       - IntentAgent
       - PriceAgent
       - MetricsAgent

examples/text2sql/agents/vllm_agent.py

@@ -3,7 +3,7 @@
 # LLM backend for SQL candidate generation, called remotely by
 # SQLGeneratorAgent. Calls AWS Bedrock (Converse API) via ventis.llm.bedrock
 # so token/cost telemetry gets recorded onto this execution's
-# future:<future_id>:metrics hash — same pattern as
+# future:<future_id> hash — same pattern as
 # examples/portfolio/agents/advisor_agent.py.
 # Configure with env vars:
 #   BEDROCK_MODEL_ID  (default: meta.llama3-8b-instruct-v1:0)