TiDB is a production-grade, open-source distributed SQL database with MySQL compatibility, horizontal scalability, strong consistency (ACID), and HTAP capabilities via TiKV (row store) and TiFlash (columnar store). The architecture separates compute (TiDB Server) from storage (TiKV/TiFlash) with PD for cluster management. Key entry points: cmd/tidb-server (SQL layer), lightning/cmd/tidb-lightning (TB-scale import), br/cmd/br (backup/restore), dumpling/cmd/dumpling (export). The codebase demonstrates exceptional craftsmanship with a modular design: the parser (pkg/parser) is a standalone MySQL-compatible SQL parser used by external projects; pkg/objstore provides a unified storage abstraction (S3/GCS/Azure/local/HDFS) with compression/encryption; pkg/dxf/framework implements a distributed execution framework with scheduler/task-executor/planner; pkg/timer provides distributed cron/interval scheduling with persistence; pkg/extension enables plugin-based authentication/session-hooks/custom-functions; pkg/util/chunk/arena/zeropool/mvmap/intset/bitmap provide high-performance columnar processing and memory management; pkg/expression uses code generation for vectorized SIMD evaluation; pkg/planner implements a Cascades-based cost-based optimizer with plan caching; pkg/ddl executes online schema changes via job-based distributed backfill with schema versioning; pkg/statistics maintains histograms/CMS/FM-sketch with auto-analyze priority queues. The repo includes comprehensive testing (unit, integration, realtikv, chaos) and modern Go practices (generics, interfaces, DI).
hardened_malloc is a security-focused replacement for the C heap allocator, compiled as a shared library and used either via LD_PRELOAD or integrated directly into a libc such as Bionic on GrapheneOS. It implements the full malloc family (malloc/calloc/realloc/free, posix_memalign/aligned_alloc, malloc_usable_size) plus hardened extensions such as malloc_object_size, free_sized, and free_aligned_sized, along with C++ operator new/delete wrappers. Its architecture is a single monolithic C library: all mutable allocator state lives in one read-only-after-init metadata region; slab allocations are partitioned into per-size-class regions inside statically reserved address space, and large allocations are mmap-backed with randomized guard pages and a quarantine. It targets 64-bit Linux and currently builds against Bionic, glibc, and musl.
Understand Anything is a developer tool that analyzes codebases to create interactive knowledge graphs for code comprehension and onboarding. It combines deterministic static analysis (using tree-sitter) with LLM-powered semantic understanding to build explorable graphs of files, functions, classes, dependencies, and business domains. The system features a multi-agent pipeline for analysis, incremental updates via fingerprinting, and a React/Vite dashboard for visualization with guided tours, semantic search, and diff impact analysis. The architecture follows a monorepo structure with pnpm workspaces separating core analysis engine, dashboard UI, and plugin interface. Key capabilities include language-agnostic structural analysis across 30+ programming languages, business domain extraction, knowledge base analysis (Obsidian/Logseq-style wikis), and multi-platform support for AI coding assistants (Claude Code, Codex, Cursor, Copilot). The system operates offline for structural analysis while optionally leveraging LLMs for enrichment, with persistence to a .ua/ directory for team collaboration via version control.
Mem0 is a production-grade memory layer for AI agents, providing persistent, searchable, and updatable memory across sessions. It operates as both an open-source Python/TypeScript SDK and a hosted platform (api.mem0.ai). The core library abstracts away LLM providers (OpenAI, Anthropic, Groq, vLLM, etc.), embedding models, vector stores (over 20 backends including Qdrant, Pinecone, PGVector, Chroma, Weaviate, and many more), and rerankers behind a consistent config-driven interface. Memory operations (add, search, update, delete) support scoping by user, agent, session, or run, and leverage hybrid search combining semantic, BM25, and entity-boost signals with configurable reranking. The repository is architected as a monorepo with a clear separation of concerns: the core SDK packages (`mem0/` for Python, `mem0-ts/` for TypeScript), a self-hosted FastAPI server with a Next.js dashboard, CLI tools in both Python and Node.js, and a comprehensive plugin ecosystem for AI coding agents (Claude Code, Cursor, Codex, Kimi, Antigravity, Pi Agent, OpenClaw, etc.). Each integration follows a uniform hook adapter pattern, translating native agent events into memory operations via MCP servers or shared core modules. The platform backend includes authentication, API key management, telemetry via PostHog, and a webhook system for event-driven workflows. Key data flows: An application calls `memory.add()` with messages and metadata → the SDK extracts facts via an LLM → embeds them using the configured embedder → stores vectors and metadata in the chosen vector store. On `memory.search()`, the query is embedded, hybrid-scored against stored vectors (with optional reranking), and results are returned with relevance scores. The system supports temporal reasoning (timestamp filtering, memory decay), entity-based partitioning, custom categories, and feedback mechanisms to refine future retrievals. External dependencies include LLM APIs, vector databases, embedding services, and optionally Redis/SQLite for local history.
