Everything Vendetta does

Grouped by where you are: at your desk, away from it, with other people, on a schedule, and what you can add with plugins.

At your desk

The chooser and the overview live in your terminal. The HUD and dictation sit over your windows.

  • The chooser

    vendetta tui, or a key you bind in the config, lists your projects with how many agents wait, work, and are ready. Every row has a jump hint, and one more key takes you to that agent.

    Docs

  • A preview you can type into

    The selected agent's terminal shows live beside the list. Type into it without leaving the chooser.

    Docs

  • Compact mode over SSH

    Below 100 columns the chooser drops to a compact layout, down to 20×6. It works over SSH from a phone.

    Docs

  • The overview on a phone, on its Needs you tab: an approval and a question, each answered with the buttons on its card

    The overview

    Every agent in lanes: needs you, working, ready, ended today. Answer a question or an approval on its card.

    Docs

  • Search on a phone: a word typed above the chat list finds the messages that say it

    Search, filters, and bulk actions

    Search across names, paths, providers, and states, or filter to the agents that need you. Select several agents and act on them together.

    Docs

  • History on a phone: ended conversations stay listed under the chats, each one a tap away from being restored

    Restore

    Ended conversations stay listed for a week. Enter, or a message in the chat or topic, restores the exact one.

    Docs

  • An agent's change on a phone: the files it touched and the diff of one of them, with a button on each line to mark it

    Changes

    Committed work shows as a change with its revisions and merged state, as a unified or side-by-side diff. Mark a line, a range, or a file, and the comment follows the code as it moves.

    Docs

  • The Issues page: open issues and pull requests from several repositories, each with its project, kind, and age, under filters for project, kind, state, and role
    The Issues page on a phone: open issues and pull requests from several repositories, with filters above the list

    Issues and pull requests

    The forges plugin lists the open issues and pull requests of your projects' repositories on GitHub, GitLab, and Forgejo, and those closed in the last 30 days. Filter by project, kind, state, and what is assigned to you or opened by you.

    Docs

  • One issue's page: New agent, Send to agent, Copy link, and Open above its description and latest comments
    One issue's page on a phone: New agent, Send to agent, Copy link, and Open above its description

    Start an agent on an issue

    An item opens with its description and latest comments. New agent starts one in its project with the link in the first message, Send to agent hands it to one that is running, and Copy link is in each row's menu.

    Docs

  • Agent states in tmux tabs

    Each project gets an Agents tab, and tab labels carry each agent's state in the tmux status bar. They show over SSH too.

    Docs

  • The desktop HUD: a card listing eight agents with their states, a to-do card with four tasks, and a notifications card where one agent needs approval and another asks a question

    Desktop HUD

    Cards over your windows for agents, to-dos, and alerts. They never take the keyboard.

    Docs

Your own HUD section
Any program can add a section to the HUD by printing JSON. Docs
Dictation
Press a key and talk, and the words go to the agent in the focused window, or are pasted where there is none. A second key sends to the Master. On macOS dictation always pastes. Docs
Layouts and keys
Arrange the chooser's panels, or keep every project in one tree. Every key can be rebound. Docs

Away from it

Every agent is a chat on the iPhone and in a browser. Telegram and Discord carry the same conversations.

  • Questions as cards

    Single or multi-select with a recommended option, steps with a review page, your own answer, or a note. The answer goes to that exact process.

    Docs

  • Approvals

    Claude Code's permission dialog arrives with the command, the reason, and its own buttons. Only you answer it.

    Docs

  • An agent in its own terminal pane: it reads a deploy file, asks where the limiter should keep its counters with three options, and once Redis is answered it writes the limiter
    Your terminal
    On iPhone, an agent adding rate limiting is working; it asks where the limiter should keep its counters, Redis is chosen, and it works on. Shown as a sequence of states.
    Your iPhone
    In the web client on a phone, an agent that was working asks where the limiter should keep its counters; Redis is tapped and the agent works on
    A browser

    One conversation on every screen

    The terminal, the browser, and the phone show the same agent, and a prompt from one appears in the others. Delivery marks say Waiting, Read, or Not delivered.

