Getting started

Accessor is a single small binary, acc. The heavy pieces — ONNX Runtime, the speech recognition model, and the Piper voice — download on first run into Accessor’s own assets folder, so the release artifacts stay around 20 MB.

At a glance

  1. Install acc.
  2. Run acc doctor to check the machine and trigger first-run downloads.
  3. Run acc -wakecode 29 speak.
  4. Say “twenty-nine”, wait for the chime, then speak.

Install

Linux and Apple Silicon macOS

curl -fsSL https://raw.githubusercontent.com/T-Lind/accessor/main/scripts/install.sh | sh

The installer detects the platform, downloads the matching archive, verifies its SHA-256 against the release SHA256SUMS, and installs to $ACC_BIN_DIR (default ~/.local/bin). Pin a version with ACC_VERSION=0.34.3.

Windows x64 (PowerShell)

irm https://raw.githubusercontent.com/T-Lind/accessor/main/scripts/install.ps1 | iex

Installs into %LOCALAPPDATA%\Programs\accessor and adds it to your user PATH. Pass -Version 0.34.3 to pin a release when running a saved copy of the script.

Build from source

Install Rust, then:

cargo install --path . --bin acc --locked
acc doctor

On Windows from a checkout you can also run .\install-accessor.ps1, and the older launcher still works: .\start-accessor.ps1 -WakeCode 29 -Speak.

Unsigned binaries. The macOS build is unnotarized, so Gatekeeper quarantines it on first launch — right-click → Open, or run xattr -d com.apple.quarantine /path/to/acc. The Windows build is unsigned, so SmartScreen warns once (More info → Run anyway).

Platform prerequisites

Platform What you need
Windows x64 Rust MSVC/C++ build tools (source builds only)
Linux x64/ARM64 libasound2-dev libdbus-1-dev libwayland-dev libxcb1-dev libxrandr-dev libxkbcommon-dev libpipewire-0.3-dev libgbm-dev libdrm-dev libegl1-mesa-dev libgles2-mesa-dev pkg-config on Debian/Ubuntu
Apple Silicon macOS Xcode command-line tools
Linux system voices espeak-ng for the System TTS provider
Linux credential store A running Secret Service such as GNOME Keyring

Intel macOS is not shipped prebuilt: there is no ONNX Runtime download entry for x86_64-apple-darwin. Build from source and supply a compatible ONNX Runtime 1.24+ dylib. 32-bit machines are not supported by the provided setup.

First run

The first launch performs one-time downloads and installs the pinned Piper engine plus en_GB-alan-medium:

  • ONNX Runtime 1.24.2 for your OS/CPU
  • Canary 180M Flash INT8 (the default speech recognizer)
  • The Piper engine and default voice

acc setup and acc doctor re-check the speech downloads if any files are still missing. Python is needed only for the optional Kokoro voice worker; Canary, Piper, and the gateway run natively in Rust.

$ acc doctor
$ acc devices          # list microphones
$ acc mic              # report ambient floor, speech level, SNR

Your first conversation

acc -wakecode 29 speak
acc run --wake-code 29 --speak   # equivalent

Say “twenty-nine” or “hey twenty-nine”, pause for the chime, then speak. “Twenty-nine, do this” in one utterance also works. The conversation stays open for 120 seconds of idle time after the last accepted request, reply, or completed work; set --idle-seconds 0 to disable automatic sleep.

The numeric wake word reduces accidental activations; it is not authentication. For a real access boundary, see Local access lock.

The dashboard

Interactive terminals use a fixed dashboard with status, a bounded conversation area, and a command field. Type / to open the command menu; filter by typing, choose with the arrow keys, and press Enter or Tab. You can type ordinary messages directly to the agent without a voice wake code.

Everything day to day is available in the screen:

In-screen command Behavior
/settings Category menu: Voice, Speech, Harnesses, Display, Metaprompt, Tests, Security, Computer
/setup Five-step setup wizard; Enter keeps defaults, Escape cancels
/config, /config set KEY VALUE Inspect and save settings
/tts, /tts provider piper, /tts voice en_GB-alan-medium Configure spoken replies (system, kokoro, piper, cartesia, off)
/tts test, /tts voices [search] Audition or list/search voices
/tts key Masked Cartesia key entry, stored in the OS credential store
/stt test, /stt off Start/stop live transcription testing
/noise Room-noise gate: calibrate, reset, on, off
/prompt Show the metaprompt; preset, extra TEXT, edit, reset
/memory Inspect shared facts; review, infer, forget N
/watch every 30m … Create a recurring, gated survey; list, edit ID, stop ID
/notifications Review raised notifications: list, read ID, readall, dismiss ID
/organizer Notes, alarms, and scheduled tasks (aliases /notes, /alarms, /tasks)
/logs Show the master log path and tail it
/devices List microphones in the conversation area
/connectors Check the agent’s connected apps
/agent Open Codex’s native UI directly
/update Check Codex / Claude Code / Antigravity and apply CLI updates
/status, /help Inspect state or controls
/quit, Ctrl+C Stop Accessor and its agent connection

Page Up/Down scroll the conversation; Escape cancels a task. Use --plain for ordinary terminal output; pipes automatically use plain mode.

CLI quick reference

acc setup                       # first-run setup
acc doctor                      # diagnose install, speech, and harnesses
acc devices                     # list microphones
acc mic [--device NAME]         # measure ambient floor and SNR
acc config show                 # print effective settings
acc config set wake-code 29     # change a setting
acc config locations            # show the settings folder
acc config export portable.json # export portable preferences
acc config import portable.json # merge them into this profile
acc connectors status           # check connected apps
acc agent login                 # sign in to the selected harness
acc organizer status            # notes, alarms, scheduled tasks
acc notifications list --unread
acc memory list
acc journal show
acc update                      # update agent CLIs
acc mcp                         # serve MCP over stdio

Settings live in the OS user configuration directory shown by acc config path. ACC_HOME selects a different settings/event directory; ACC_ASSETS selects a model/runtime directory. Credentials are never stored in config.json.

Choosing speech

New profiles default to local Canary after-wake recognition and fast local Piper speech, so the default path needs no key.

acc tts setup
acc tts test
acc stt test
  • Piper (default): fast local neural speech, pinned build plus one voice.
  • System: Windows System.Speech, macOS say, Linux espeak-ng.
  • Kokoro: local neural speech via an optional Python/ONNX worker.
  • Cartesia: opt-in cloud speech, key in the OS credential store.
  • Off: display replies without speech.

See Features and Voice performance for the trade-offs.

Next steps

Before unattended use, read Local access lock. Accessor protects an Accessor interface, not an unlocked OS account.