Mira

Mira is Undra’s built-in assistant. Mira works inside your workspace: reading your notes and plans, answering from that material, and making changes you can see and undo.

Mira runs on the AI provider you connect: a Claude, OpenAI, or Cursor subscription, or a local model on your own machine. See Setting up AI.

Mira, and a team when needed

There is no roster of helpers to learn or manage. You talk to Mira. Ask for a write-up, a status report, or a diagram and Mira handles it. For a big job Mira can quietly split the work across several helpers running at once. You watch them in a live view (each one’s task, tools, and progress), and one Undo covers everything they did (see Chats).

What Mira can do

• Answer from your work. Ask about the current note, a selection, or your library, and get an answer grounded in your real material rather than the open web.
• Summarize and draft. Turn a messy meeting note into a summary, draft a brief from scattered ideas, or pull the open tasks out of a project.
• Edit your workspace. Create a note, update a plan, or fix wording across files. Those writes land as Mira works, show up in chat, and stay undoable.
• Build a whole plan. Mira can create an entire plan in one step: a kanban item pre-filled with task cards, ready to refine.
• Remember what matters. Tell Mira a durable fact or preference once (how you like your summaries, what your projects are called) and it keeps a long-term memory it carries into later chats. Memories are real editable notes in a Mira/Memory folder, Mira suggests new ones for your approval, you can pin hard rules it must always follow, and you can review or clear everything in Settings → AI → Memory. See What Mira remembers.

Grounded in your workspace

A generic chatbot forgets you the moment the tab closes. Mira works from your actual files: the note in front of you, the items you select, the things they link to. Ask “what’s left before the Phoenix launch?” and Mira reads the real plan instead of guessing.

Write, see, undo

Trust here is built on reversibility, not on approving every keystroke first.

• Workspace items (notes, plans, canvases) are created and updated as Mira works. You see what changed in chat, and an “AI made N changes, Undo” bar can roll the whole run back in one click.
• Folder portals use the same undo idea, with risk-scaled gates: safe file edits (create, edit, rename, move) auto-apply by default with a backup; deletions always ask before anything is removed. You can turn off auto-apply in Settings → AI if you want every portal change set to wait for Apply. Details live in Folder Portals.

That safety net is what makes leaning on Mira a low-stakes decision rather than a leap of faith.

Skills

A skill is reusable guidance Mira carries into the work: your house style, formatting rules, how you like release notes written. Instead of repeating the same instructions in every prompt, you write them once and Mira applies them when they fit.

In Undra, a skill is just a note. Right-click any note and choose Use as skill: its content becomes standing instructions, and a short description on the note tells Mira when to reach for it. Skills live in your folders like any other note, show a badge in the sidebar, and are edited in the normal editor.

Mira keeps skills honest over time: it ranks them by how well they are actually working, offers to sharpen one whose guidance is getting weak, and offers to retire ones it never reaches for, so your skill set stays useful instead of piling up.

Already keep personal skills for Claude Code? An opt-in importer copies the ones you pick into Undra as skill notes.

Custom agents and routines

In the current release, Undra ships Mira only. Creating your own saved agents, pinning specialists beside Mira, and scheduling routines are not available in production builds yet.

What you can do today: chat with Mira, teach lasting guidance with skill notes, and let Mira pull notes and plans into the answer.

Trust is the point

Mira should make you faster without quietly taking ownership of your files. Your workspace stays on your machine. Changes are inspectable and undoable, and the riskiest portal actions still ask first. That is what makes leaning on AI here feel safe rather than risky. See Local-First Trust for the foundation underneath it.

Where to go next

• Chats: where you talk to Mira, and how conversations are saved.
• What Mira remembers: the long-term memory behind Mira, and how to manage it.
• Notes: the material Mira reads and writes, including notes flagged as skills.
• Setting up AI: connect a provider and turn AI on.
• Local-First Trust: why AI here can be powerful and safe.