AI Agent

Raven Docs includes a powerful AI agent that can assist with planning, task management, research, and documentation. The agent learns from your workspace and provides contextual assistance.

Overview

The agent operates as an intelligent assistant with access to your workspace:

User ←→ Agent Chat ←→ Agent Core
                         ├── Memory System
                         ├── Planning Engine
                         ├── Task Manager
                         └── Research Tools

Agent Chat

Starting a Conversation

Access the agent chat:

• Click the chat icon in the header
• Use Cmd/Ctrl + Shift + A
• Type in the chat panel

Capabilities

The agent can:

• Answer questions about your documentation
• Create and manage tasks
• Generate content drafts
• Research topics from docs and web
• Plan your day or week
• Summarize long documents

Example Conversations

You: What tasks are due this week?
Agent: You have 5 tasks due this week:
       1. Review API docs (Tomorrow)
       2. Update deployment guide (Wednesday)
       ...

You: Create a task to update the README
Agent: Created task "Update README" in your inbox.
       Would you like to add a due date or link it to a goal?

You: Research best practices for API versioning
Agent: I'll research API versioning. This may take a few minutes.
       [Generates research report]

Agent Planning

Planning Horizons

The agent can create plans across multiple horizons:

Horizon	Timeframe	Focus
Daily	Today	Immediate priorities
Short	This week	Weekly objectives
Mid	This month/quarter	Project milestones
Long	This year+	Strategic goals

Generating Plans

Ask the agent to plan:

You: Plan my day
Agent: Based on your tasks and goals, here's a suggested plan:

       Morning:
       - [ ] Review PR #123 (30 min)
       - [ ] Team standup (15 min)

       Afternoon:
       - [ ] Complete API documentation (2 hrs)
       - [ ] Respond to feedback (30 min)

       Would you like me to create these as scheduled tasks?

Plan Cascading

Plans cascade from long-term to daily:

Long-term: Launch product
    ↓
Mid-term: Complete beta features
    ↓
Short-term: Finish authentication
    ↓
Daily: Implement OAuth flow

Agent Memory

The agent maintains memory of your workspace:

What's Remembered

• Conversations - Past discussions and decisions
• Preferences - Your working patterns
• Context - Project details and documentation
• Activity - Pages viewed, tasks completed

Memory Features

• Contextual recall - Agent remembers relevant past discussions
• Learning - Improves suggestions over time
• Graph connections - Links between entities and concepts

See Memory System for detailed configuration.

Autonomous Mode

Scheduled Runs

Configure the agent to run autonomously:

Schedule	When	Purpose
Daily	Each morning	Plan the day, surface priorities
Weekly	Monday morning	Weekly review, plan the week
Monthly	First of month	Monthly reflection, goal review

Autonomous Actions

When running autonomously, the agent can:

• Generate daily plans
• Surface overdue tasks
• Create weekly review pages
• Suggest goal updates

Approvals

Some actions require approval:

Agent: I'd like to create 3 tasks based on your weekly goals.
       [View proposed tasks]

       [Approve] [Modify] [Reject]

Agent Settings

Configuration Options

Setting	Description	Default
Autonomy Level	How independently agent acts	Assisted
Schedule	When agent runs autonomously	Daily
Timezone	For scheduling	System
Approval Required	Actions needing approval	Create, Delete

Per-Space Settings

Override settings for specific spaces:

Space Settings {
  agentEnabled: true
  autonomyLevel: "full"
  scheduleOverride: "weekly"
}

Agent Suggestions

Proactive Suggestions

The agent proactively suggests:

• Task priorities - "Consider focusing on X today"
• Missing links - "This task might relate to goal Y"
• Content updates - "This doc hasn't been updated in 30 days"
• Review reminders - "Time for your weekly review"

Reflection Questions

During planning, the agent asks:

• "What's the most important thing to accomplish today?"
• "Are there any blockers I should know about?"
• "How did yesterday's plan work out?"

