What Agent V is

Agent V is a coding workspace for working through real repositories. It combines a natural-language harness, workspace tools, provider-hosted models, live context management, and explicit file-backed memory.

The application and workspaces run on this machine. Model requests go to the active provider host. New settings select local Ollama, but the local endpoint must still be installed and reachable before it can serve a request.

Product surfaces

Three top-level views—chat, Settings, and Marketplace—organize the product. Chat includes workspace and chat navigation, a persisted transcript, the composer, and six dock panels: Files, Browser, Terminal, Changes, Pull Request, and Plan.

Ask, Plan, and Agent are enforced interaction modes. The built-in tool catalog contains 61 tools, including Skill. Connected MCP servers can add separate Agent-only tools.

Local and remote boundaries

Workspace files, run records, indexes, logs, settings, notifications, and local model caches are stored locally. Content leaves the machine when a configured model provider, remote MCP server, remote Ollama host, GitHub, or website receives it.

API keys use operating-system secure storage. That protects storage at rest; it does not change the fact that a provider receives prompts and attached content sent to it.

Explicit non-goals

• Agent V is not a hosted cloud IDE; the repository remains a local folder.
• Memory is not automatic retrieval-augmented generation. It is Markdown under .vyotiq/memory/.
• Semantic code search is a derived local index, not memory.
• Marketplace is not one undifferentiated bucket. Manage separates MCPs, Skills, Rules, and Packages.
• Composer @ Docs means project documentation, not this product manual.

Run model

A chat stores an evolving run history. Each send starts a new invocation in that run. /clear creates a fresh chat. Child agent instances have their own run and transcript; they are not another desktop window or a UI split.