Memori is agent-native memory infrastructure designed to transform agent execution and conversation into structured, persistent state for production systems. The project addresses a fundamental challenge in AI agent deployment: agents typically forget everything between sessions, losing context about user preferences, coding patterns, project conventions, and decision history. Memori solves this by automatically capturing and structuring memories from agent interactions without requiring changes to existing agent code or prompts.
The platform is LLM-agnostic and datastore-agnostic, integrating seamlessly into existing architectures without requiring rip-and-replace migrations. It supports deployment across managed cloud, single-tenant cloud, VPC, and on-premises environments. Memori offers both a managed cloud service through Memori Cloud with zero-config setup via API key, and a bring-your-own-database option for organizations requiring data sovereignty. The project provides SDKs in both Python and TypeScript, with support for major LLM providers including Anthropic, OpenAI, Gemini, DeepSeek, Grok, and others, as well as frameworks like LangChain, Pydantic AI, and Agno.
The repository demonstrates strong community engagement with median issue and pull request response latency of 2.3 hours according to GitGenius tracking data. The most active contributors include harshalmore31 with 89 tracked events, devwdave with 77 events, and Boburmirzo with 58 events. Bug reports and enhancement requests represent the most active issue categories with 19 and 15 items respectively, alongside 14 hacktoberfest-tagged items. The project overlaps with contributors from notable repositories including PingCAP's TiDB, Argo Workflows, and Anthropic's Claude Code.
Memori's Advanced Augmentation system tracks memories at entity, process, and session levels, enriching them with attributes, events, facts, people, preferences, relationships, rules, and skills. This multi-level approach creates contextual understanding between users and agents without incurring latency. The platform includes specialized integrations: an OpenClaw plugin that captures structured memory from agent execution and tool calls, a Hermes Agent memory provider with explicit recall tools, and Model Context Protocol support for tools like Claude Code, Cursor, and Warp.
Performance benchmarking on the LoCoMo benchmark for long-conversation memory shows Memori achieving 81.95% overall accuracy while using only 1,294 tokens per query, representing just 4.97% of full-context footprint. This efficiency demonstrates 67% prompt size reduction compared to Zep and over 20x lower context cost versus full-context prompting. The project includes a command-line interface for managing accounts, API keys, and quotas, plus a dashboard for exploring memories, analytics, and testing.
The codebase is classified by GitGenius across multiple AI domains including Continual Learning, Embodied AI, Memory Systems, Lifelong Learning, Experience Management, Knowledge Representation, and AI Agents. The project is licensed under Apache 2.0 and maintains active documentation for both cloud and self-hosted deployments, with a companion cookbook repository providing examples and demonstrations.
