smolagents Memory
https://huggingface.co/docs/smolagents/tutorials/memory
Hugging Face barebones agent framework. Memory lives as a Python list; no built-in persistence. Step callbacks let you mutate memory between steps. Persistence is left to the user — community proposals exist for "memory bank"-style additions.
At a glance
- Type
- In-process Python list (no persistence)
- Tier
- T2
- Section
- Framework-embedded memory
- Created
- 2024-12
- Latest release
- v1.24.0 2026-01-16
- License
- Apache-2.0
- GitHub
- 27.1k★ +104/mo Python
- Pricing
- OSS free (Apache-2.0); Hugging Face Enterprise Hub starts at $20/user/mo for managed services
- Funding
- Hugging Face (parent) raised $400M total across 8 rounds from 34 investors
Taxonomy
- storage
- kv
- retrieval
- injection
- persistence
- session
- update
- append-only
- unit
- episode
- governance
- inspectable
- conflict
- append
When to use
Optimised for: simplicity / barebones agent loop
Anti-fit: not for production multi-session use - in-process Python list, no persistence
Pros & cons
Pros
HuggingFace's minimal-agent framework; tiny dependency surface; useful for embedded / edge deployments.
Cons
Memory features are intentionally minimal; not the right fit for sophisticated long-term memory needs.
Claims & capabilities
HuggingFace lightweight code-first agent framework (~26k★ vs LangChain's ~122k); memory exposed as ordered list of past steps (history of plan/exec/error); step callbacks let users dynamically modify memory; long-term memory and memory-bank are open community-requested features
Technical surface
- API surface
- SDK: Python (agent.memory.steps; CodeAgent/ToolCallingAgent)
- Backend storage
- in-memory
- Deployment
- Both (OSS self-hosted; Hugging Face Inference Endpoints / Enterprise Hub managed)
- Embedding model
- multiple supported
- Multi-tenancy
- not applicable — library
- MCP
- via official adapter — smolagents.MCPClient
- A2A
- searched not found
- OpenTelemetry
- via adapter — Arize Phoenix
Similar systems
Other framework-embedded memory in the catalog, ranked by inbound references.
- LangGraph Persistence T2
Distinct from LangMem. Built-in checkpointer saves graph state per super-step (short-term, thread-scoped). Store System adds long-term hierarchical key-value memory across threads with optional vector search + TTL. Postgres / Mongo / Redis stores for production.
- AutoGen Memory T2
ListMemory chronological context + teachable agents that vectorise corrections. Integrates with Mem0/Zep rather than building deep memory natively.
- CrewAI Memory T2
Memory subsystem inside the CrewAI orchestration framework; integrates with Mem0 for the long-term tier.
- AGiXT Adaptive Memory T2
Open-source AI automation platform. Routes between short-term and long-term memory adaptively across any LLM provider; plugin system for storage backends. Memory managed at the instruction-management layer — task context, instruction state, conversation history as unified agent state.
- Agno (Phidata) Memory T2
Agno (formerly Phidata). AgentStorage persists sessions to a DB; AgentMemory auto-classifies/store user preferences and conversation summaries. Single-line integrations with LanceDB, Pinecone, Weaviate, Qdrant.
- Botpress LLMz T1
Per-plan vector-DB storage quota + LLMz autonomous engine (in-session working memory) + Knowledge Base (semantic search over uploaded docs). Long-term user memory persists across sessions.