What problem does Agent Terminal solve?

AI voice-to-agent developer terminal

Speak a goal. Route it to memory and execution. Stay in flow.

Capture voice, classify intent, enrich with semantic memory, and launch bounded coding tasks — all from a single desktop terminal built for AI-native development.

Desktop app optimized for solo developers using voice + local AI in daily coding sessions.

Start Here Read the Docs View GitHub

Pair the terminal with the free,MIT-licensed DSH Agent Harness pluginwhen local model execution is the right fit.

agent-terminal://sessioncomposer tab

# 1. clone and start
git clone https://github.com/joshl26/agentterminal
npm install && npm run tauri dev

# 2. open Composer tab, type your goal
# Brain recall enriches context automatically

# 3. pre-send guard validates the request
# then routes to Agent Harness workers

# 4. check Runs tab for live status

UI tabs: Composer, Runs, History, Transcripts, Context

162

tests protecting the core session flow, auth, and SSO

Tauri

native desktop shell on Windows, macOS, and Linux

Local-first

local-first, scales to team operation

See it in action

Watch a real session, not a script.

Agent Terminal overview

See voice capture, project memory, and bounded task delegation working together in one session.

What it does

Voice, memory, and execution in one terminal — not three tools.

Agent Terminal is a real, shipping desktop application — not a demo. It captures voice through Agent Wispr, recalls project memory from Agent Brain before every session, validates each outgoing request with a pre-send guard, and delegates bounded coding tasks to Agent Harness workers with live status you can watch in the Runs tab.

Voice transcript ingestion

Capture speech via Agent Wispr, classify intent, and route transcripts directly into the terminal session without breaking flow.

Semantic memory integration

Recall relevant project context from Agent Brain before each session — facts, decisions, and prior work surface automatically.

Harness task launch

Compose a goal in the terminal, run the pre-send guard, and delegate bounded coding tasks directly to Agent Harness workers.

Persistent session history

Every session and turn is persisted to a local SQLite database. Browse and replay past sessions from the History tab.

Native Tauri desktop shell

Runs as a first-class desktop application on Windows (WSL2), macOS, and Linux — not a browser tab.

Pre-send guard

A pure, side-effect-free guard evaluates every outgoing message before dispatch — blocking risky or malformed requests before they leave.

Single sign-on across the suite

Log in once and stay authenticated across Brain, Terminal, and Harness. Tokens are stored in the OS credential store — no re-login when switching apps.

How it works

Start with a goal. Add memory and execution only where it helps.

Speak or type a goal

Use Agent Wispr to capture voice transcripts, or type directly in the Composer tab. The terminal classifies intent and prepares context.

Enrich with memory

Agent Brain recalls relevant project facts, prior decisions, and session history. The enriched context is attached before any task is launched.

Run the pre-send guard

Every outgoing message passes through a pure guard function that validates scope, detects risky patterns, and blocks malformed requests.

Launch to Harness, review in Runs

Bounded coding tasks are dispatched to Agent Harness workers. Live run status appears in the Runs tab with full attempt history.

Tauri desktop shell

Voice ingestion via Agent Wispr

Semantic memory via Agent Brain

Pre-send guard (pure, no side effects)

Harness task launcher

SQLite session history

First-run shape

Start with `npm run tauri dev`, open the Composer tab, type a goal. Brain recall and Harness delegation are available immediately — add Wispr voice input only when you want hands-free operation.

Positioning