Skip to content
M.A.V.E.Ndocs

Quickstart

This assumes M.A.V.E.N is installed and open at http://localhost:3000.

1. Your first chat

Section titled “1. Your first chat”

Open Chat from the sidebar (or go to /chat). The engine row across the top lists every model engine this install can reach - Claude, Grok, Codex, Kimi, Z.AI, Qwen, DeepSeek, Qwen Code, Antigravity, a local Ollama model, OmniRoute, PROVIDERS. An engine that is not set up is greyed out and says what it wants (a CLI on PATH or a key in Settings → Secrets).

Pick one that is lit, type a prompt, press Enter. The same window streams the reply whether it comes from a CLI agent, a cloud API or a local model. History is kept per engine under ~/.maven/chat/; Resume brings a past thread back.

With no key and no CLI at all, the fastest way to a working engine is Local: install Ollama, ollama pull gemma2, and the Local engine lights up.

2. Set the global model

Section titled “2. Set the global model”

The global model is the one the apps that take “a model” without asking use - Loop, Pipeline, SEO-Forge’s generator, and every picker left on global. It is the box at the bottom of the sidebar. Open it and choose any family this install can actually run; the list is exactly the set the Model Manager has switched on.

A saved choice is an instruction: M.A.V.E.N honours it as-is and does not silently swap in another model when a probe is slow. If the pick stops working, the optional failover (Settings → Model routing) answers from the next model of the same kind and tells you who answered.

3. Hand work to an agent

Section titled “3. Hand work to an agent”

Agents differ from chat models: they have tools and a working directory. In Chat, pick Claude, Codex, Grok or Qwen Code, open Settings on the right, set a Working dir, and give it a task that touches files. Tool calls show inline; the Workspace tab browses what it wrote. Hermes, OpenClaw and OpenCode have their own consoles under Agents in the sidebar.

4. Your first Mastermind

Section titled “4. Your first Mastermind”

When one answer is not enough, open AI Mastermind (/room). The header chips show every reachable agent; trim the roster to two or three. Pick a mode:

  • ASK - each agent answers blind; you compare.
  • DISCUSS - a round table over 1-4 passes; each speaker reads the ones before.
  • MoA - every agent proposes, layers refine, one aggregator returns one answer. The header says what the run will cost in calls before it starts.

Type the question, send, and watch the seats answer in turn. Details on AI Mastermind.

Where the output goes

Section titled “Where the output goes”

Everything you make is written to your Obsidian vault (~/Obsidian/maven-os unless MAVEN_VAULT_DIR says otherwise): chats, goals and journal entries into one daily note under Memories/, and each app into its own folder. Generated media shows up in the Media Library. Nothing is uploaded anywhere.

Next

Section titled “Next”