A local-first
personal knowledge base
your AI agents can use.

Cloud AI workspaces work by design: your data is uploaded, indexed, embedded, sometimes used to train, and then served back to you through a model the vendor hosts. Every step is a surface for leakage. Gartner forecasts that over 40% of AI-related breaches in 2027 will involve cross-organisational generative-AI use; enterprise reviews already terminate 73% of AI coding-tool implementations for the same reason.

Morion has nothing to breach. Your notes, tasks, comments, and attachments live in one local file in your user-data folder. Tool calls from MCP clients are localhost-to-localhost. There is no cloud index, no embedding service, no remote backup — even when Mo is running, only the slice of activity Mo needs for the current task ever leaves your machine, and Mo never holds the whole notebook. Pair Mo with a local LLM (Ollama) and even that slice stays on-device.

Bonus: pair Morion with a local LLM (Ollama, llama.cpp, LM Studio, MLX) and the entire loop runs offline. Plane, train, tent, airgapped machine — nothing blocks you. You don't have to commit to local LLMs to get value from Morion; the option is just there.

Notes apps capture context. They don't move work forward. Task apps move work forward but have no idea what your AI is doing. Morion is both, on the same data: a task is just a note with a status. Same search, same folders, same MCP, same audit log.

Six statuses (note · backlog · todo · doing · review · done) and atomic claim semantics mean an agent picks up a card with a single tool call, gets exclusive ownership, and either finishes it or hands it back. Two agents racing on the same kanban can't accidentally take the same job. "Drop something into Todo and an agent ships it overnight" is a real workflow on day one — Free tier, no Pro required.

This is the wedge no one is shipping. Linear has the queue but no AI integration. Notion has notes + AI but no atomic claim or agent-aware semantics. Obsidian is local but tasks are a plug-in patchwork. Morion makes the kanban a first-class agent surface.

Model Context Protocol is how every serious AI client (Claude Desktop, Claude Code, Cursor, Cline, Zed, Codex CLI, Windsurf, Google Antigravity) talks to outside tools. Morion ships a full MCP server in the box: 33 tools across notes, folders, tags, comments, attachments, tasks, and audit. One JSON snippet per client and they're reading your notebook and claiming kanban cards.

Cloud-hosted MCP servers add 300ms+ of network overhead per call. One call is fine. Twenty calls — which is normal for any non-trivial agent workflow — feel broken. Morion is a local process; tool calls round-trip in under 100ms, typically under 30ms for reads from a warm local store. Half the latency you attribute to the model is the tool-call pipeline; remove it and agents feel sharper.

Every write is logged with the actor name (claude-code, cursor, Mo, user) so you always know who did what. Full MCP reference.

Your external AI clients are great at doing tasks. They are bad at remembering the project. A raw card title loses the sibling tickets, earlier decisions, known risks, unresolved questions, and the reason the work matters. Mo reads the local workspace and turns a card into a work packet an agent can actually use.

That packet can include related tickets in the same group, linked notes, folder memory, recent activity, decisions, risks, questions, and workflow rules. Mo also remains the in-app assistant you can ask directly, and the decision-maker inside Auto-code loops when a build needs to be accepted, reviewed, reopened, or escalated.

Mo runs on your own LLM key: OpenRouter, OpenAI, Anthropic, Groq, or local Ollama. You control the bill and where prompts go. Your data stays on your disk; Mo only sees the slice it needs for the current task. See how Mo briefs your agents.

One stage writes code. A second reviews. Mo supplies context and routes the outcome. A stronger outside agent can inspect the board through MCP and choose the next safe batch. Auto-code makes that pattern explicit: each ticket runs through a visible graph instead of a fragile chat ritual.

The story is not a demo project or a fixed model lineup. A planning chat can create the backlog, a fast model can draft the first pass, a stronger reviewer can inspect the diff, Mo can attach the right context and move the card, and a supervisor can coordinate batches from the terminal.

Closest products hit one or two pieces of this. Cline Kanban has a local board but not a graph. Cursor agents edit code but do not use your local kanban as the queue. LangGraph-style tools compose graphs for developers, not a personal notebook. Auto-code puts the harness on top of the same local board your agents already use. Full Auto-code overview.

Cloud workspaces bill per seat, per query, per stored document. A Notion AI seat is roughly $15–20/month. Mem.ai Pro is $12/month. Devin's self-serve plan is $20/month for ~15 minutes of autonomous work; team plans run $500/month. Cloud-hosted memory layers charge per embedding, per retrieval, per token. The friction shows up as you scaling your agent activity down to fit the budget.

Morion is flat: $0 on Free; $8/mo on Paid (billed annually as $96, or $16/mo monthly) for unlimited boards + per-folder/per-note MCP permissions + Mo on your own LLM key + Auto-code (closed beta) + every future paid feature as it ships. Your agents can make a thousand MCP tool calls while you sleep — Morion's month-end bill is the same. Full pricing.

This matters most when you're a team of one plus three agents, which is increasingly common. The cloud-AI workspace pricing model (per seat, per token) was designed for hundred-person product teams. You shouldn't be paying enterprise rates to run a personal workflow.

Morion's storage format is the most boring thing on this page: a local data file with markdown bodies. Open, well-specified formats with decades of stability ahead of them. If we get hit by a bus, your notes still open. If you decide to leave, export to plain markdown is one menu click and your folder hierarchy comes with you.

Your backup strategy is whatever you already use — Time Machine, restic, Arq, rsync to a NAS. No proprietary backup service, no "premium" export tier, no API quota that locks the door behind you. The whole product passes the test: if Morion shut down tomorrow, would you still own your work? Yes.

This compounds with §1: when nothing of yours is hosted by us, there's also nothing for us to take hostage at price-renewal time. Cloud-tool migrations are painful precisely because the inverse is true.