    Docs

  • Voice notes

    Hold the mic and release to send, slide left to cancel, or slide up to lock and keep talking. The transcript shows under the waveform, and a note that fails is kept with Retry.

    Docs

  • Listen

    Any reply read aloud, skipping code, tables, and thinking. A player bar pauses, jumps 15 seconds, and speeds up to 2×.

    Docs

  • Voice mode with the Master on iPhone: a sheet over the chat list with a microphone, Speak now, and Send Now

    Voice mode

    A hands-free conversation. It sends when you pause, reads the reply, and listens again. Talk to Master starts it with the Master.

    Docs

  • The working line and the plan

    Above the composer: elapsed time, the tool call running now, and the step it's on. It opens to the agent's own todo list, live.

    Docs

  • Usage and context on a phone, from a chat's menu: the model, how full its context is, and the 5-hour and weekly limits with their reset times

    Usage

    Weekly and 5-hour limits with their reset times, in the web client on a wide screen and on the iPhone's lock screen. Each chat shows its model and how full its context is.

    Docs

  • The marks editor on a photo on iPhone: a box drawn on a trail sign with the note Is this the fork before the hut?, listed as mark 1

    Marks on photos

    Draw boxes and highlight strokes on an agent's image, each with a typed or spoken note. The agent gets numbered marks and a picture showing where each one is.

    Docs

  • Marks on video

    Mark a single frame or an In and Out segment, with a box on it if needed. A filmstrip to scrub, frame steps, and speeds from 0.25× to 2×.

    Docs

  • Narration and on-screen text

    When a video has text tracks, the cue list shows a frame at each cue. Change a cue to your exact wording, remove it, add one after it, or leave a note.

    Docs

  • A sound effect open in the marks editor on a phone: its waveform, the play control, and the In and Out buttons that mark a segment

    Marks on audio

    Mark a time or a segment on a waveform, so a note names the sample it is about.

    Docs

  • Pages in the chat

    An agent answers with an interactive page, such as a checklist, a comparison, or a form. It can send your choices back, and you can pin up to ten per chat. What you tick is saved for you, on each of your devices.

    Docs

  • An agent's request for a secret, as a card in its chat: the item and its vault, the agent's reason and the command it will run, and This session, Deny, Once, Project, and Always
    On a phone, an agent's request for a secret as a card in its chat, with its reason, its command, and the answers This session, Deny, Once, Project, and Always

    Secrets on a card

    An agent that needs a password or a token asks for it by name. The card shows the item, the agent's reason, and its command, and you choose This session, Once, Project, Always, or Deny. The value goes to that command, and [secret] replaces it in what the command prints.

    Docs

  • Settings, Secrets: a config change waiting for Confirm, the password manager and its state, and a switch for each vault agents see
    Settings, Secrets on a phone: a config change waiting for Confirm, the password manager, and the vaults agents see

    Grants and the audit

    Settings, Secrets has a switch for each vault agents see, the grants in force with Revoke, and every read, save, request, and decision. Agents see no vault until you name one. Vendetta ships a plugin for Proton Pass.

    Docs

  • A running agent's chat with Usage & Context beside it: the model it is on, and Switch model and Switch effort under it
    Usage & Context on a phone: a running agent's model, with Switch model and Switch effort under it

    Switch model and effort

    A running Claude or omp agent switches model or effort from Usage & Context in its chat. Claude waits for a running turn to end, omp takes it while idle, and later restores keep the new pair.

    Docs

  • The queue on a phone: three messages held for the end of the agent's turn, each with controls to reorder, edit, or remove it

    Queue

    Up to 50 messages held for after the agent's turn. Reorder, edit, or remove them.

