Software Agent v0.7.0 - Talk to Nova

Nova voice conversations

• Press Ctrl+R or run /voice inside the project room to start explicit push-to-talk.
• Speak for up to two minutes, press Enter to transcribe, edit the text, then press Enter again to submit it.
• Nova speaks the committed response for that exact task, so unrelated background agent events are never selected.
• Ask the full team or select a specific active agent with Tab or /target before recording.

Private and governed by design

• The microphone stays dormant until the user explicitly starts voice input.
• Captured audio remains bounded in memory and is zeroed after transcription, cancellation, validation failure, or abort.
• --offline blocks voice before configuration, credential, microphone, or network access.
• OpenAI credentials remain in the configured OS credential store and are cleared from the bounded request lease.
• Voice does not bypass Software Agent approvals or tool policy.

Easier terminal experience

• Nova is discoverable from /, help, settings, setup, footers, and doctor --json.
• README and installation docs now include copy-paste setup, the complete voice workflow, technology highlights, privacy boundaries, and troubleshooting.
• Uses the official Picovoice PvRecorder Node SDK, OpenAI transcription, OpenAI speech generation, and native OS WAV playback.

Voice requires a configured OpenAI API key. Coding agents may still use OpenAI, Anthropic, or per-role routes. Voice-provider usage is separate from the controller run-token ledger.

Verification

• 134 TypeScript tests passed across 15 files.
• 60 Python/MCP compatibility tests passed.
• Typecheck, ESLint, production build, clean tarball install, doctor --json, and dependency audit passed.
• SHA-256: 94A4FB44BA759AA17E4E8D88C64FB96B385B678BFDC100D780395B5C26361DD3

Install

npm install -g https://github.com/khajaaijaz26/software-agent/releases/download/v0.7.0/software-agent-0.7.0.tgz
software-agent --version