mycontext Architecture

context engineering pipeline — from pattern to response

The core pipeline: a Pattern builds a structured Context, which assembles into a prompt and executes via any LLM provider. The same Context can be exported to LangChain, CrewAI, LlamaIndex, or any framework.

scroll to zoom · drag to pan · click ⛶ to expand

100%

flowchart TD
    subgraph input_layer [Input Layer]
        P["Pattern
88 cognitive templates"]
        BC["build_context()
+ output_format"]
    end

    subgraph context_layer [Context Object]
        CTX["Context
guidance + directive
+ constraints + data"]
        ASM["assemble()
research-backed
prompt flow"]
    end

    subgraph export_layer [Export Layer]
        LC["to_langchain()"]
        LI["to_llamaindex()"]
        CR["to_crewai()"]
        OAI["to_openai()"]
        MK["to_markdown()"]
        JS["to_json()"]
    end

    subgraph exec_layer [Execution Layer]
        EX["execute()
provider + model"]
        LP["LiteLLMProvider
OpenAI, Anthropic,
Google, etc."]
    end

    subgraph output_layer [Response]
        PR["ProviderResponse
.response (string)
.usage, .metadata"]
    end

    P -->|"problem, depth,
output_format"| BC
    BC --> CTX
    CTX --> ASM
    ASM --> EX
    ASM --> LC
    ASM --> LI
    ASM --> CR
    ASM --> OAI
    ASM --> MK
    ASM --> JS
    EX --> LP
    LP --> PR

Input

Core

Export / Output

input

Pattern

88 cognitive templates. Each defines a research-backed methodology: RootCauseAnalyzer, DataAnalyzer, CodeReviewer, etc.

input

build_context()

Assembles guidance, directive, constraints, and data into a Context object. Accepts output_format for presentation hints.

core

Context

The structured prompt carrier. Contains guidance (role/rules), directive (task), constraints (guardrails), and data (user inputs). Research-backed assembly order.

core

assemble()

Renders the Context into a final prompt string following primacy/recency research: role first, task last, reasoning in the middle.

export

Export Methods

to_langchain(), to_llamaindex(), to_crewai(), to_openai(), to_markdown(), to_json() — ship the same Context anywhere.

output

ProviderResponse

The LLM's reply: .response (text), .usage (tokens), .metadata (model, latency). Always a string — format depends on the prompt hint.

Three lines to insight: analyzer = DataAnalyzer() → ctx = analyzer.build_context(data_description=..., goal=..., intent="executive") → resp = ctx.execute(provider="openai")