Synced from Hive. This page is pulled from hivecommons/hive@v5 during the docs build. Edit the canonical source in the Hive repository.
Input methods and the Inbox
The Inbox is a web page, so anything that can move a pointer and enter text in a browser can use it. This page says plainly which input methods reach the Inbox today, which cannot yet, and why. It backs the quiet row of chips under the Inbox’s source chips (“Works today” and “Neural input (coming soon)”).
Hive is not affiliated with any company or project named here. A listing reflects interest in the input method, not a partnership, an announced integration or a commitment from the project. Names are shown as plain text with a link to the project’s own page. Facts were checked against each project’s own site or repository on 2026-11-17.
What the row does
- The chips are not sources and not connectors. There is nothing to
configure, connect or sign in to, and they never appear under Knowledge →
Integrations, in
knowledge.connectorsor ininbox.sources. - Clicking or focusing a chip shows a short explanation under the row; Escape or Close hides it. Nothing else happens.
- The row shows a handful of entries; more opens this page, which has the full list.
Works today
These tools enter text like a keyboard, or dictate into the focused text box, so they already reach the Inbox: open + Add and use Quick note. Nothing needs to be set up on the hive.
| Name | Kind | What it is | Licence |
|---|---|---|---|
| Talon | voice, gaze | Voice dictation and commands, eye-tracking pointer control and noise input for macOS, Windows and Linux | free to use, not open source |
| OptiKey | gaze | On-screen keyboard for Windows, driven by gaze, a mouse or a webcam | GPL-3.0 |
| ACAT (Assistive Context-Aware Toolkit) | switch | Windows toolkit from Intel: switch and camera input with a scanning keyboard and word prediction | Apache-2.0 |
| macOS Switch Control | switch | Built into macOS: a scanning on-screen keyboard operated by a switch or joystick | part of macOS |
| nerd-dictation | voice | Offline dictation for Linux | GPL-3.0 |
The operating system’s own dictation and any other tool that types into the focused field work the same way. Speech can also arrive as audio: record in the dashboard (Recording into the Inbox) or drop an audio file, and the hive transcribes it when transcription is on.
The Inbox’s small controls (item checkbox, small buttons, pills and the
“+N / less” links) answer over an area of at least 40px (--tap-target), so
they are easier to hit with gaze, dwell, switch scanning or an imprecise
pointer. Their visible size is unchanged.
Neural input
Brain-computer interfaces are the reason this page exists, so the facts first:
- No vendor publishes an API, SDK or export that yields text or speech. No connector can be built today, and hive does not ship a placeholder one.
- People in the clinical trials use these devices to drive an ordinary computer (pointer, on-screen keyboard, or a speech app on a laptop). They reach the Inbox through the dashboard like anyone else, and Quick note takes whatever text entry they already use.
- The devices are investigational. Nothing here is a statement about what they can do beyond what each company says on its own site.
- When a project publishes a way to read text or speech, hive will add it as a source. Until then it is listed as “coming soon”.
Coming soon / exploring
| Name | Kind | What it is | Why it is not a source yet |
|---|---|---|---|
| Neuralink | neural | Implanted brain-computer interface in clinical trials for computer control and speech | no public API, SDK or export |
| Synchron | neural | Implanted brain-computer interface in clinical trials; it has been shown controlling an iPad as an accessibility input | no public API, SDK or export |
| Paradromics | neural | Implanted brain-computer interface in a clinical study designed to turn brain signals into text, a digital voice or device commands | no public API, SDK or export |
| Precision Neuroscience | neural | Investigational brain-surface electrode array (Layer 7) for controlling devices | no public API, SDK or export |
| Blackrock Neurotech | neural | Implantable electrode arrays used in research | its public research SDK (CereLink, BSD-2-Clause) streams raw neural signals, not text |
| Brain-to-Text (UC Davis Neuroprosthetics Lab) | neural | Published research code and data that decode attempted speech into text from implanted electrodes | runs offline on a recorded dataset; needs an implant and a lab to produce new text |
| speechBCI (Willett et al.) | neural | Published research code and data for decoding attempted speech into text from implanted electrodes | runs offline on a recorded dataset |
| silent_speech (Gaddy and Klein) | silent speech | Research code (MIT licence) that turns facial muscle (EMG) signals from silently mouthed speech into text and audio | needs an electrode rig; single-speaker research data |
| AlterEgo | silent speech | Silent-speech wearable | its site offers a waitlist and no public API or SDK |
Open brain-signal toolkits such as BrainFlow, Lab Streaming Layer, OpenViBE, BCI2000, MNE-Python and MetaBCI are not listed in the Inbox: they stream or analyse raw signals and classified intents, not free text, so there is nothing for an inbox to read. A speller built on one of them types on screen like any other keyboard and is covered by “Works today”.
Adding or moving an entry
The catalogue is one Go slice, inputMethods in
src/pkg/inbox/input_methods.go; GET /api/inbox returns it as
input_methods (id, name, kind, state, note, url, featured).
Adding a project is one line there and one row here.
stateisworks_todayorcoming_soon;kindisneural,silent-speech,gaze,switchorvoice.featuredputs the entry in the Inbox row (at most 8, so the row stays readable); the rest are listed on this page only.noteis one factual sentence, checked against the project’s own site. It must not suggest a partnership, an integration or an endorsement; a test rejects those words.urlis an https link to the project’s own page.- An id must not match a connector type or an Inbox source: an input method is never a connector.
When a “coming soon” project publishes an API or export that yields text or audio, remove it from the catalogue and build a real connector for it (Inbox and voice sources).