I voice-code from my phone while walking my dog
I dictate coding tasks to Claude Code via Telegram voice notes while walking my dog. Voice input is 4x faster than typing on a phone.
Read moreCode from anywhere with Claude Code, Codex CLI, Gemini CLI, OpenCode, Pi, and Amp
A realistic Telegram phone mockup demonstrating Untether. The animation shows switching between project chats, sending text and voice messages, receiving tool call progress, approving actions, and getting results — all within Telegram.
Untether Bot VPS
starting · claude code · 0s 🟧 dir: lba-web
Untether Bot VPS
working · claude code · 2m 28s · step 24 ✓ for file in...
Untether Bot VPS
working · codex · 6m 17s · step 124 ✓ grep -rn...
Untether Bot VPS
finished · claude code · 4m 52s · step 67
Untether Bot VPS
working · opencode · 32m 14s · step 203 ✓ npm run...
Untether Bot Local
finished · claude code · 1m 38s · step 12
Untether Bot VPS
finished · claude code · 8m 45s · step 89
Untether Bot VPS
cancelled · claude code · 5m 07s · step 12
Voice, text, files, and approvals — all from Telegram, wherever you are.
Every engine gets the core features. Some support deeper integration.
| Feature | ||||||
|---|---|---|---|---|---|---|
Progress streaming | ||||||
Session resume | ||||||
Model override | 1 | |||||
Model in footer | — | — | ||||
Approval mode in footer | 4 | — | — | 2 | — | |
Voice input | ||||||
Verbose progress | ||||||
Error hints | ||||||
Preamble injection | ||||||
Cost tracking | 3 | 3 | 3 | 3 | ||
Interactive permissions | — | — | — | — | — | |
Approval policy | 4 | — | — | 2 | — | |
Plan mode | — | — | — | — | — | |
Ask mode (option buttons) | — | — | — | — | — | |
Diff preview | — | — | — | — | — | |
Auto-approve safe tools | — | — | — | — | — | |
Progressive cooldown | — | — | — | — | — | |
Subscription usage | — | — | — | — | — | |
Reasoning/effort levels | — | — | — | — | ||
Device re-auth (/auth) | — | — | — | — | — | |
Context compaction | — | — | — | — | — | |
Cross-env resume (/continue) | 5 | 6 |
1 Amp model override maps to --mode (deep/free/rush/smart)
2 Toggle via /config between read-only, edit files (--approval-mode=auto_edit, files OK but no shell), and full access (--approval-mode=yolo, default in headless mode); pre-run policy, not interactive mid-run approval
3 Token usage counts only — no USD cost reporting
4 Toggle via /config between full auto (default) and safe (--ask-for-approval=untrusted, untrusted tools blocked); pre-run policy, not interactive mid-run approval
5 Pi requires provider = "openai-codex" in engine config for OAuth subscriptions in headless mode
6 Amp requires an explicit thread ID; no "most recent" mode
Requires a AI Coding Agents account
Copy this prompt and paste it into any AI assistant:
"I was walking Normi and had an idea for a fix. By the time I got home and opened my laptop, I'd forgotten half of it."
Largest release since initial publication — 147 issues across 14 release candidates. Three workflow modes (assistant/workspace/handoff), cross-environment resume (/continue), /config UX overhaul, agent-initiated file delivery (outbox), interactive plan outline with hold-open approval, Codex/Gemini approval policies, auto-continue for Claude Code, stall diagnostics with CPU-aware suppression, orphan progress cleanup, cost budget toggles, model metadata for all 6 engines, and 60+ bug fixes.
Acknowledged ideas and investigations — not yet scheduled.
Also by Little Bear Apps