topoteretes--cognee/cognee/tasks/codingagents/README.md

Coding Agents

Overview

The coding agents module provides utilities for extracting developer coding rules and best practices from text and associating them with their original sources in Cognee’s knowledge graph. It uses LLM-powered structured extraction to identify rules from conversations, documentation, or commit messages.

Note

This module runs automatically in cognee.memify() (the enrichment pipeline), but is not enabled by default in the standard cognee.cognify() pipeline.

Pipeline Position: ingestion → graph extraction → coding rule association → storage / indexing

Components

Functions

Function	Description
add_rule_associations(data, rules_nodeset_name, ...)	Extracts rules via LLM from the data, adds them to the graph, and creates source links.
get_existing_rules(rules_nodeset_name)	Retrieves existing rules from the graph for a specific nodeset
get_origin_edges(data, rules)	Searches for the original DocumentChunk that matches the input data and creates rule_associated_from edges linking the new Rule nodes to that source chunk.

Data Models (extend DataPoint)

Rule

Represents a single extracted developer rule.

Field	Type	Description
text	str	The coding rule text content.
belongs_to_set	NodeSet	Reference to the parent NodeSet (e.g., "coding_agent_rules").
metadata	dict	Indexing configuration (indexes rule field).

RuleSet

A collection of rules extracted in a single pass.

Field	Type	Description
rules	List[Rule]	List of extracted Rule objects.

Usage

Automatic (via Memify)

The memify pipeline includes rule associations by default.

import cognee
from cognee.tasks.codingagents.coding_rule_associations import get_existing_rules


await cognee.add(["agent.md"])# Add data (text or file paths)
await cognee.cognify() # Create Knowledge Graph
await cognee.memify()# Enrich Graph (Extract Rules automatically)

rules = await get_existing_rules("coding_agent_rules")
if rules:
    for rule in rules:
        print(f"{rule}")

Manual Rule Association

You can run the task directly on specific data.

from cognee.tasks.codingagents.coding_rule_associations import add_rule_associations

await add_rule_associations(
    data="Always use type hints in Python functions.",
    rules_nodeset_name="coding_agent_rules"
)

Retrieval

from cognee.tasks.codingagents.coding_rule_associations import get_existing_rules

rules = await get_existing_rules("coding_agent_rules")
for rule in rules:
    print(f"{rule}")

Advanced: Manual Graph Construction

You can manually create Rule objects and link them to content using get_origin_edges if you want to bypass the LLM extraction.

from cognee.tasks.codingagents.coding_rule_associations import Rule, get_origin_edges

# 1. Define your rule (the abstract guideline)
rule = Rule(text="Use snake_case for function names.")

# 2. Link it to the source (the text that implies the rule)
# 'data' is used to find the original document chunk in the graph
edges = await get_origin_edges(
    data="We strictly follow PEP8. Function names must use snake_case.",
    rules=[rule]
)

Configuration

Environment Variables:

Variable	Description
LLM_API_KEY	API key for LLM provider (required)
LLM_PROVIDER	Provider name (default: openai)
LLM_MODEL	Model name

Dependencies

Internal: cognee.infrastructure.databases.graph, cognee.infrastructure.databases.vector, cognee.infrastructure.llm, cognee.modules.engine.models

External: pydantic, LLM provider

Related

• cognee docs | Ingestion | Graph