OpenViking is an agent-native context database that unifies memories, resources, and skills under a virtual filesystem paradigm (viking:// URIs). It provides a three-tier context model (L0 abstract, L1 overview, L2 full content), recursive directory-based retrieval with visualized trajectories, and automatic session memory extraction. The system consists of a FastAPI HTTP server with embedded Rust components (RAGFS filesystem, vector index engine), a Rust CLI (ov), Python SDKs, a React/Vite Web Studio dashboard, and VikingBot - an agent framework with multi-channel support (Telegram, Slack, Feishu, etc.) and MCP integration. It integrates with OpenClaw, Claude Code, Codex, OpenCode, Hermes, and LangChain/LangGraph. The codebase demonstrates strong engineering: clean separation of concerns, async-first design, comprehensive test coverage (unit, integration, e2e), and production-ready deployment artifacts (Docker, Helm, multi-arch builds). Benchmarks show significant improvements over baseline agents on LoCoMo (82% vs 24-57% accuracy), tau2-bench (+6-12pp), and HotpotQA (91% at top-20).
Memlab is a comprehensive JavaScript memory leak detection and heap analysis framework developed by Meta. It provides a CLI, a programmatic API, and an MCP (Model Context Protocol) server for AI-assisted debugging. The system captures heap snapshots from Chromium (via Puppeteer), Node.js, Electron.js, and Hermes, then parses and analyzes them to identify memory leaks, optimization opportunities, and retention patterns. It includes specialized detectors for detached DOM elements, React component leaks, event listener accumulation, duplicate strings, and global variable bloat. The architecture is a monorepo with packages for core parsing and algorithms, E2E browser automation, heap analysis plugins, a public API, a CLI, an in-browser visualizer (MemLens), and an MCP server that exposes over 100 analysis tools. Key data flows involve warmup, browser interaction, snapshot capture, parsing, leak filtering, trace clustering (heuristic and ML-based), and report generation. External dependencies include Puppeteer, Chromium, and the MCP SDK.
OpenRath is a framework for building LLM-powered AI agent applications with a torch-like API design. It centers on the Session abstraction (analogous to a Tensor) that flows through Sandboxes (execution environments), Memories (persistent state), and Tools (callable operations), orchestrated by Agents and Workflows. The framework provides session lineage tracking, sandboxed tool execution, memory persistence, and LLM provider abstraction, enabling the creation of multi-agent workflows with durable state and traceable provenance. The architecture is modular and extensible, featuring pluggable backends for sandboxes (local, OpenSandbox) and memory systems (local, OpenViking), a registry pattern for dynamic backend loading, and lazy evaluation of sessions for performance. Key innovations include write-ahead logging for crash-resistant session persistence, resource-keyed parallelism for safe concurrent tool execution, and unified garbage collection across session, sandbox, and memory domains. The framework is designed for production use, with extensive testing, clear documentation, and active maintenance.
