Best memory and evaluation tools for AI agents
Give agents durable memory, then measure whether they actually work. Memory tools first, evals and observability second. Updated October 2026 · 24 picks.
1. Mem0
Memory layer that extracts and retrieves user/agent facts across sessions.
Source →2. Letta (MemGPT)
Stateful agents with self-editing tiered memory, from the MemGPT research.
Source →3. Zep / Graphiti
Temporal knowledge graph for agent memory that tracks how facts change over time.
Source →4. LangMem
LangChain library for long-term memory extraction and prompt optimization in agents.
Source →5. Cognee
Builds knowledge graphs plus vectors from your data as agent memory.
Source →6. MCP Memory server
Reference knowledge-graph memory server any MCP client can use.
Source →7. Anthropic memory tool
Claude API tool for storing and retrieving information in a file directory across conversations.
Source →8. Anthropic: Effective context engineering
Guide to curating context: compaction, structured notes and sub-agent architectures.
Source →9. Chroma
Open-source embedding database commonly used for agent retrieval memory.
Source →10. SWE-bench
Benchmark of real GitHub issues for evaluating coding agents.
Source →11. Inspect AI
UK AISI framework for LLM and agent evaluations with sandboxed tools.
Source →12. OpenAI Evals
Framework and registry for evaluating LLMs and LLM systems.
Source →13. promptfoo
CLI for testing prompts, agents and RAG with assertions and red-teaming.
Source →14. DeepEval
Pytest-like LLM evaluation framework with agentic metrics like task completion.
Source →15. Langfuse
Open-source LLM observability: traces, evals and prompt management for agents.
Source →16. Arize Phoenix
Open-source tracing and evaluation for LLM apps built on OpenTelemetry.
Source →17. tau-bench
Sierra benchmark for tool-agent-user interaction in realistic domains.
Source →18. Terminal-Bench
Benchmark for agents completing tasks in real terminal environments.
Source →19. Braintrust
Eval and observability platform for iterating on AI products.
Source →20. Ragas
Evaluation toolkit for RAG and agent pipelines.
Source →21. Anthropic: Demystifying evals for AI agents
Practical guidance on designing graders and tasks for agent evaluation.
Source →22. LangSmith
Tracing, evals and monitoring for LLM apps.
Source →23. W&B Weave
Weights & Biases toolkit for tracing and evaluating LLM apps.
Source →24. Helicone
Open-source LLM gateway and observability.
Source →