Not on a framework
Python SDK (Python 3.11 and above), TypeScript SDK (Node 20 and above, native rather than a wrapper), and a REST API from any language.
SDK referenceNew research from Maximem. Agentic Context Management: Solving Agent Memory and Cost by Treating Them as Lifecycle and Architecture ProblemsAgent Memory is an architecture problem. Read the paper →
Synap · Integrations
Maximem Synap ships 23 native framework integrations today — each with a dedicated docs page and, for our flagship integrations, a deep-dive blog post. Every integration is in production and verified against docs.maximem.ai/integrations.
Drop-in memory provider for LangChain chains and agents — replaces ConversationBufferMemory with persistent, cross-session recall.
Persistent memory across LangGraph nodes — state that survives graph reruns, checkpoints, and human-in-the-loop pauses.
Long-term agent memory for LlamaIndex query engines, workflows, and agentic retrievers.
Native memory tool for the OpenAI Agents SDK — drop into any agent loop without rewriting tool definitions.
Typed memory backing for Pydantic AI agents — model-validated reads and writes, no string parsing.
Shared memory across CrewAI crews and tasks — every agent in the crew sees the same evolving context.

Cross-conversation memory for AutoGen multi-agent groups — durable state outside a single GroupChat run.

Memory layer for Google Agent Development Kit agents — built for ADK sessions and tool-calling loops.

Persistent context for Haystack pipelines and agents — memory components that compose with existing nodes.
Memory provider for Agno agent workflows — keeps team and session memory in sync across runs.
Memory plugin for Microsoft Semantic Kernel — works in both the Python and .NET SDKs.
Memory backing for the Microsoft Agent Framework — drop-in for any AgentChat or Group conversation.
Memory toolkit for NVIDIA NeMo Agent Toolkit — integrates with NeMo agent and workflow primitives.
In-conversation memory for LiveKit voice agents, under 15ms at P75, with anticipatory retrieval fast enough for real-time turn-taking.
Real-time conversational memory for Pipecat pipelines — built for voice-first agent loops.
Memory tool for the Claude Agent SDK — usable from any Claude agent loop without custom glue code.
TypeScript memory module for Mastra agent flows — typed reads and writes from any Mastra workflow node.
Server-side memory layer for the Vercel AI SDK — wraps streamText / generateText with persistent context.
Memory tools plus a per-turn short-term-context resolver for Vercel eve agents, in TypeScript.
Native MemoryStore for Strands Agents — short-term context hook, memory tools, and a real-time anticipation feed.
Native AgentMemory for CAMEL-AI — long-term recall layered over the agent's own conversation history.
Memory tools and a per-step turn recorder for Hugging Face Smolagents — persistence across agent runs.
Backend and two middlewares for deepagents agents: persistence across runs and sub-agents.
Not on one of the frameworks above, or not on a framework at all. Synap works from anywhere.
Python SDK (Python 3.11 and above), TypeScript SDK (Node 20 and above, native rather than a wrapper), and a REST API from any language.
SDK referenceA hosted MCP endpoint, or self-host the adapter. Point any MCP-aware tool at it and it has memory.
MCP guideThe CLI, for provisioning and operating Synap without opening a dashboard. Install with pipx, Homebrew, or a one-line installer.
CLI guideconst { createClient } = require('@maximem/synap-js-sdk');
const client = createClient({ apiKey: 'your-api-key' });
await client.init();
// Send the conversation as it happens. Returns immediately.
await client.conversation.recordMessage({
conversationId: 'mon-standup',
userId: 'alice',
role: 'user',
content: "I'm migrating our auth service to OAuth2 this sprint.",
});
// Ask what is known, before your agent replies.
const context = await client.fetchUserContext({
userId: 'alice',
query: 'what is alice working on?',
});
console.log(context.formattedContext);Real import jobs from other memory providers, with field mapping and read-back verification, not an export file and instructions. Migrations in the docs →
First-party connectors that pull your organisation’s own context continuously, so agents know your product, policies and pricing, not only the user.
Or upload files, or send it over the API. Whatever the source, it lands at the client level of your hierarchy and reaches every agent that should see it. How scoping works →
The Synap SDK (Python and TypeScript) and REST API work from anywhere — and most new framework adapters land within a release. Tell us what you're shipping on.