Agent Terminal

Agent Terminal is a voice-to-agent developer terminal for AI-native coding sessions.

It is built around one simple operating model:

1. capture intent — typed or via voice through Agent Wispr
2. enrich with semantic memory from Agent Brain
3. run the pre-send guard to validate the request
4. delegate bounded coding tasks to Agent Harness workers

What it is not

It is not a general chat UI, a hosted IDE plugin, or a vague "AI assistant."

The current product path is deliberately focused:

• standalone desktop app first
• reliable daily-driver for solo developers
• voice + memory + execution in one terminal
• auditable pre-send guard with no side effects

Public documentation boundary

These docs describe the shipped desktop workflow and supported integrations. They intentionally omit private prompts, credentials, deployment addresses, and unreleased implementation details.

Product priorities

The current release focuses on a reliable local operator loop: capture an intent, enrich it with available context, review the guard decision, and route bounded work to the configured execution service. The public roadmap covers future improvements without exposing private experiments or evaluation plans.