Advanced

Advanced topics for power users and contributors.

Topics

Writing Adapters

Create custom adapters for new frameworks.

Learn

Vector Backends

Configure and optimize vector databases.

Configure

Cloud Embeddings

Use cloud embedding providers.

Set up

Agent Integration

Integrate SpecMem with AI agents.

Integrate

Agent Memory Patterns

Design layered memory for coding agents.

Review patterns

Kiro Powers

SpecMem is available as a Kiro Power for seamless integration with Kiro IDE. Install it from the Powers panel or configure manually.

Architecture Deep Dive

graph TB
    subgraph Input
        A1[Kiro Specs]
        A2[Agent Guidance]
        A3[Claude.md]
        A4[Custom Specs]
    end

    subgraph Adapters
        B1[KiroAdapter]
        B2[CursorAdapter]
        B3[ClaudeAdapter]
        B4[CustomAdapter]
    end

    subgraph Core
        C1[SpecIR]
        C2[Memory Bank]
        C3[Impact Graph]
        C4[SpecDiff]
        C5[Validator]
    end

    subgraph Storage
        D1[LanceDB]
        D2[ChromaDB]
        D3[Qdrant]
    end

    subgraph Output
        E1[CLI]
        E2[Python API]
        E3[Web UI]
        E4[Agent Pack]
    end

    A1 --> B1
    A2 --> B2
    A3 --> B3
    A4 --> B4

    B1 --> C1
    B2 --> C1
    B3 --> C1
    B4 --> C1

    C1 --> C2
    C1 --> C3
    C1 --> C4
    C1 --> C5

    C2 --> D1
    C2 --> D2
    C2 --> D3

    C2 --> E1
    C2 --> E2
    C2 --> E3
    C3 --> E4

Performance Tuning

Embedding Batch Size

[embedding]
batch_size = 64  # Increase for faster indexing

Vector Index Settings

[vectordb]
# LanceDB specific
index_type = "IVF_PQ"
num_partitions = 256
num_sub_vectors = 96

Memory Usage

[memory]
# Limit in-memory cache
max_cache_size = 1000
# Enable disk-based storage
use_mmap = true

Extending SpecMem

Custom Validation Rules

from specmem.validator import ValidationRule, register_rule

@register_rule
class MyRule(ValidationRule):
    name = "my_rule"

    def validate(self, spec):
        # Custom validation logic
        pass

Custom Embedding Providers

from specmem.vectordb.embeddings import EmbeddingProvider, register_provider

@register_provider("custom")
class CustomEmbeddings(EmbeddingProvider):
    def embed(self, text: str) -> list[float]:
        # Custom embedding logic
        pass

Hooks and Events

from specmem import on_scan, on_build

@on_scan
def my_scan_hook(result):
    print(f"Scanned {result.count} specs")

@on_build
def my_build_hook(result):
    print(f"Built pack at {result.path}")