The arrival of the agent as your eyes.
A screen reader that reads, keyboard nav, and high-contrast themes have always been here. 1.0 crosses the real frontier: the AI does the looking and the reaching. It reviews the diff, tours the repo, and reads the failing build — narrating each one back — while you drive the whole app by keyboard or voice.
Everything that landed in 1.0.
Three brand-new capabilities lead the release; the mature foundation is refined around them.
- NewThe agent as your eyes — Change Review, a Codebase Orientation Tour, and Narrated Run & Verify
- NewVoice-command grammar — drive the entire app by speech, with spoken confirmation before anything risky
- NewA guided tutorial that’s authored and fully offline, and resumes where you left off
- RefinedRead-aloud and dictation, with a fully local option so your audio never leaves your machine
- RefinedConsent-gated PeerLink pairing — view, prompt, and approve granted separately and revocable instantly
- RefinedSafe-Deny approvals, keys in OS secure storage, and a host-only, offline-capable core
The AI does the looking, so you don’t have to.
A screen reader that reads, keyboard nav, high-contrast themes — that’s table stakes, and Accessible Claude has always had it. 1.0 crosses the real frontier: using the AI as your eyes and your hands. It does the seeing — reviewing a diff, orienting in a new repo, reading a failing build — and narrates it back to you.
Change Review
Walk me through what you just changed.
The glance you’re missing
You can’t scan a diff. When an agent edits ten files, verifying its work non-visually is the single hardest, most trust-critical moment in the loop.
A navigable, spoken review of the agent’s own edits — drill from turn to file to change. It narrates the structure (“three files changed, 40 lines added”), the content, and an on-demand “why” for any edit. Undo at a coarse grain if you don’t like what you hear. Sourced from the agent’s edits directly — no git required.
Codebase Orientation Tour
Give me a tour of this project.
The glance you’re missing
Dropped into an unfamiliar repo, a sighted developer skims the file tree and gets their bearings in seconds. Without that glance, orientation is slow and disorienting.
A narrated, queryable map of the codebase — architecture, entry points, the key modules, how data flows. Navigable and conversational: “where is authentication handled?”, “what calls this function?” You build a mental model by asking and listening, not by tabbing blindly through hundreds of files.
Narrated Run & Verify
Run it and tell me what happened.
The glance you’re missing
“Did it pass?” shouldn’t mean paging a screen reader through a thousand lines of build and test log to find the one line that matters.
The agent runs your build, tests, or app and narrates the outcome — passed or failed, the single error that actually matters, and a suggested fix — instead of dumping logs. The raw output is always one command away when you want it. You get the signal, not the noise.
Drive the whole app by voice.
Hold one command key and speak. A deterministic, offline grammar navigates, reads, and acts — and echoes anything risky back for a one-word confirmation before it happens.
Navigate
instantMove around the app without a keystroke.
- go to settings
- show the tour
- go to changes
- new session
Move & read
instantStep through lists and control what’s spoken.
- next change
- read the last response
- read the error
- stop
- slower
Act
confirmed if riskyDecide on the agent’s requests, out loud.
- approve
- deny
- undo that
The foundation, refined.
The accessible core that everything else stands on — the tutorial, the voice layer, pairing, and the security model.
Drive the whole app by voice
Hold one key and speak — “go to changes”, “read the error”, “approve”. A deterministic, offline command grammar navigates, reads, and acts, with a spoken confirmation before anything risky.
The agent as your eyes
Ask it to walk you through what it changed, give you a spoken tour of an unfamiliar codebase, or run the build and tell you the one error that matters — never a wall of logs.
Screen-reader-first from the ground up
A streaming transcript that never floods or steals focus, safe-Deny approvals, high-contrast themes, and a fully keyboard-operable interface — verified with NVDA and Narrator.
Read aloud and dictate
Hear responses spoken and dictate prompts, using cloud voices or a fully local, offline engine — your audio need never leave your machine.
Consent-gated pair programming
Invite a trusted person to view, prompt, or approve on your session — each capability granted explicitly and revocable instantly, so you stay in control.
A guided tutorial that works offline
Learn the app hands-on with an authored, screen-reader-friendly walkthrough — no API key, no network, resumes where you left off.
This is 1.0. Come and hear it.
Download Accessible Claude for Windows, or join the waitlist for what comes next.