How To Use MaiaChat

Chat is conversation only. Agent is a coding/work harness for the current turn with tools. Loop keeps working until the goal is verified done and can spawn subtasks.

1. 1Use Chat for Q&A, drafting, and explanation with tools off.
2. 2Use Agent for vibe coding, file edits, shell, search, and one focused tool-using turn.
3. 3Use Loop for end-to-end goals that need observe → act → verify → repeat.

The Agents picker lets you attach specialists and choose Sequential, Parallel, Hierarchical, Consensus, Mixture, or Auto. This is manual team orchestration, not the same as Loop subagent spawning.

1. 1Open Agents in chat or the Agents page.
2. 2Add presets or your own templates to the conversation.
3. 3Choose an orchestration mode for how those specialists collaborate.
4. 4Keep Loop off when you want multi-agent orchestration; Loop uses one agent at a time.

In Agent mode you can ask MaiaChat to create reusable agent templates and attach them to the current chat, for example: create a coding agent on OpenRouter with Grok that has file access.

1. 1Switch to Agent mode.
2. 2Describe the agent: role, provider/model, tools, and whether it should be a reusable template.
3. 3Ask MaiaChat to create it and optionally add it to this conversation.
4. 4Use the Agents picker or ask Agent to run/select those specialists when needed.

Everyday coding: Agent + Project folder. Long build/verify goals: Agent + Loop. Debate/review/mixture: Agents picker with Loop off.

1. 1Vibe code: Agent + Choose project + Supervised or Full access.
2. 2Ship a feature end-to-end: Agent + Loop + coding preset.
3. 3Architecture review: add coder + reviewer + architect agents, then Mixture or Consensus.

Open Chat, choose a model, type your message, and send. MaiaChat stores the conversation and can generate a title automatically.

1. 1Use New Chat from the sidebar.
2. 2Choose a model if you do not want the default.
3. 3Enable tools, skills, memory, or retrieval only when needed.

Tools let the assistant search the web, calculate, work with files, call approved integrations, or run configured local capabilities.

1. 1Enable Tools in chat.
2. 2Select only the tools needed for the task.
3. 3Use approvals for sensitive shell or workflow actions when configured.

Documents can be searched through local RAG, Gemini stores, or both. The assistant receives retrieved context as a system message.

1. 1Upload documents in Documents.
2. 2Enable retrieval in Chat.
3. 3Select relevant documents or stores when the task needs source material.

Memory lets MaiaChat reuse relevant previous conversation context, user profile facts, and curated notes.

1. 1Enable Memory in Chat.
2. 2Ask naturally; the assistant will use relevant memories without forcing them into every answer.
3. 3Update or inspect memory from Settings when needed.

Create an agent with a name, role, provider/model, system prompt, tools, priority, and optional retrieval settings.

1. 1Open Agents.
2. 2Create or edit an agent.
3. 3Use the system prompt to define the agent's specialty and boundaries.
4. 4Set priority higher for agents that should influence coordination more.

In Agent mode, ask MaiaChat to create reusable templates or conversation agents. Example: create a coding agent using OpenRouter and Grok with file tools enabled.

1. 1Switch to Agent mode with tools on.
2. 2Describe provider, model, role, tools, and whether it should be reusable.
3. 3Ask MaiaChat to create the agent and optionally add it to the current chat.

Profiles collect multiple agents and set the default orchestration mode for that team.

1. 1Open Profiles.
2. 2Add the agents that belong together.
3. 3Choose single, sequential, parallel, hierarchical, consensus, or MOA based on the task.

Single uses one agent. Sequential passes work agent-to-agent. Parallel gets independent answers. Consensus synthesizes. Hierarchical coordinates. MOA fans out to references and synthesizes with an aggregator.

Normal Agent mode does not auto-spawn specialists. Loop mode can spawn subtasks. Multi-agent mode runs the agents you attached. Ask Agent to create/attach specialists first when you want a custom team.

Use MOA for complex planning, architecture, research, high-stakes writing, debugging strategy, or any task where independent perspectives improve the result.

1. 1Select at least two active agents.
2. 2Use diverse models or roles when possible.
3. 3Give a clear task prompt.
4. 4Let the aggregator produce the final response.

Reference agents run concurrently. Their drafts are hidden by default. The crowned aggregator receives the successful drafts, critiques them, resolves conflicts, and returns one answer.

The chat selector marks the actual MOA aggregator with a crown. MaiaChat chooses a selected coordinator first, reviewer second, and otherwise the highest-priority selected active agent.

MOA requires at least two successful reference drafts by default. If aggregation fails after enough references succeeded, MaiaChat falls back to the strongest successful reference answer.

MOA is available in multi-agent chat, profiles, channels, scheduled tasks, and the scheduled-task tool interface.

Run prompts on a schedule using a model, isolated agent, loop, script, or MOA agent set.

1. 1Open Scheduled Tasks.
2. 2Choose the execution mode.
3. 3For MOA, select at least two agents.
4. 4Choose a schedule and optional delivery channel.

Trigger workflows from incoming events or webhooks rather than a time schedule.

Use background execution for work that should continue outside the active chat screen.

Inspect autonomous work, blockers, artifacts, approvals, and recent task state from one operational view.

The app includes channel infrastructure for webchat and messaging platforms such as Slack, Discord, Telegram, WhatsApp, Teams, Signal, SMS, and email depending on configured connectors.

Channels can use sequential, parallel, consensus, or MOA mode with selected agents.

1. 1Open Channels.
2. 2Choose a channel.
3. 3Enable multi-agent behavior.
4. 4Select the mode and the agent set.

Channel permission profiles control which tools are exposed, whether filesystem access is available, and how unknown contacts are handled.

Add keys for providers such as OpenAI, Anthropic, Google, xAI, Perplexity, OpenRouter, or local model providers where available.

Enable reusable skills and connected services from Settings. The chat can use only the capabilities that are enabled and allowed.

Usage tracking and cost settings help monitor token spend and provider routing behavior.

Sensitive actions can require approval before execution, especially command, workflow, and channel-triggered operations.