How RecCall Works

Technical deep dive into RecCall's architecture and ML-powered context management

Universal Context System

RecCall uses a unified context model that supports three types of contexts:

Static Contexts

Manually created templates for project documentation, coding standards, and architectural decisions. Perfect for onboarding and reference.

Dynamic Contexts

Automatically generated from conversations using ML. Captures summaries, key topics, code references, and embeddings for semantic search.

Hybrid Contexts

Static templates enhanced with ML insights. Combine the structure of templates with the intelligence of dynamic context.

ML-Powered Intelligence

RecCall's ML layer automatically processes conversations to extract meaningful context artifacts.

🏗️ ML Architecture

The ML Intelligence Layer processes conversations through multiple pipelines to generate rich context artifacts.

Conversation Messages Input user/assistant Context Engine createFromConversation() enhanceContext() Orchestrates ML Pipeline 📝 Summarizer Extract Key Points Generate Summary 💻 Code Extractor Find Code Blocks Extract File Refs 🔗 Embedder Generate Vectors 384-dim Embeddings 🏷️ Topic Extractor Keyword Analysis Topic Identification ML Artifacts • embedding: number[384] • summary: string • topics: string[] • codeRefs: CodeRef[] Context Storage Filesystem + Index Search Engine Hybrid Search Input Orchestration Processing Consolidation Storage
📝

Summarization

Extracts key points and generates concise summaries from conversations, preserving important decisions and insights.

🔍

Topic Extraction

Identifies key topics and themes from conversations, enabling better organization and discovery.

💻

Code Reference Extraction

Finds code blocks, file references, and line ranges mentioned in conversations, creating traceable links.

🔗

Semantic Embeddings

Generates vector embeddings for semantic search, enabling intelligent context matching based on meaning.

Hybrid Search

RecCall combines keyword and semantic search to find the most relevant context quickly.

Keyword Search

Fast text-based search for exact matches, file names, and specific terms.

Semantic Search

AI-powered search that understands meaning and intent, finding related context even without exact keyword matches.

Architecture & Access Layers

RecCall provides multiple ways to access and manage context, fitting into any workflow.

MCP Server

Model Context Protocol integration for AI IDEs like Cursor. Automatic context injection.

Native

CLI

Command-line interface for terminal-based workflows. Full context management from command line.

Terminal

REST API

HTTP API for web applications and custom integrations. Express.js and Fastify adapters available.

Web

SDKs

JavaScript SDK for programmatic access. Python and Go SDKs coming soon.

SDK

Storage & Sync

File System (Default)

Local file-based storage with atomic writes and in-memory indexing. Fast and reliable for single-user scenarios.

Git Sync (Coming Soon)

Version-controlled context repositories. Share and sync contexts across teams using Git.

PostgreSQL & Redis (Optional)

Enterprise storage backends for scalable deployments. ACID compliance and high-performance caching.

Security & Privacy

🔒 Local-First

All context data stored locally by default. No data leaves your machine unless explicitly configured.

🔐 Encryption Ready

Storage backends support encryption at rest. Enterprise deployments can use encrypted databases.

👥 Access Control

Context-level permissions and sharing controls. Manage who has access to which contexts.

Ready to Try RecCall?

Get started in 2 minutes and see how it works.