EverOS is a local-first, markdown-based memory framework for AI agents and chat applications. It ingests conversational sessions via a FastAPI REST API, extracts structured memories (episodes, atomic facts, agent cases, skills, user profiles) using LLMs, persists them as version-controllable Markdown files, and maintains derived search indexes in LanceDB (vector + BM25) and SQLite. The system provides a rich retrieval layer with keyword, vector, hybrid (RRF), and agentic (LLM-guided multi-round) search methods, plus a knowledge wiki feature. An Offline Memory Engine (OME) runs background strategies (reflection, clustering, extraction) on cron/idle/immediate triggers, and a cascade daemon watches the Markdown filesystem to reconcile indexes. The architecture is a layered monolith (entrypoints → service → memory → infrastructure) with cross-cutting components for LLM, embedding, rerank, tokenizer, and config. It ships with a CLI (Typer), a TUI demo (Textual), and extensive integration tests. While primarily an application, it includes a benchmark harness (LoCoMo, LongMemEval, EverMemBench) and evaluation fixtures, giving it a hybrid character with a strong research-oriented testing pipeline.
MemPalace is a local-first, persistent memory system for AI agents. It stores and retrieves structured 'drawers' (memory chunks) organized into wings, rooms, and closets, with hybrid vector+BM25 search, a temporal knowledge graph, and agent coordination via a logstream. The system provides a CLI, an MCP server (both full and lightweight), and shell hooks for integration with Claude Code, Cursor, Codex, Antigravity, and DeepSeek Harness. It supports multiple storage backends (ChromaDB, SQLite exact, pgvector, Qdrant, Milvus, and a Rust native backend) and offers privacy-first operation with no mandatory API keys or internet connectivity. Architecturally, MemPalace is a modular Python package with optional Rust acceleration (PyO3). The core is split into several packages: `mempalace.cli` (fragment-exec-based CLI), `mempalace.mcp_server` (MCP tool server), `mempalace.palace` (palace operations and locks), `mempalace.searcher` (hybrid ranking and filters), and `mempalace.backends` (pluggable storage). The knowledge graph is SQLite-backed with temporal validity windows. A write-ahead log provides an audit trail. The system uses a hybrid logical clock (HLC) for causal ordering and a source adapter system for pluggable ingest. Although primarily an application, the project includes extensive benchmarking and model evaluation suites, giving it a mild hybrid character.
Swarm is a production-grade Swift-native agent runtime framework for building multi-agent conversational systems with type-safe tools, on-device Apple Foundation Models, composable workflows, memory, guardrails, and observability. The architecture is modular: core agent execution (Agent, Workflow, Job) with pluggable inference providers (Foundation Models, OpenAI-compatible, mock), memory backends (conversation, sliding window, summary, vector, persistent via SwiftData or Wax SQLite), and tool integration (built-in tools, MCP protocol, web search). Key entry points include Agent.run()/stream(), Workflow.execute(), and the SwarmMCPServerService for exposing tools via MCP. Data flows involve agent turns with tool calls, memory retrieval/consolidation, and durable workflow checkpointing via the HiveCore deterministic runtime. External dependencies include Apple FoundationModels (macOS 26+), swift-log, swift-syntax (for macros), and optional Wax (SQLite vector store) and OpenTelemetry SDK. The framework emphasizes compile-time safety via @Tool and @Agent macros, and provides resilience patterns (retry, fallback, circuit breaker, rate limiting) and observability (tracing, metrics, OpenTelemetry).
AgentMemory is a persistent memory system designed for AI coding agents (Claude Code, Cursor, Codex, etc.) that captures agent observations via hooks, compresses and stores them, and enables semantic/recall search across sessions. Built on the iii-engine framework, it provides a modular architecture with CLI, REST API, MCP server, and web viewer interfaces. The system eliminates repetitive context re-explanation by automatically injecting relevant memories into agent sessions, implements memory consolidation to extract stable facts, and includes features like working memory management, audit trails, and health monitoring. Agent integrations occur through platform-specific hooks and MCP tools, allowing agents to store observations and retrieve context without manual intervention.
openacid/slim is a Go library for building memory-efficient, serializable trie indexes and sparse arrays, primarily designed for in-memory indexing of large on-disk key-value data. The core artifact is SlimTrie, a compressed trie that achieves roughly 11 bits per key while supporting fast Get, range queries, prefix scans, and versioned marshal/unmarshal. The repository also contains a standalone sparse-array package and a type-safe encoder framework, with extensive benchmarks comparing against Go maps and btree indexes. Its architecture is modular: the trie package defines the index structure, the array package provides bitmap-backed dense/sparse storage primitives, and the encode package abstracts value serialization with generated concrete implementations.
