Docs

Install Agent V, complete a first useful run, learn each work surface, and recover when something fails.

Start here

Build and install from source Pack Agent V with electron-builder and verify the runtime on Windows, macOS, or Linux.
Quickstart: first useful change Open a workspace, configure a provider and model, run a small task, then review the result.

Choose Ask, Plan, or Agent Match the mode to the job and understand exactly which actions each mode permits.
Prompt and attach context Use precise prompts, @ mentions, files, images, audio, paste, and drop without exceeding composer limits.
Workspaces, chats, tabs, and split panes Understand the four containers Agent V can open and what closing each one changes.
Background and interrupted runs Read run status, return to work that continued off-screen, and resume only the interrupted chat you open.
Context and compaction Read the context meter, understand its estimates, and compact a long chat without treating summaries as lossless.
Checkpoints and undo Review agent writes, keep or discard a checkpoint, and understand what checkpoints do not recover.
Agent instances Delegate independent work to depth-one child runs, inspect their progress, and merge isolated work deliberately.
Questions, todos, and plans Respond to blocking forms, follow live task state, and move an approved run plan into Agent mode.
Long-lived goals Keep a chat working an objective until complete, pause or resume it, and arm a prompt loop.

Providers and API keys Configure one of eleven provider hosts, store credentials securely, and make a configured provider active.
Models, thinking, and speed Refresh model catalogs, choose favorites, and use capability-dependent thinking and service-tier controls.
Marketplace overview Use Browse for registry packages and Manage for MCPs, Skills, Rules, and installed Packages.
MCP servers Add local or remote MCP servers, configure bearer or OAuth authentication, inspect catalogs, and control workspace use.
Skills Create personal or project SKILL.md packages, understand discovery precedence, and load them in a run.
Rules Author user-global and workspace instructions, understand precedence, and control automatic injection.
Packages and workspace overrides Install, update, enable, remove, and scope Marketplace packages without confusing global and workspace state.
Slash commands Use all 15 built-in commands and understand dynamic commands from skills, MCP tools, rules, and command files.

Files editor Browse and edit workspace files, handle external changes, and use diff, blame, LSP, and formatting integrations.
Browser Share the embedded browser with the agent, control navigation, manage session data, and enforce a domain allowlist.
Terminal Distinguish interactive terminal sessions from agent commands, choose a shell, and recover from PTY fallback.
Changes and Git Inspect working-tree status and diffs, separate checkpoints from Git history, and commit through the correct boundary.
Pull requests and GitHub Connect GitHub, satisfy repository prerequisites, and create or inspect a pull request from the dock.
Memory files Store deliberate workspace knowledge in bounded Markdown files without confusing memory with semantic search.
Codebase search and indexing Choose exact or semantic search, configure the local index, and understand model downloads and derived storage.
Voice dictation Choose OpenAI, OpenRouter, or Local transcription and manage the two supported Whisper caches.
Notifications Configure the inbox and desktop toasts for completed, failed, blocked, and system events.

What Agent V is Understand the Agent V coding workspace, supported workflows, and run-model boundaries.
Runs, sessions, and state Learn how chats, invocations, artifacts, receipts, instances, interruption, and cleanup relate on disk.
Security and approval model Understand mode boundaries, tool approval, path guards, browser restrictions, and residual risk.
Privacy and data storage Map local data, provider-bound content, credentials, logs, indexes, memory, and deletion boundaries.

Settings reference Exact controls, options, defaults, and scope across all nine Settings sections.
Built-in tools reference All 61 built-in tool names, grouped by job with mode, approval, output, and limit boundaries.
Keyboard shortcuts Keyboard chords as listed in Settings → Shortcuts. They are not rebindable.
Attachment limits and formats Authoritative count, size, format, extraction, and model-capability limits for composer attachments.
Storage locations Locate settings, workspaces, runs, indexes, logs, notifications, models, packages, skills, rules, and memory.

Provider and model issues Recover from missing keys, inactive providers, empty catalogs, refresh errors, and unsupported model capabilities.
Run, network, and recovery issues Diagnose offline queues, interrupted turns, repeated Continue states, provider cutoffs, and crash recovery.
Browser and terminal issues Resolve blocked hosts, navigation failures, shared-control stalls, missing shells, and PTY fallback.
Marketplace and MCP issues Recover from registry acknowledgement, package install, MCP import, authentication, connection, and timeout failures.
Indexing and dictation issues Recover from model downloads, hash fallback, reindex failures, Ollama embedding errors, and voice setup problems.
Git and pull-request issues Resolve missing repositories, commits, remotes, Git identity, GitHub CLI, and authentication.