Odin
Roadmap

What we're building.

Where Odin is today, what's landing next, and what's coming after that. Updated as we ship.

Planned

6

Designed. Not started.

  • Voice cloning

    Train Odin on a sample of your voice and have it reply in your own.

    Voice

  • Team-shared skills

    Skills that travel across teammates so you teach Odin once, the whole team benefits.

    Skills

  • iOS companion

    See and approve background agents from your phone while you’re away from the desk.

    Platform

  • Multiple STT providers

    Choose Deepgram, Whisper, or on-device transcription per workspace.

    Voice

  • Desktop widgets

    Pin commonly-used agent outputs to your desktop so they’re always one glance away.

    Agent

  • Long-context retrieval

    Search across every session, file, and skill Odin has touched.

    Memory

In Progress

4

Active work, near the finish.

  • Skills marketplace

    Install pre-made skills from a registry instead of teaching every one yourself.

    Skills

  • Multi-agent dock

    Multiple background agents running side-by-side with live status, not one-at-a-time.

    Agent

  • ElevenLabs voice selection

    Choose how Odin sounds. Or sound like you.

    Voice

  • Windows parity

    Close the remaining gap between macOS and Windows — AX path, hotkeys, multi-display.

    Platform

Shipped

9

Live for private beta users.

  • Voice loop end-to-end

    Push-to-talk, point at things on screen, reply in natural speech. Sub-second turn-around.

    Voice

  • Act Mode

    Odin drives the UI itself — clicks, types, navigates menus — while you watch each step land.

    Agent

  • Background agents

    Spawn a headless agent that uses tools, writes files, browses the web, and reports back when done.

    Agent

  • Durable memory

    A persistent memory file that survives restarts, archives before replacement, and is injected into every agent task.

    Memory

  • Learnable skills

    Teach Odin a skill once; it applies silently after, without announcing which skill fired.

    Skills

  • Accessibility-API navigation

    Cursor lands on the exact element using the OS accessibility tree, not pixel guessing.

    Agent

  • Multi-display awareness

    Overlay follows you across screens, DPI-aware, per-display.

    Platform

  • Bring-your-own-model

    OpenRouter-backed: pick GPT, Claude, Gemini, or anything else with one setting.

    Platform

  • Silent auto-updates

    Background updates with delta downloads. Nothing to click.

    Platform

Got something you want us to add? Email hello@heyodinai.com .

Latest updates

  1. Rebranded to Odin AI

    New name, new identity, new home at heyodinai.com. The cursor and the menu-bar eye stayed the same.

    Jun 1, 2026

  2. Act Mode gets a new navigation path

    Integrated the cua-driver accessibility path, fixed browser navigation reliability, dropped the rate of mis-clicks substantially.

    May 25, 2026

  3. Background agents v2

    Todos, parallel tool calls, context compaction, a stack of bug fixes. Long-running tasks finish more reliably.

    May 6, 2026

  4. AX element lookup + memory/skills wiring

    OS accessibility tree now informs cursor targeting. Memory and skills are reachable from every mode.

    Apr 30, 2026

  5. Cursor turns teal

    Brand color update — the on-screen pointer is now #00c4b4, matching the eye in the menu bar.

    Apr 29, 2026

  6. Pointing accuracy pass

    Tightened coordinate math so the cursor lands on the element you meant, not the one next to it.

    Apr 29, 2026

  7. v1.0 — memory, skills, SOUL

    First milestone build with persistent memory, learnable skills, and the soul + agent-instructions framework.

    Apr 28, 2026

  8. Act Mode launches

    Odin can now take control of the mouse and keyboard to complete multi-step tasks.

    Apr 21, 2026