Dawarich is a production Ruby on Rails 8 monolith for personal location history tracking, positioning itself as a self-hostable Google Timeline replacement. It ingests timestamped geospatial points from a broad range of trackers (OwnTracks, Overland, Traccar) and file formats (GPX, GeoJSON, KML/KMZ, TCX, FIT, CSV, Google Takeout), then drives them through heavy background processing: reverse geocoding, anomaly filtering, visit detection, track generation, transportation-mode classification, trip building, and monthly/yearly statistics with digest emails. The web frontend is a MapLibre GL JavaScript application using vector-tile rendering, WebSocket real-time updates, and novel in-browser encoding studios for route videos (WebCodecs MP4) and printable posters (hand-rolled PNG encoder). The system also integrates photo libraries (Immich, PhotoPrism), flight logs (AirTrail), TeslaMate, and TREK trip planning, plus family location sharing, achievements, and an MCP API endpoint. Architecturally it is a conventional but well-executed Rails service-object monolith: controllers are thin, domain logic lives in app/services (command pattern with call methods), heavy processing is delegated to a large Sidekiq job surface with named queues, and cross-process coordination uses Redis-based debouncers, per-user advisory locks, and tile epochs for cache invalidation. PostGIS powers spatial queries (ST_DWithin, lonlat geography, H3 hexagons, vector tiles). The codebase ships dual-mode: self-hosted Docker deployments and a cloud SaaS with subscription/entitlement gating, with many env-conditional branches. Test coverage is extensive (RSpec with regression specs for timezone/race/geospatial edge cases, Playwright e2e, locale-parity and schedule-consistency specs), CI includes RuboCop, Trivy, and Dependabot. The strongest engineering is in the data-ingestion and processing layer: streaming SAX-style JSON parsers for huge Google Takeout files, batched upsert with deduplication and Null Island filtering, deterministic nearest-centroid transportation-mode classification with explicit explainability, and Redis coordination patterns for geocoding claims and per-user job locks. Weaknesses include a sprawling surface area (35+ service subdomains in one repo), significant schema churn via migration-enqueued backfill jobs, and dual self-hosted/cloud branches that add cognitive load. Overall this is a well-maintained, feature-complete application with genuinely reusable geospatial and streaming-data patterns.
ByteRover CLI is an AI-powered development assistant that provides persistent, structured memory for AI coding agents through a context tree. It features a multi-layered architecture with CLI (OCLIF), TUI (React Ink), and WebUI (React/Vite) interfaces, backed by a daemon service handling core agent logic. The system enables knowledge curation, synthesis, version control for context trees, swarm-based querying across multiple LLM providers, and integration with external AI agents via MCP. Key capabilities include tool execution (file operations, code execution), review workflows for AI-generated changes, and telemetry tracking.
This repository implements a system to use NVIDIA GPU VRAM as swap space on Linux via the NBD (Network Block Device) protocol. The core component is a userspace daemon (nbd-vram) that allocates VRAM using the CUDA driver API and serves it as a block device over a Unix socket. The Linux kernel's NBD driver connects to this socket, exposing /dev/nbdX as a swap device. The system includes scripts for connection/disconnection, power management (automatically disabling swap on battery), and proper handling of suspend/resume to avoid GPU-related deadlocks. Key features include memory locking to prevent the daemon from being swapped out, using PR_SET_IO_FLUSHER to avoid deadlocks under memory pressure, and multi-threaded request batching to amortize CUDA synchronization overhead. The solution works with any CUDA-capable NVIDIA GPU without requiring kernel modifications or NVIDIA's P2P API, which is restricted to professional GPUs.
