From Your Voice to Your Coding Agent
1. Dictate on your phone
Speak an idea or a prompt, in English, Chinese, or both in one sentence. The iPhone recognizes it on the device itself, with iOS or the speech model inside Fastxt: the audio never leaves the phone and is never stored. Correct it by typing, and keep it as an idea or a prompt.
2. Send it to a session
Pair the phone with your computer and pick one of the coding-agent sessions you have open there in Orca: Claude Code, Codex and the others it runs. Fastxt types the prompt in, as if you had, and never into a session that is asking you a question.
3. Or let the agent take it
Any agent with Fastxt's MCP server can take your newest prompt when you ask; in Claude Code, that's /mcp__fastxt__take. Every device records which agent had which prompt, and in which project.
![[object Object]](/img/fastxt-sync-server-qr-mac.png)
No Server in Between
Your notes live in a SQLite database on each of your devices and sync directly between them, over an encrypted connection pinned by a pairing code. No account, no Fastxt server.
On-Device AI
Tags and summaries on the phone, from Apple's on-device models; semantic search and categories too on the desktop, from a model server on your own machine, such as Ollama.
Cross Platform
iPhone and iPad, Android, macOS, Windows and Linux, plus one fastxt command for your computer. Devices on the same network pair by scanning a QR code.
Roadmap
Android
Dictation and sending to coding agents on Android, with the same speech model.
Semantic Search on the Phone
One multilingual embedding model on every device, so the phones search by meaning too.
