thedotmack/claude-mem
Claude Mem
Claude-mem is an agentic memory persistence system designed to provide AI assistants with long-term context across multiple development sessions. It functions as a background orchestrator that captures, summarizes, and indexes interaction history, allowing models to maintain continuity and recall technical decisions from past tasks. By utilizing a vector-augmented context engine, the system injects relevant historical observations into active sessions, ensuring that AI agents remain informed without exceeding finite token budgets.
The project distinguishes itself through an endless memory architecture that compresses tool observations into concise summaries, preventing context window exhaustion during extended workflows. It employs a multi-layered retrieval framework that enforces progressive disclosure, fetching compact indices before retrieving full details to optimize performance. Users can further refine this behavior through granular context filtering, custom model selection for processing, and the ability to route requests through unified API gateways to support various AI providers.
Beyond its core memory capabilities, the system includes a comprehensive suite of development and maintenance tools. It features a real-time dashboard for monitoring memory streams, automated diagnostics for system health, and utilities for managing database integrity. The infrastructure is built to handle intensive tasks asynchronously, ensuring that data capture and processing do not interfere with the responsiveness of the primary host application.
Features
- Agent Memory Persistence - Stores and retrieves historical session data to ensure AI assistants maintain continuity across development tasks.
- AI Memory Layers - Captures, summarizes, and indexes development session history for long-term context retrieval.
- Context Compression - Compresses tool observations into summaries to maintain linear context complexity.
- Agentic Orchestrators - Manages lifecycle hooks and worker processes for autonomous development assistants.
- Agentic Process Managers - Manages the lifecycle and health of autonomous AI worker processes to ensure reliable task execution.
- Context Injection - Injects historical session summaries and observations into the current context to provide relevant background.
- Vector Context Engines - Manages token-efficient memory access through progressive disclosure and vector-augmented retrieval.
- Agent Knowledge Bases - Enables the creation of filtered observation corpora to prime AI sessions with synthesized historical knowledge.
- Knowledge Retrieval Systems - Searches through past observations and documentation to recall technical details and previous problem-solving steps.
- Model Gateways - Routes requests through unified gateways to support multiple AI providers and interchangeable models.
- Observation Processing - Processes observations using an AI agent SDK that builds structured prompts and stores results.
- Session Management - Executes custom logic at specific points in a session lifecycle, such as starting, prompting, or ending.
- Workflow Orchestration - Enforces progressive disclosure by separating search indexing, chronological navigation, and detail fetching.
- Context Engineering - Provides techniques to curate token usage across turns to maximize model performance.
- Context Retrieval Systems - Fetches compact indices before full details to minimize token usage and maintain context efficiency.
- Context Window Management - Manages token usage by compressing historical data and selectively injecting relevant project information.
- Lifecycle Hooks - Hooks into session events to inject context, track prompts, and perform background cleanup.
- Retrieval Workflows - Executes a three-layer retrieval workflow to minimize token usage while fetching relevant details.
- Context Filters - Filters historical observations by type or concept to control information injected into sessions.
- Development History Automation - Automatically captures and summarizes development activity to help AI models understand project evolution.
- State Persistence - Manages session state by capturing identifiers and ensuring graceful cleanup to prevent data loss.
- Search Indices - Provides full-text search indices with support for project, type, and date range filtering.
- Asynchronous Task Runners - Executes hooks asynchronously by enqueuing tasks for background processing to maintain application responsiveness.
- Task Queuing Systems - Decouples data capture from intensive processing using a queue-based architecture to handle parallel events and retries.
- Process Management - Manages worker processes using standardized scripts for status checks, starting, stopping, and logging.
- AI Provider Integrations - Supports connecting to self-hosted or bridged API endpoints via base URL configuration.
- Context Generation Tools - Generates markdown summaries of project directories to provide AI models with historical development context.
- Model Selection Tools - Allows selection of specific models to balance processing speed, quality, and costs.
- Chronological Retrieval - Retrieves chronological context around memory observations to understand narrative arcs and relationships.
- Full-Text Search Engines - Performs fast full-text searches across observations and prompts using synchronized virtual tables.
- Task Queues - Decouples intensive tasks from the main flow to ensure responsiveness and handle retries.
- Service Discovery - Resolves service ports from configuration to facilitate communication between development environments and background processes.
- Data Sanitization - Strips sensitive or system-level tags from user prompts and tool data to ensure privacy.
- Health Monitoring - Monitors worker health and provides real-time updates via Server-Sent Events.
- Data Visualization Interfaces - Serves a web-based viewer UI that provides real-time memory stream visualization and pagination.
- Memory Dashboards - Provides a web-based interface for live visualization and administrative control over stored session data.
