2026.10.16
IWE: Shared Markdown memory for people and AI agents
IWE turns local Markdown files into a connected knowledge graph with editor LSP features, a CLI, and MCP tools for AI agents. It suits developers who value readable local files, but editor support, agent access, and deployment details should be verified first.
What it is for
IWE, from iwe-org/iwe, organizes local Markdown documents into a connected knowledge graph intended to serve as shared project memory for people and AI agents. Notes can hold specifications, plans, decisions, and lessons. People can read, edit, reorganize, or delete them, while an agent can read, write, and organize the same documents during a workflow.
IWE is not a standalone AI model service. Its README states that neither the CLI nor the Mac app includes an AI model. AI assistance comes from an existing Claude, Codex, or another MCP-compatible tool. In practice, IWE is best understood as a local notes layer combined with editor integration and an agent-facing tool interface, rather than as an independent chat product.
Features and suitable users
The project presents four ways to work: a native Mac app, editor integrations, a CLI, and agent access through MCP. The documented editor integrations include Neovim, Helix, VS Code, and Zed, with link completion, backlinks, search, and safe renaming. This may suit developers, technical writers, researchers, and users of Zettelkasten-style or project knowledge-base workflows who already write Markdown in those editors.
The CLI is aimed at terminal and scripted workflows. It can search, create, and reorganize notes, including content search. Through MCP, an AI agent can read, save, and organize memory. The same Markdown files can be used by the editor, CLI, Mac app, and other compatible tools. The Mac app is described as providing local browsing and writing, search, navigation between related notes, and Mermaid diagram viewing.
Limitations before adoption
IWE does not provide the model, model quota, or agent account. Availability, authentication, permissions, and data handling for Claude, Codex, or another MCP client must be evaluated separately. Once an agent can access local notes, users should decide carefully which folders and files it may read or modify. The supplied project materials do not support a broad security guarantee.
The available sources also do not confirm a complete platform matrix, minimum hardware requirements, scale limits, or compatibility across every editor version. The Mac app is described as a native Mac application; equivalent graphical support on other platforms should not be assumed. Consult the official installation and editor documentation and validate the workflow in the target environment.
The repository identifies Apache License 2.0 and is not archived. Release v0.26.1 fixed cases where missing-document or non-Markdown links could make queries crash, hang, or return non-document objects. That release note describes specific fixes, not a general stability promise. Keep backups of the Markdown files and plan for synchronization, git conflicts, and review of agent-generated edits.
Deployment considerations
A practical starting point is to choose a notes folder as the local Markdown source, then select the editor, CLI, Mac app, or MCP agent that matches the workflow. Users relying on version control should define naming, structure, and change-review conventions before inviting an agent to edit files. Start with low-risk notes, limit the agent’s working directory and write scope, and retain human review.
IWE’s central appeal is that memory remains visible and portable as ordinary files. Whether it fits a particular environment depends on the existing editor, agent account, MCP client, and local file-management practices. Those integrations should be confirmed before committing important project knowledge to the workflow.
Sources
Sources checked: 2026-10-16