    Docs

  • New agent from the phone

    Start one with a first message that can carry photos, files, video, or a voice note. Pick the project, provider, model, and effort, or make it ephemeral.

    Docs

  • A chat with an agent on iPhone: a chime alert saying the snapshot is restored, and a running 25-minute countdown to look at the upgrade check, with a cancel button

    Alerts and timers

    An agent on your computer can sound an alert or start a countdown on your devices. You hear it even with notifications muted. Agents in containers can't send alerts, timers, or Live Activities.

    Docs

Lock screen
A Live Activity counts the agents that need you and shows the question that has waited longest. An agent can run its own, with a title, a progress figure, and a countdown. Docs
Share sheet
From Photos or any app on iPhone, send files, a link, or text to the Master, a recent chat, or a new one. Docs
Telegram and Discord
A topic or channel per conversation, with progress edited into one message and questions as buttons. Voice notes are transcribed, and photos arrive as prompts. Docs
Notifications
iPhone push with an Answer action, web push, desktop notifications, and ntfy. Reading on one device clears the others. Docs

With other people, and other agents

Agents can lead agents, and the people you add get a Vendetta of their own on your computer.

  • The new chat composer on iPhone set to Manager: What should it get done?, the project, and chips for provider, model, and effort

    Managers and teams

    A manager is a Master for one project: select agents and start one over them. Teams nest, members start muted, and their manager relays what matters.

    Docs

  • A company's agents on a phone, as a tree: the chief of staff, the leads under it, and each lead's members with their states

    The team tree

    The overview draws each project as a frame, with a line from every agent to the ones it started.

    Docs

  • The Add Person sheet on iPhone with a display name typed and the name it becomes

    People

    A second person gets their own Vendetta on your computer, in containers, with their own chats and Master. You set limits on memory, CPUs, and machines for each.

    Docs

  • Share a project

    Share a whole project and you both prompt, answer, and start agents in it. Every message says who sent it, and notes between people never reach the agent.

    Docs

  • Ending a share on iPhone: the Share sheet for trail-maps asks Stop sharing trail-maps with Rui? Its chats go from their devices at once, with Remove and Cancel

    End a share

    Stopping and deleting stay with the person who shared. They can end the share at any time.

    Docs

  • Simple and Advanced

    Every device starts in Simple: chats by project, newest first, with three states. One switch at the top of Settings turns on Advanced.

    Docs

  • A container project's machine on a phone: its repository, memory, and the controls to stop it, fetch branches, or recreate it

    Containers

    A project's agents in a Docker or Podman container on a private clone. Your logins are served through a proxy and never copied in.

    Docs

The Master
One trusted agent that lists, reads, prompts, answers, starts, resumes, and renames the others, and manages schedules. It has no shell and never answers an approval. Docs

On a schedule

Prompts that run without you, each with a record of what happened.

  • A schedule on a phone: the smoke test failed this morning and stays paused until it is acknowledged, above its timing and the agent it prompts

    Schedules

    Cron, once, or on demand, to a new agent, the same agent, or one exact conversation. History keeps 1,000 runs, and a failed run stops until you look. In a container project a run starts a new agent in the container, or with --reuse prompts the conversation its first run started.

    Docs

  • An on-demand schedule on a phone: the support inbox's webhook address and the prompt each call runs

    Webhooks

    A URL that runs a schedule when a service calls it. Requests are signed, from GitHub, Forgejo, or any HMAC sender.

    Docs

  • Tasks grouped by project, with due dates, priorities, and a snoozed task
    Tasks on a phone, grouped by project, with due dates and priorities

    Tasks

    Tasks from any plugin that serves them, such as an Obsidian vault. A list and a week calendar, with snooze and priority, in the terminal, the apps, and the HUD.

