Skip to content

agent_advisor

agent_advisor

LM-guided agent restructuring — analyzes traces and suggests improvements.

Classes

AgentAdvisorPolicy 

AgentAdvisorPolicy(*, advisor_engine: Any = None, advisor_model: str = '', max_traces: int = 50)

Bases: AgentLearningPolicy

Higher-level LM analyzes traces, suggests agent structure changes.

Does NOT auto-apply changes — returns recommendations that can be reviewed or applied via config.

Source code in src/openjarvis/learning/agent_advisor.py 
22
23
24
25
26
27
28
29
30
31
def __init__(
    self,
    *,
    advisor_engine: Any = None,
    advisor_model: str = "",
    max_traces: int = 50,
) -> None:
    self._advisor_engine = advisor_engine
    self._advisor_model = advisor_model
    self._max_traces = max_traces
Functions
update 
update(trace_store: Any, **kwargs: object) -> Dict[str, Any]

Analyze traces and return agent improvement recommendations.

Source code in src/openjarvis/learning/agent_advisor.py 
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
def update(self, trace_store: Any, **kwargs: object) -> Dict[str, Any]:
    """Analyze traces and return agent improvement recommendations."""
    try:
        traces = trace_store.list_traces()
    except Exception as exc:
        logger.warning("Agent advisor analysis failed: %s", exc)
        return {"recommendations": [], "confidence": 0.0}

    # Collect failing or slow traces
    problem_traces = []
    for trace in traces[-self._max_traces :]:
        is_failing = trace.outcome != "success"
        is_slow = (trace.total_latency_seconds or 0) > 5.0
        if is_failing or is_slow:
            problem_traces.append(trace)

    if not problem_traces:
        return {
            "recommendations": [],
            "confidence": 1.0,
            "message": "No problematic traces found",
        }

    # Analyze patterns without LM (structural analysis)
    recommendations = self._analyze_patterns(problem_traces)

    # If advisor engine available, get LM-guided recommendations
    if self._advisor_engine and self._advisor_model:
        try:
            lm_recs = self._get_lm_recommendations(problem_traces)
            recommendations.extend(lm_recs)
        except Exception as exc:
            logger.debug("Failed to parse agent advisor recommendation: %s", exc)

    confidence = 1.0 - (len(problem_traces) / max(len(traces), 1))
    return {
        "recommendations": recommendations,
        "confidence": round(confidence, 2),
        "analyzed_traces": len(traces),
        "problem_traces": len(problem_traces),
    }