ClawVoice Agent Voice + TTS

AI voice and x402 wallet capabilities for OpenClaw agents.

ClawVoice by ForgeMesh Labs gives OpenClaw agents voice output, optional push-to-talk voice input, and a local Base USDC x402 wallet for paid agent calls.

Package name: @forgemeshlabs/x402-agent-voice

Commands: clawvoice or x402-agent-voice

Homepage: https://forgemesh.io

Feedback: [email protected]

What It Does

• Text-to-speech for OpenClaw agents.
• Local, hosted, or hybrid voice modes.
• Optional push-to-talk speech-to-text on supported platforms.
• x402 payments for hosted ForgeMesh Voice fallback.
• Local Base USDC x402 wallet for paid agent calls.
• Spend caps, balance checks, and withdrawal for leftover USDC.
• Plug-and-play OpenClaw / ClawHub skill packaging.

Search terms: ClawVoice, OpenClaw Voice, OpenClaw Agent Voice, AI Voice, Agent Voice, x402 Voice, Wallet Voice, Voice Skill, OpenClaw Skill.

Install

From the release tarball:

npm install -g ./forgemeshlabs-x402-agent-voice-0.3.10.tgz

Then run setup:

clawvoice init --mode hybrid --mic

Use hosted mode if you do not want local voice setup:

clawvoice init --mode hosted

The old command still works:

x402-agent-voice init --mode hybrid --mic

Register With OpenClaw

Global npm installs are not automatically scanned by OpenClaw. To make ClawVoice appear in the OpenClaw Skills sidebar, link the installed package into your OpenClaw skills folder:

mkdir -p ~/.openclaw/skills
ln -s "$(npm root -g)/@forgemeshlabs/x402-agent-voice" ~/.openclaw/skills/clawvoice-x402
openclaw gateway restart

Use the restart command your OpenClaw install supports if it differs. The symlink keeps future npm update -g @forgemeshlabs/x402-agent-voice updates connected to the same skill registration.

What Setup Does

Setup asks how your agent should speak:

| Mode | Meaning | |---|---| | local | Free local voice after setup. The voice runtime runs on your machine. | | hosted | No local voice install. Your agent pays ForgeMesh Voice per call through x402. | | hybrid | Local first, hosted fallback. This is the recommended mode. |

Setup creates a unique local Base USDC wallet the first time it runs for a user profile. If ~/.x402-agent-voice/wallet.json already exists, setup reuses that wallet instead of replacing it.

Each machine, container, or user profile gets its own wallet unless you deliberately copy the local wallet file. Do not share wallet.json, check it into git, paste it into chat, or upload it to support.

Quick Commands

clawvoice voice-check
clawvoice wallet
clawvoice balance
clawvoice withdraw --to 0xYourWallet --amount all
clawvoice products
clawvoice speak "hello world"

wallet shows the public address, chain, token, and spend policy. It does not show the private key.

Recover Leftover Funds

The agent wallet is meant to hold a small working balance. If there is leftover USDC, send it back to your main wallet:

clawvoice withdraw --to 0xYourWallet --amount all

You can also withdraw a specific USDC amount:

clawvoice withdraw --to 0xYourWallet --amount 1.25

Withdrawals send Base USDC from the local agent hot wallet. The command prints a summary first and asks for confirmation before broadcasting. Use --yes only for automation you trust.

Important: USDC transfers on Base require a small amount of Base ETH for gas. x402 payments usually do not require you to think about ETH, but a normal wallet-to-wallet USDC transfer does.

Voice Input

Push-to-talk voice input is optional functionality.

clawvoice install-mic
clawvoice listen
clawvoice talk --agent "claude -p"

listen records from the microphone until you press Enter, then transcribes locally with faster-whisper.

talk runs a conversation loop:

record your voice -> transcribe locally -> send words to your agent -> speak the reply

Wake word and always-on listening are not shipped yet.

Platform Support

| Capability | macOS | Linux | Windows | |---|---|---|---| | Wallet creation | Supported | Supported | Supported | | Hosted x402 voice | Supported | Supported | Supported | | Local text-to-speech | Supported | Supported | Use WSL2 | | Playback | Supported | Supported | Partial / WSL2 recommended | | Push-to-talk voice input | Supported with FFmpeg + faster-whisper | Supported with FFmpeg + faster-whisper | WSL2 recommended | | Wake word / always-on listening | Not shipped | Not shipped | Not shipped |

Windows users can use wallet and hosted voice flows. Native Windows microphone recording, local voice install, and playback still need a compatibility pass. Use WSL2 for the full local path.

Spend Safety

Hosted voice and other paid x402 calls are guarded before any payment is signed.

Default policy:

| Setting | Default | |---|---:| | Per-call cap | $0.05 | | Daily cap | $0.25 | | Session voice cap | $0.10 | | Approval before paid calls | On |

Spend records are stored locally at:

~/.x402-agent-voice/spend.jsonl

Configuration is stored locally at:

~/.x402-agent-voice/config.json

Local Voice

For local or hybrid mode:

clawvoice install-voice

To see what it would do first:

clawvoice install-voice --dry-run

Local voice may download Python packages and model files. It needs Python 3.9-3.12 (onnxruntime has no wheels for 3.13+); the installer probes python3.12 through python3, or you can point X402_VOICE_PYTHON at a specific interpreter. Hosted mode avoids that setup and uses x402 payments instead.

Built-In ForgeMesh Products

The package includes a starter list of ForgeMesh x402 products:

clawvoice products

AgentCash and X402Scan discovery can still be used for broader x402 discovery.

Security Notes

• Treat the generated wallet as a small-balance hot wallet.
• Each fresh user profile gets a unique local wallet. Reinstalls reuse the existing local wallet unless you reset local state.
• Do not fund it with more than you are comfortable letting an agent spend.
• Do not share ~/.x402-agent-voice/wallet.json; it contains the private key.
• Use clawvoice withdraw --to 0xYourWallet --amount all to recover leftover USDC.
• The private key stays on your machine.
• The private key is not printed by setup, wallet, voice-check, or normal commands.
• Paid calls require approval unless you explicitly pass --approve.

Feedback

Please send feedback, bugs, and confusing setup steps to:

[email protected]

Helpful details:

• OS and Node version.
• Setup mode: local, hosted, or hybrid.
• Whether voice-check reports local server, FFmpeg, and Whisper as ready.
• Whether speak, listen, and talk worked.
• Any payment approval prompt that felt unclear.