Memory Systems

OpenClaw v2026.4.23-beta.5 introduces significant enhancements to image generation across multiple providers (OpenAI, OpenRouter), adds optional forked context for subagents, implements per-call timeout support for generation tools, and includes configurable local embedding context sizes. The release fixes numerous issues across Codex harness routing, media handling, provider authentication, and UI persistence, while updating Pi dependencies to v0.70.0 and improving debug logging for harness selection decisions.

crewAI 1.14.2 introduces checkpoint management features including resume, diff, and prune commands with lineage tracking and forking capabilities. The release enhances LLM token tracking with reasoning and cache tokens, improves CLI ergonomics for deploy validation and template management, and fixes multiple security vulnerabilities across dependencies. Key improvements include scoped streaming handlers to prevent data contamination, recursive glob for checkpoint discovery, and better HITL (Human-in-the-Loop) resume event handling.

Google ADK Python v1.31.0 introduces significant enhancements including Vertex AI Agent Engine Sandbox integration for computer use, Firestore support, and improved memory management with VertexAiMemoryBankService. The release adds user agent tracking to Parameter and Secret Manager clients, includes live session tracking in LlmResponse, and resolves multiple bugs related to event callbacks, BigQuery plugin issues, and authentication headers. Several refactoring improvements modernize the live session API and update MCP Toolbox documentation.

OpenClaw v2026.4.15 is a major release featuring Claude Opus 4.7 as the default Anthropic model, new Google Gemini text-to-speech support, a Model Auth status card for OAuth health monitoring, cloud storage support for LanceDB memory, and GitHub Copilot embedding provider integration. The release includes numerous stability fixes across gateway tools, agent replay recovery, Docker builds, Matrix E2EE, memory management, and channel integrations, plus packaging improvements to reduce bundle size and plugin runtime dependencies.

OpenAI's openai-agents-python v0.14.0 introduces Sandbox Agents, a beta feature enabling agents to run in persistent, isolated workspaces with file access, command execution, and cross-run continuity. The release includes multiple execution backends (local, Docker, hosted providers), workspace management with snapshots and resume capabilities, sandbox memory for learning across runs, and comprehensive examples. Runtime enhancements provide tracing, session lifecycle management, and seamless integration with the existing Agent and Runner framework.

OpenClaw v2026.4.15-beta.1 introduces significant enhancements across control UI, memory management, and security. Key additions include OAuth token health monitoring, cloud storage support for LanceDB memory indexes, GitHub Copilot embedding provider integration, and experimental lean mode for local models. The release focuses on security hardening with credential redaction, symlink protection, and improved authentication handling, alongside numerous bug fixes for CLI configuration, memory access control, and provider failover logic.

Claude Code v2.1.108 introduces prompt caching TTL configuration options, a new recap feature for session context, and improved slash command discovery via the Skill tool. The release includes numerous UX improvements such as better error messaging, enhanced session resumption, and a warning system for model switching mid-conversation. Multiple bug fixes address transcript handling, telemetry interactions, file operations, and terminal compatibility issues.

OpenClaw v2026.4.12 is a comprehensive quality release emphasizing plugin loading reliability, memory system improvements, and expanded local model support. Key additions include an Active Memory sub-agent plugin, local MLX speech provider for Talk Mode, LM Studio provider for self-hosted models, and enhanced Feishu setup workflows. The release also addresses critical security fixes, dreaming/narrative stability, and improves gateway lifecycle management and Matrix client streaming support.

OpenClaw v2026.4.12-beta.1 introduces significant improvements to plugin loading, memory recall, and dreaming functionality. Key changes include narrowing plugin activation to manifest-declared needs, improving QMD recall search telemetry, and fixing critical issues with memory retrieval, dreaming narrative cleanup, and gateway authentication. The release also expands documentation for providers and memory-wiki setups while addressing numerous edge cases in CLI updates, Unicode handling, and WebSocket keepalive mechanisms.