    Docs

Project file
vendetta.toml declares a project's always-running agents, teams, and schedules. Nothing runs until you trust it. Docs
Housekeeping
Auto-delete for stopped conversations, with a dry run. An idle timeout pauses agents, and a message resumes them. Docs

Plugins

Most of Vendetta is plugins. Each is a separate program that the worker starts, keeps running, and stops with itself, and each reaches Vendetta only through the worker's private control socket. Yours can be written in any language.

A plugin can add buttons and pages to the iPhone and web apps as data the apps draw with their own pieces, serve a command your agents call, and react when an agent asks, finishes, or fails.

A plugin's page in the web client on a phone: the week's deploys, drawn by the app from what the plugin published
Ships with VendettaWhat it does
telegramA Telegram topic for each conversation
discordA Discord channel for each conversation, with status cards
appThe endpoint the iPhone and web apps connect to
forgesIssues and pull requests from GitHub, GitLab, and Forgejo for your projects' repositories
proton-passYour Proton Pass vaults, for the secrets agents and plugins use
hudCards over every window: agents, tasks, alerts, and your own sections. Hyprland, or macOS with the Vendetta app
hud-kittyA panel in the corner of your Kitty terminal on Hyprland
host-tmuxtmux as a terminal host: chooser, panes, launches, shortcuts, and tab labels
host-kittyKitty as a terminal host: chooser, panes, launches, shortcuts
shortcutsDesktop-wide keys on Hyprland and macOS
speechSpeech to text for voice notes and dictation, and the voice the apps read replies in
dictationPush-to-talk into the focused agent or window
ntfyAlerts on your phone through an ntfy topic
webhooksURLs that run a schedule when a service calls them
claude-proxyYour Claude and git logins for containers and people, through sockets of their own, never a copy
githubYour GitHub account: your repositories for new projects, and git for you and your containers
personal-hubTasks from your vault for the chooser, the apps, and the HUD
dashboardThe apps' Overview: lanes of agents by what they need
diffChanges: each agent's committed work, by revision, with your line comments
tailscaleYour app endpoint, and each person's, on your tailnet

The rest

Chat
Rich-text composer with shortcuts and a full-screen editor. Drafts that survive a relaunch. Reply with a quote. A photo editor with pen, shapes, arrow, text, crop, and rotate. @Photo 2 to name an attachment. Slash-command suggestions from the agent's own commands, skills, and MCP prompts. Links between chats that open at the exact reply. Interrupt, mute, keep, and ephemeral chats. A picture per chat. Search in a chat and across chats.
Media
Videos play in place and full screen. A viewer pages through every photo and video in a chat. Picture in Picture and a captions menu. Uploads up to 256 MB with progress and resume. Camera, drag and drop, and paste.
Lists
Tree or projects layout. Pinned projects and chats. History with resume. Mark all as read. Twelve project colours, the same on every surface.
Look
Light, dark, or system. An app colour, or match the project colours.
Web
A key for everything: jump mode, jump to a chat, step through chats, answer and approve on the overview. Every shortcut can be rebound, and ? lists them.
iPad
Split view, with the list beside the chat.
Teams
An agent can have several managers. A team has its own model and effort, and reminders for members not acted on.
People
Per-person switches for your Claude, git, services, and Master. Skills and MCP servers shared one by one. Their own GitHub. Managed from the app, with each step shown.
Terminals
tmux, or Kitty, as the terminal host. Desktop-wide shortcuts on Hyprland and macOS. A HUD panel in Kitty. Agents outside tmux are listed too.
Connections
GitHub sign-in, with your repositories as a source for new projects. Tailscale names for your endpoint and each person's.
Transcription
Yours to choose: a local command such as Whisper, or a provider.
For agents
A skill tells each agent what reaches you: images, video, audio, and subtitle tracks in a reply, questions as buttons, pages, alerts, timers, and Live Activities, follow-ups it schedules for itself, links to chats and changes, and vendetta agent to start and drive other agents.
Checks
vendetta config check reads your configuration and says what is wrong with it.