Configuration

All configuration and data is stored in ~/.tota/.

File / Directory	Description
`~/.tota/config.json`	Main config: agent name, user name, providers, models, Telegram token
`~/.tota/tota.yaml`	Extended config: loop guard, web search, MCP servers, API channel
`~/.tota/.env`	API keys and tokens (sourced automatically)
`~/.tota/memory/`	SQLite Second Brain database and FTS index
`~/.tota/daemon.pid`	PID of the running background daemon
`~/.tota/daemon.log`	Background mode logs (last 1000 lines kept, then rotated)
`~/.tota/daemon-error.log`	Error log for the daemon process (macOS LaunchAgent only)
`~/.tota/skills/`	Installed custom skill directories
`~/.tota/tasks.json`	Persisted scheduled tasks

Nothing is stored in your project directory. You can delete ~/.tota/ to start completely fresh.

Web Search

When WEB_SEARCH_PROVIDER=auto, tota picks whichever key it finds first (Brave → Serper → Tavily).

Env Var	Description	Default
`API_CHANNEL_ENABLED`	Enable the HTTP API channel	`false`
`API_CHANNEL_PORT`	Port to listen on	`3001`
`API_CHANNEL_KEY`	Optional bearer token for auth	—

Browser tools work out of the box — no API keys needed. Install browser binaries once:

npx playwright install chromium firefox webkit
# or via the wizard:
tota setup browser

Env Var	Description	Default
`BROWSER_ENGINE`	Default engine: `chromium`, `firefox`, or `webkit`	`chromium`
`PLAYWRIGHT_HEADLESS`	Set to `true` to run without a visible window	`false`

Tool	Description
`browser_open`	Open a URL in the browser (visible window by default)
`browser_click`	Click an element (selector or visible text)
`browser_type`	Type text into an input field (SPA-safe)
`browser_key`	Press a keyboard key (`Enter`, `Tab`, `Escape`, combos)
`browser_wait`	Wait for a CSS selector or navigation
`browser_screenshot`	Capture the current page
`browser_extract`	Extract text / attributes via CSS selector
`browser_scroll`	Scroll up, down, left, or right
`browser_close`	Close the shared browser instance
`browser_engine`	Switch engine to `chromium`, `firefox`, or `webkit`

Env Var	Description	Default
`COMPUTER_USE_ENABLED`	Enable computer-use and Android ADB tools	`false`

Or in ~/.tota/tota.yaml:

capabilities:
  computer:
    enabled: true

Enable via wizard: tota setup computer

Uses @nut-tree-fork/nut-js. On Linux, first run sudo apt install libxtst-dev.

Tool	Description
`computer_screenshot`	Capture the full screen or a region
`computer_see`	Screenshot + vision AI analysis in one step
`computer_click`	Left, right, or double-click at (x, y)
`computer_move`	Move cursor to (x, y)
`computer_type`	Type text at the current keyboard focus
`computer_key`	Press a key or combo (`cmd+c`, `ctrl+z`, `enter`, etc.)
`computer_scroll`	Scroll at a position
`computer_drag`	Click and drag between two coordinates
`computer_screen_size`	Get display resolution

Requires adb in your PATH and a connected device or emulator.

Tool	Description
`adb_devices`	List connected Android devices
`adb_screenshot`	Take a screenshot of an Android device
`adb_see`	Android screenshot + vision AI analysis
`adb_tap`	Tap at (x, y) on the device
`adb_swipe`	Swipe between two points
`adb_type`	Type text into the focused field
`adb_key`	Send key event (3=HOME, 4=BACK, 66=ENTER…)
`adb_shell`	Run any `adb shell` command
`adb_pull`	Pull a file from the device
`adb_push`	Push a local file to the device

Env Var	Description	Default
`LOOP_MAX_STEPS`	Max agentic steps per request	`50`
`LOOP_ABSOLUTE_MAX`	Hard cap on total tool calls	`100`
`LOOP_FAILED_ABSOLUTE_MAX`	Stop after this many failed calls	`25`
`LOOP_IDENTICAL_THRESHOLD`	Consecutive identical calls before warning	`5`
`LOOP_SIMILAR_THRESHOLD`	Consecutive failing calls before warning	`8`
`LOOP_SAME_TOOL_THRESHOLD`	Consecutive same-tool calls before warning	`10`
`LOOP_NO_ACTION_MAX`	Steps with no tool calls before stopping	`10`
`LOOP_TEXT_REPEAT_THRESHOLD`	Repeated step texts before warning	`3`

mcp:
  servers:
    - name: my-tools
      url: http://localhost:8080/mcp
      enabled: true
      apiKey: optional-bearer-token

Each server must implement the Model Context Protocol JSON-RPC tools/list and tools/call methods over HTTP.