Settings¶
The options of the /settings page: stored in the database, changed at run time, no deploy. Labels and help are the ones the page shows (here in the English base language).
How the agent works (agent)¶
| Setting | Type | What it does |
|---|---|---|
Commit after each completed task (commit_after_task) |
bool | When a task is closed the agent commits without asking. If off, the work stays uncommitted until you ask. |
Automatic push to GitHub (push_after_commit) |
bool | After the automatic commit it also pushes. If off, you ask for the push. |
Question level (autonomy) |
select: autonomous, essential, ask, many, paranoid |
How many questions the agent asks before it really starts working on a task, from a fully autonomous agent to a paranoid one. The rules of the chosen level are previewed below and written into the agent context when you save. |
Notifications when a task is closed (notify_on_done) |
bool | One switch for both layers: the board notifies the list owner (bell, Web Push, mail) when a task is closed, and the agent is told to notify you on its own channel too. Off = neither of them says anything. |
Notifications for questions (notify_on_question) |
bool | One switch for both layers: the board notifies the list owner (bell, Web Push, mail) when the agent asks a question, and the agent is told to notify you on its own channel too. Off = neither of them says anything. |
Verify before closing (verify_before_close) |
bool | Automatic mobile+desktop screenshots and Livewire tests before closing a task (slower, safer). |
Agent comment (comment_detail) |
select: short, detailed |
How detailed the answer under the note is. |
Response tone (response_tone) |
select: clear, technical, conversational |
How the agent writes in chat and summaries. “Clear and structured” keeps technical accuracy while explaining jargon and using paragraphs, lists and headings when helpful. |
Response length (response_length) |
select: concise, balanced, detailed |
How deeply a normal response should go. It does not limit code, verification or details explicitly requested. |
Git (git_flow) |
select: main, branch_pr |
Where the commits of each task go. |
Evening summary (daily_summary) |
bool | A push notification in the evening with what was closed during the day. |
Summary time (daily_summary_time) |
time | When the evening summary arrives. |
Tick sub-tasks on close (check_subtasks_on_done) |
bool | When the agent closes a task, all its sub-tasks are ticked. |
Task working mode (task_mode) |
select: ordered, multitasking |
Whether the agent works «open to work» tasks one at a time in order, or several in parallel. |
Optimization (optimization)¶
| Setting | Type | What it does |
|---|---|---|
Compact command output (compact_check) |
bool | After --take/--done/--ask/--progress the command prints only the result line, without settings and list (about 700 characters saved per call, which would otherwise stay in the chat for the whole session). |
Terse mode (terse_agent) |
bool | The agent writes almost nothing in the chat (no explanations or recaps), batches commands, reads only what it needs and does not repeat checks. The chat becomes hard to read for a human; the board (agent comment, notifications) keeps working as usual. |
Max context characters (context_max_chars) |
int | Cuts the previous notes/comments printed by the command (resumed items, agent comments) to N characters. 0 = no cut. |
Progress % only piggybacked (progress_piggyback) |
bool | The agent updates the percentage only together with other commands, never with a step just for that (each step costs a whole context re-read). |
Record tokens on close (token_report) |
bool | When closing a task the agent counts and records the tokens spent (one more command per task). Off = time only, no tokens. |
Suggest clearing the session (thousands of tokens) (clear_reminder_k) |
int | The context is re-read at every turn: past a certain weight every single step costs more. When the session goes over this, the agent tells you to run /clear (the agent cannot run it for you). 0 = never. |
App (app)¶
| Setting | Type | What it does |
|---|---|---|
Board language (locale) |
select | Language of the board interface (menus, buttons, settings). Empty = the language of the application (APP_LOCALE). |
Theme (default_style) |
select | Theme used by the board at its main address. |
Maximum title length (title_max_length) |
int | Maximum characters for the title of an item (10–200). |
Automatic archiving (days) (auto_archive_days) |
int | Items completed more than N days ago are archived automatically every night. 0 = never. |
AI description of images (ai_describe_images) |
bool | Every uploaded image is described by an AI model for the search (needs an API key in .env). |
AI provider for images (ai_image_provider) |
select | Which provider describes the images. |
AI model for images (ai_image_model) |
text | Model name (empty = cheapest of the provider / AI_IMAGE_MODEL). |
Toast for console changes (toast_console_changes) |
bool | On-page notice when the agent changes the state of an item (working, done, question). |
Board mode (mode) |
select | Server = login required, each user has their own lists (access can be restricted, see config). Local = NO authentication, one global set of lists: only for a board on your own machine. Empty = as set in the config (GRIGLIA_MODE). |
DASHBOARD side tab (show_dashboard_tab) |
bool | Show the slide-out dashboard tab at the edge of the window (desktop). |
Speech to text (speech_mode) |
select | Auto = server transcription (AI SDK, OpenAI…) when configured, otherwise the browser. Server = record and transcribe on the server (better quality, works on every browser with a microphone). Browser = the browser's own recognition (free, no server). |
Price per 1M input tokens (cost_per_m_in) |
text | Used by the statistics to turn tokens into cost (e.g. 3 for 3 €/M). 0 = cost not shown. |
Price per 1M output tokens (cost_per_m_out) |
text | Same, for output tokens (e.g. 15). |
Currency (cost_currency) |
text | Symbol or code shown next to costs (EUR, $, …). |
Generate instruction files from the board (context_sync) |
bool | When off, the host sync restores the original CLAUDE.md / AGENTS.md and stops touching them. |
In-app notifications (notify_in_app) |
bool | Bell at the top of the board with what the agent did (task closed, question asked), live. |
Web Push notifications (notify_webpush) |
bool | Notifications on the devices where you enabled them (button below), even with the app closed. iPhone: add the app to the Home screen first. |
E-mail notifications (notify_mail) |
bool | Also by e-mail to your account address (needs a configured mailer: MAIL_MAILER). |
Dashboard tab side (tab_side) |
select | Which side of the window the slide-out dashboard tab opens from (desktop). |