OpenClaw v2026.4.11-beta.1 introduces significant enhancements to memory/dreaming features with ChatGPT import ingestion, improved media rendering in chat UIs, expanded video generation capabilities, and better integration with Feishu and Microsoft Teams. The release includes numerous fixes for OAuth flows, audio transcription, TTS persistence, and configuration handling, plus improved plugin manifest descriptors and model provider diagnostics.

crewAI 1.14.0 introduces checkpoint management with CLI commands and SqliteProvider storage, implements runtime state checkpointing with an event system, and refactors the executor. Security enhancements include SSRF/path traversal protections and RAG tool validation. The release also optimizes memory serialization by excluding embedding vectors and addresses critical CVE-2026-35030 via litellm upgrade.

The Claude Agent SDK provides built-in session management functions to build conversation history sidebars without writing custom parsers. This cookbook demonstrates how to list, read, rename, tag, and fork sessions stored as JSONL transcripts on disk. By leveraging SDK primitives like list_sessions(), get_session_messages(), and rename_session(), developers can implement session browsers for desktop apps, IDE extensions, and chatbots without managing file I/O directly.

crewAI 1.14.0a4 introduces checkpoint storage capabilities with SqliteProvider and CheckpointConfig, implements runtime state checkpointing with an event system, and refactors the executor. Key improvements include guardrail tracing enhancements, embedding vector exclusion from memory serialization to reduce token usage, and a security bump to litellm. The release also deprecates CodeInterpreterTool and code execution parameters while improving documentation and performance with JSONB storage.

CrewAI v1.13.0 introduces significant architectural improvements including RuntimeState RootModel for unified serialization, enhanced event telemetry for skills and memory, and A2UI extension support. The release includes critical bug fixes for GPT-5.x model compatibility, RBAC permission alignment, and Windows lancedb compatibility. Performance optimizations reduce framework overhead through lazy event bus implementation and conditional tracing.

Claude Code v2.1.89 is a major release introducing deferred permission decisions for headless sessions, flicker-free rendering, improved MCP connection handling, and numerous bug fixes across tool execution, file handling, and UI rendering. Key improvements include named subagents, enhanced auto-mode permission workflows, and fixes for critical issues like memory leaks, LSP crashes, and prompt cache misses in long sessions.

This cookbook teaches context engineering strategies for long-running AI agents, focusing on three key techniques: compaction (summarizing context), tool-result clearing (removing re-fetchable tool outputs), and memory (persistent external storage). The guide addresses context rot—the degradation of model performance as context windows grow—and provides practical implementations using Claude's API. Through a research agent example, it demonstrates how to combine these strategies to manage token growth, maintain conversation continuity, and persist knowledge across sessions.

This guide teaches how to build knowledge graphs from unstructured text using Claude's structured outputs for entity extraction, relation mining, and entity resolution. Rather than training traditional NER and relation classifiers, Claude handles each stage via prompts, enabling multi-hop graph reasoning without a database. The approach uses Haiku for high-volume extraction and Sonnet for entity resolution, with techniques transferable to production databases like Neo4j or PostgreSQL.

CrewAI 1.12.0 introduces significant enhancements including Qdrant Edge storage backend, native OpenAI-compatible provider support (OpenRouter, DeepSeek, Ollama, vLLM, Cerebras, Dashscope), agent skills implementation, and automatic root_scope for hierarchical memory isolation. The release adds Arabic language support, a new docs-check CLI command, logout functionality, and resolves multiple bugs across type checking, HITL flow system, and memory management. Documentation improvements include a new CONTRIBUTING.md guide and instructions for using CrewAI without LiteLLM.

CrewAI 1.12.0a1 introduces native OpenAI-compatible provider support (OpenRouter, DeepSeek, Ollama, vLLM, Cerebras, Dashscope), agent skills, and automatic hierarchical memory isolation via root_scope. The release includes Arabic language support for documentation and tooling, a new docs-check command, and significant refactoring to decouple from LiteLLM while resolving mypy type-checking errors across all packages.

OpenAI Agents Python v0.13.0 is a minor release that updates the default Realtime websocket model to gpt-realtime-1.5, expands MCP server capabilities with resource management and session tracking, and introduces opt-in reasoning content replay for chat completions. The release includes critical runtime stability fixes for concurrent writes, tool removal edge cases, and batch executor races, with no breaking changes.