Skip to main content

Intelligence Layer

The Intelligence Layer sits above the core Context/Pattern API and adds automatic reasoning — it analyzes your question, selects the right cognitive patterns, composes multi-pattern prompts, and orchestrates complex workflows.

from mycontext.intelligence import (
    transform,                  # Auto pattern selection → Context
    suggest_patterns,           # Recommend patterns for a question
    suggest_routes,             # Multi-route agent pipelines (v0.10+)
    smart_execute,              # Auto-route + execute in one call
    smart_prompt,               # Auto-route + compose optimized prompt
    smart_generic_prompt,       # Zero-cost auto-route + compile
    build_workflow_chain,       # Deprecated — prefer suggest_routes
    TemplateIntegratorAgent,    # Fuse multiple templates into one
    PromptComposer,             # Merge template prompts
    QualityMetrics,             # Measure context quality
    ContextAmplificationIndex,  # CAI quality signal
    PromptArchitect,            # Parse → score → rewrite any raw prompt
    GuidanceOptimizer,          # Upgrade Guidance rules to binding language
    get_criteria,               # Pre-built DeepEval GEval criteria bundles
)

The Core Intelligence Functions​

FunctionWhat it doesLLM calls
transform()Analyze input → auto-select pattern → return Context0
suggest_patterns()Suggest best patterns + chain order0 (keyword) or 1 (hybrid/llm)
suggest_routes()Multiple differentiated routes + agent steps (receives / produces)1
smart_execute()Route → execute → return response2 (assess + execute)
smart_prompt()Route → compose optimized prompt2–4
smart_generic_prompt()Route → compile generic prompt1 (assess only)
build_workflow_chain()Deprecated — single-chain workflow result1+ (may delegate to suggest_routes)
generate_context()LLM generates a full Context from role + goal1

Architecture Overview​

Three Complexity Tiers​

smart_execute() and smart_prompt() automatically route questions into three tiers:

Tier 1: Raw​

Simple, well-known questions answered directly. No templates needed.

"What is Python?" → Raw LLM call

Tier 2: Single Template​

Specialized questions where one pattern adds clear value.

"Review this code for security issues" → CodeReviewer template

Tier 3: Integrated​

Multi-domain, complex questions requiring multiple frameworks fused together.

"Should we migrate to microservices given our 10-year-old monolith?" 
→ DecisionFramework + RiskAssessor + TechnicalTranslator (integrated)

Quick Start​

from mycontext.intelligence import smart_execute, suggest_patterns, transform

# Option 1: Let it figure everything out (2 LLM calls)
response, meta = smart_execute(
    "Why did our conversion rate drop 30% after the latest release?",
    provider="openai",
)
print(response)
print(f"Mode: {meta['mode']}, Templates: {meta['templates_used']}")

# Option 2: See recommendations first (0 LLM calls)
result = suggest_patterns(
    "Why did our conversion rate drop 30%?",
    mode="keyword",
)
print(result.suggested_chain)
# → ['root_cause_analyzer', 'data_analyzer']
print(result.to_markdown())

# Option 3: Transform to Context (0 LLM calls)
ctx = transform("Why did our conversion rate drop 30%?")
print(ctx.metadata["transformation_metadata"]["patterns_applied"])
# → ['root_cause_analyzer']
result = ctx.execute(provider="openai")

All Exports​

from mycontext.intelligence import (
    # Core functions
    transform,
    suggest_patterns,
    suggest_routes,
    smart_execute,
    smart_prompt,
    smart_generic_prompt,
    build_workflow_chain,

    # Route types (suggest_routes)
    RouteAnalysis,
    AnalysisRoute,
    RouteStep,

    # Classes
    TransformationEngine,
    TemplateIntegratorAgent,
    PromptComposer,
    QualityMetrics,
    ContextAmplificationIndex,
    OutputEvaluator,
    TemplateBenchmark,

    # Data classes
    InputAnalysis,
    SuggestionResult,
    PatternSuggestion,
    WorkflowChainResult,
    IntegrationResult,
    ComposedPrompt,
    QualityScore,
    CAIResult,
    ComplexityResult,

    # Enums
    InputType,
    ComplexityLevel,
    QualityDimension,

    # Utilities
    get_pattern_class,
    get_generic_prompt_for,
    assess_complexity,
    PATTERN_BUILD_CONTEXT_REGISTRY,
    FULL_PATTERN_CATALOG,
    VALID_PATTERN_NAMES,
)

Prompt Quality Tools​

These tools live in the Intelligence Layer and focus on improving the quality of prompts and template rules — before execution, not just after.

ToolInputWhat it doesLLM call
PromptArchitectAny raw prompt stringParse sections → score → rewrite weak/missing → diff0 (parse) or 1 (build/improve)
GuidanceOptimizerA Guidance objectAudit rules for suggestive modals + vague directives → rewrite only weak ones0 (audit) or 1 (optimize)
QualityMetricsAny ContextScore on 6 dimensions: clarity, completeness, specificity, relevance, structure, efficiency0
OutputEvaluatorLLM output + ContextScore on 5 dimensions: instruction_following, reasoning_depth, actionability, structure_compliance, cognitive_scaffolding0
get_criteria()Bundle nameReturn pre-built DeepEval GEval rubrics0

Choosing the Right Entry Point​

Your situationUse
I want to just get an answersmart_execute()
I want to see what patterns are recommendedsuggest_patterns()
I want a prompt string (not execution)smart_prompt() or smart_generic_prompt()
I want to control which pattern runspattern.execute() directly
I want multiple analytical angles / agent pipelinessuggest_routes()
I want a legacy single-chain planbuild_workflow_chain() (deprecated)
I want to merge multiple templatesTemplateIntegratorAgent
I have a raw prompt and want it upgradedPromptArchitect.improve()
I want to audit/rewrite template rulesGuidanceOptimizer.optimize()
I want to measure prompt qualityQualityMetrics.evaluate()
I want to measure output qualityOutputEvaluator.evaluate()
I want to measure amplificationContextAmplificationIndex.compute()

Next Steps​