OpenCode-Mem is a plugin for OpenCode, an AI coding assistant, that provides persistent memory using vector embeddings and Turso/libSQL. It stores project-level technical memories (decisions, code facts, patterns) and learns user preferences/configs across sessions. The system uses local ONNX embeddings (or remote API) and supports multiple LLM providers (Anthropic, OpenAI, Google Gemini, MiniMax). It features auto-capture of session content, user profile learning with confidence decay, deduplication, and migration services. A full React web UI allows browsing, editing, and managing memories and user profiles. The architecture is modular: plugin entry, AI provider abstraction, Turso shard management, embedding service, web server, and shared API schemas. While primarily an application, it includes robust data pipeline patterns for vector search, ETL (migration), and user modeling.
Activeloop's DeepLake is a production-grade, multi-vector database engine for AI data. At its core, it is a columnar, chunked, compressed storage system for tensors (multi-dimensional arrays) with native support for embeddings, images, video, text, and other complex AI data types. It exposes a Python library for dataset management, ingestion (COCO, Labelbox, CSV, Parquet), version control, and integration with PyTorch/TensorFlow/MMDetection. A high-performance C++ backend provides lazy NumPy-like array operations (nd::array), a composable async framework (async::promise), and a Tensor Query Language (TQL). The most distinctive component is a PostgreSQL extension (pg_deeplake) that allows SQL access to DeepLake datasets via custom table and index access methods, supporting vector similarity search (cosine, BM25) and hybrid search. Query execution is delegated to DuckDB through a custom scan node, enabling efficient analytical queries. The extension also includes WAL-based DDL replication across instances. This is primarily an application (a database system), with substantial research-oriented features such as medical image formats, compression codecs, and TPC-H benchmarking.
This is a production-grade long-term memory plugin for the OpenClaw AI agent platform, built as a TypeScript/ESM module. It provides a complete memory lifecycle: LLM-powered extraction from conversation transcripts (6-category classification with L0/L1/L2 layering), hybrid vector+BM25 retrieval with cross-encoder reranking, Weibull decay-based importance scoring, tier management (core/working/peripheral), and multi-scope isolation (global, agent, user, project). The plugin integrates deeply into OpenClaw via lifecycle hooks, agent tools, CLI commands, and a dreaming engine for periodic consolidation. It stores all data in LanceDB and supports multiple embedding providers, LLM backends, and OAuth flows. The codebase is large and complex, with extensive test coverage (100+ tests) and many edge-case handlers, but exhibits significant technical debt in a single massive index.ts file and multiple fallback/legacy migration paths. It is not a research project; it is a highly functional, actively maintained plugin with a clear engineering focus on correctness, resilience, and performance.
Total Recall is a memory system for Claude Code that provides persistent, curated memory across sessions. It implements a tiered architecture with daily logs for raw capture, registers for structured domain knowledge, working memory for essential context (auto-loaded every session), and archive for historical data. The core innovation is its write gate mechanism that evaluates whether information should be saved based on whether it changes future behavior, preventing memory bloat. The system includes automatic hooks for session context loading and pre-compaction flushing, a sophisticated ID-based metadata system for tracking memory entries, and maintenance routines for pressure-based cleanup. It's designed as both a plugin and standalone installation with comprehensive slash commands for all memory operations.
Supermemory is a commercial-grade AI memory and context engine that provides persistent long-term memory for AI agents. The repository implements the client-side integration SDKs, an MCP (Model Context Protocol) server, a memory graph visualization component, and supporting infrastructure. In production, it connects to an external cloud API (api.supermemory.ai) that handles document ingestion, semantic search, user profiles, and knowledge graph construction. The codebase is organized as a monorepo with distinct packages for TypeScript/Python SDKs covering multiple AI frameworks (Vercel AI SDK, OpenAI, Mastra, Claude, etc.), a Cloudflare Workers-based MCP server with OAuth and widget UI, a ship-grade memory-graph React component with force-directed canvas rendering, and a comprehensive validation schema package. The architecture demonstrates strong engineering practices: strict Zod validation, middleware injection patterns, caching and deduplication logic, and thorough testing across languages. Notable skill is evident in the memory-graph canvas engine (viewport, simulation, spatial indexing, version chain visualization) and the shared memory-context injection utilities. The main downside is significant logic duplication across four Python SDK packages and their TypeScript counterparts, which increases maintenance risk. The repo is actively developed with CI/CD pipelines publishing to npm and PyPI.