CMEM launches persistent memory layer for AI coding agents

CMEM Cloud is a memory system designed to let AI coding agents retain and reuse context across projects and time. The core product consists of two parts: claude-mem, a free open-source engine, and CMEM Cloud, a paid managed service.

The open-source claude-mem engine installs in 30 seconds via npm (npx claude-mem install) with no configuration or account required. It runs locally on your machine and continuously captures observations — decisions, code fixes, dead ends — from agent work, storing them in a local observations database. It includes semantic vector recall (search by meaning, not keywords) and temporal context (tracking what changed, when, and what it replaced). The engine also bundles 11 skills and works with Claude Code, Cursor, Windsurf, OpenCode, OpenClaw, Codex CLI, and Gemini CLI, as well as any MCP (Model Context Protocol) client.

CMEM Cloud, now live, adds a cloud layer on top of claude-mem. It mirrors the observations database to the cloud and makes it reachable from anywhere through a private MCP link. The service includes an offline-first cloud mirror that syncs automatically, a live memory feed accessible on your phone, and end-to-end privacy (your link, your keys). When significant events occur — such as a checkout error rate spiking, a pricing decision landing in a team chat, or a new API contract being merged — the system fires a "brainbeat": a bespoke context notification that briefs the right agent and sets it working. For example, when a pricing decision lands in #general, the enablement agent is briefed and begins drafting an update; when a new API contract merges, the docs agent rewrites the quickstart.

Pricing for CMEM Cloud is $20/month for individuals and can be cancelled anytime. For teams of 3–50 seats, the price is $333/seat/month; every team plan includes a Forward Deployed Engineer to lead onboarding, either remote or on-site. The company claims that the system offers "0–12× cheaper codebase exploration versus full reads" and "~0% lower agent cost via model tier routing," suggesting it reduces the computational cost of agent operations by intelligently routing work to cheaper models when sufficient context is available. According to the product description, the system allows the whole memory (observations DB, skills, and context) to be accessed "on your phone" in real time, so teams can watch every machine and agent read and write to one shared timeline.

CMEM addresses a common friction in multi-agent and multi-developer workflows: context loss across projects and sessions. When AI agents or teams work on different codebases or revisit old projects, they typically start from scratch because the agent has no persistent memory of prior decisions, failed approaches, or architectural choices. CMEM's core insight is that a structured memory layer — combining a local database, vector search, and a sync service — allows agents to recall the reasoning behind decisions months later, ranked by semantic relevance rather than keywords alone.

The product strategy splits the offering into an open-source foundation (claude-mem) and a managed cloud layer (CMEM Cloud). The free open-source version lowers the barrier to adoption and allows teams to test the concept locally; the cloud tier adds synchronization across devices and agents, mobile access, and team governance. The pricing structure ($20/month for solo developers, $333/seat/month for teams) positions CMEM as an add-on to existing agent and IDE tooling rather than a replacement. The claims of "0–12× cheaper codebase exploration" and "model tier routing" suggest the system is also marketed as a cost-optimization tool, enabling smarter routing of agent work to cheaper models when full context is available.

No comments yet. Be the first to share your thoughts!