2026.10.06
SoloMD: A Markdown Vault for Local-First AI Workflows
SoloMD is an MIT-licensed, local-first Markdown editor with MCP access and agent recipes. Review data flows, automated writes, platform packaging, and deployment requirements before adoption.
What it does
SoloMD treats a folder of Markdown files as a knowledge vault, adds an editor-side AI surface, and bundles solomd-mcp so external MCP clients can work with the same workspace. It is aimed at people who want to keep readable text files while allowing tools such as Claude Code, Cursor, or other MCP clients to help organize and rewrite notes. The repository is MIT-licensed and not archived.
Known features
The editor includes WYSIWYG editing, tabs and split panes, KaTeX, Mermaid, image pasting into _assets/, slideshow mode, Vim mode, Hunspell and CJK proofreading, semantic search, wikilinks, backlinks, and Pandoc export. The README also lists automatic detection for GBK, Big5, and Shift-JIS encodings.
Knowledge-base features include a YAML frontmatter inspector, type-driven navigation, typed relationships, a per-note relationship graph, saved filtered views, and Markdown-backed tldraw whiteboards. Users who value portable files should inspect how these features affect the on-disk format and their existing workflow.
The AI layer includes an Agent Panel, YAML-defined recipes with scheduled or event-based triggers, and an AutoGit branch sandbox. The README says agent writes are placed on a separate branch for acceptance or rejection, with a replayable trace for each run. The MCP service is described as read-only by default, with writing requiring an explicit option. The project also lists 14 AI providers and BYOK, but provider behavior, pricing, and data policies must be checked separately.
Who it suits
SoloMD is a plausible fit for Markdown-vault users, researchers, writers, and developers who want AI-assisted organization without abandoning local files. Users familiar with Git, MCP, and YAML will likely find the integrations easier to operate. People seeking only a minimal Markdown editor may find the agent, graph, and integration layers add learning and maintenance overhead.
Limitations before adoption
“Local-first” does not mean every AI request is offline. The README lists Ollama as a local provider alongside cloud providers; whether data leaves the device depends on configuration, model choice, and the request path. BYOK also means that users remain responsible for provider accounts, keys, and costs.
Automated writes should be tested in a copy or disposable workspace first. Branch isolation, write caps, and dirty-tree checks described by the project do not replace backups, change review, or verification of model output. Sensitive, copyrighted, or high-stakes material should not rely solely on an agent.
Platform availability and release packaging are separate questions. The README lists several operating systems, while release v4.14.8 provides installation guidance for macOS, Windows, and Linux. The release notes state that Windows builds are not code-signed, so users may see an operating-system warning. The sources do not confirm feature parity, minimum hardware, or long-term mobile behavior across platforms.
Deployment considerations
Before deployment, confirm the workspace path, backup plan, Git permissions, and each AI provider’s data policy. If MCP is enabled, expose only the workspaces that are necessary and begin in read-only mode. When writes are required, review each change. GitHub synchronization and other HTTPS Git services also require separate decisions about repository visibility, credential handling, and conflict resolution.
Windows users should obtain packages from the release attachments and use the provided checksum file for independent verification; v4.14.8 mentions SHA256SUMS.txt. This article is based on the supplied project sources and does not claim independent testing, security validation, or performance measurement.
Sources
Sources checked: 2026-10-06