Engram is a persistent memory layer for LLMs and AI Agents, designed to let Claude, ChatGPT, and virtually any model share the same long-term memory. It targets both everyday Claude/ChatGPT users and Agent developers: the former can carry context across different AI tools, while the latter can connect structured memory to their Agents via API, MCP, or SDK.
Engram’s focus is not simply saving chat logs or running similarity search. Instead, it extracts conversations into typed bullets such as FACT, DECISION, PREFERENCE, GOAL, PROCEDURE, and PRINCIPLE. Each memory item receives a salience score, and the system combines importance, relevance, and token budget to generate context that can be injected into prompts. It also emphasizes handling updates and contradictions—for example, when a user “moves from Boston to Vancouver,” the old fact is superseded while version history is preserved. Over long-term use, the system deduplicates, merges, forms schemas, and decays stale knowledge.
The free plan is $0/month and includes 3 contexts, 1,000 bullets per context, BYOK, Dashboard, Claude MCP, ChatGPT Actions, and community support. The Enterprise plan is custom-priced and offers unlimited contexts, 50,000 bullets per context, graph browsing, dedicated tenants, data isolation, SSO/SAML, audit logs, and a dedicated SLA. Integration options are fairly comprehensive, including Claude MCP, ChatGPT Actions, REST API, and Python SDK, with references to use in stacks such as LangGraph, CrewAI, Claude, and OpenAI. The full engine is MIT open source and can be self-hosted, which is a major advantage.
Its main strength is that its memory model is closer to a knowledge layer: it supports types, priority, decay, conflict handling, and token budget control, making it better suited to long-running Agents than a simple vector database or context concatenation. The open-source codebase and free tier also reduce the cost of trying it out. The limitations are that the main documentation does not disclose the underlying models, quality benchmarks, or multilingual performance; Chinese support is not clearly stated. On privacy, while it offers dashboard controls, enterprise data isolation, and self-hosting, it lacks detailed information on compliance certifications, encryption, and data usage policies.
Engram is suitable for developers building personalized assistants, long-term project assistants, multi-model workflows, and Agent memory systems. It is also useful for users who frequently switch between Claude and ChatGPT. Access from mainland China and supported payment methods are not disclosed. Since Claude and ChatGPT themselves typically face network and account restrictions in China, actual usability may depend on the connected platform. Alternatives worth watching include Mem0, Zep, LangGraph Memory, or a self-built vector database/RAG setup.
⚠ This review is compiled from public sources and does not constitute a purchase recommendation. Verify all facts on the vendor's official site. Verify on engram.so official site.
engram.so is an Unknown AI Apps provider. TG4G tracks its product information, an overall rating of 8.0/10, and a China-accessibility score of China direct-connect friendly. Click "Visit Official Site" to reach engram.so directly.