External Agent Integration

Raven Docs supports external AI agents connecting via the MCP API. Workspace administrators can control which agents have access and what they can do.

Agent Registration

External agents can register with your workspace through two methods:

1. Invite-based registration (Recommended) - Create an invite token with specific permissions
2. Public registration - Allow any agent to request access (requires workspace setting)

Creating Agent Invites

Workspace admins can create invite tokens for external agents:

Settings → Agents → Invites → Create Invite

Configure the invite:

Setting	Description
Name	Descriptive name (e.g., "GitHub Copilot", "Internal Bot")
Permissions	What actions the agent can perform
Max Uses	How many agents can use this invite (null = unlimited)
Expiration	When the invite expires

Agent Permissions

Control what agents can do with granular permissions:

• page.read - Read page content
• page.write - Create and update pages
• task.read - View tasks
• task.write - Create and update tasks
• memory.read - Access workspace memory
• research.run - Execute research jobs

Managing Registered Agents

View and manage all registered agents:

Settings → Agents → Agents List

For each agent, you can:

• View activity logs
• Revoke access
• Update permissions

Resource-Level Access Control

Control agent access at the resource level for sensitive content.

Page Agent Access

Each page has an "Agent Accessible" toggle:

• Enabled (default) - Agents can read and interact with this page
• Disabled - Page is hidden from all agents

Toggle via:

• Page menu → Agent accessible switch
• Or via the API when creating/updating pages

Use cases:

• Personal journals or notes
• Sensitive internal documents
• Draft content not ready for agent consumption

Task Agent Access

Tasks also support agent accessibility controls, either explicitly set or inherited from the project.

Security & Privacy

Data Access

The agent only accesses:

• Content you have permission to view
• Your own tasks and activity
• Shared workspace knowledge
• Resources marked as agent-accessible

Privacy Controls

• Disable agent for specific spaces
• Clear agent memory
• Audit agent actions
• Control data retention
• Toggle agent access on individual pages/tasks
• Revoke external agent access at any time

Agent Runtime Hosting

Raven Docs supports multiple hosting modes for agent runtimes, allowing flexibility in where and how agents execute.

Hosting Modes

Mode	Description	Best For
Local	Runtime runs on your machine or local VM	Quick setup, interactive auth
Parallax Cloud	Managed Kubernetes infrastructure	Enterprise workflows, SLA guarantees
Custom	Your own VPC or self-hosted cluster	Compliance requirements

Configuring Hosting

1. Navigate to Settings → Agents
2. Under Agent Runtime Hosting, select your hosting mode
3. Configure the runtime endpoint (for Local/Custom modes)
4. Set authentication type and credentials

Runtime Endpoint

For Local and Custom modes, specify the runtime endpoint URL:

http://localhost:8765  # Local development
https://runtime.mycompany.com  # Custom deployment

Spawning Agents

Request new agents from the configured runtime:

1. Click "Spawn Agents" in the agent management panel
2. Select agent type (Claude Code, Codex, Gemini CLI, Aider)
3. Configure count and optional settings
4. Monitor spawn progress in the activity feed

Supported Agent Types

Type	Description
Claude Code	Anthropic's Claude-powered coding assistant
Codex	OpenAI's code generation model
Gemini CLI	Google's Gemini-powered CLI agent
Aider	AI pair programming assistant
Custom	Your own agent implementation

Login Flow (Local Runtime)

When spawning agents that require authentication:

1. Runtime starts agent CLI in PTY session
2. If login required, Raven Docs displays login instructions
3. Complete device code flow or OAuth
4. Agent registers and connects to workspace

Runtime Status

Monitor your runtime connection:

• Connected - Runtime is healthy and responding
• Disconnected - No heartbeat received
• Active Agents - Number of agents currently running
• Version - Runtime version for compatibility checks

Best Practices

1. Be specific - Clear questions get better answers
2. Provide context - Mention relevant projects or goals
3. Review suggestions - Agent learns from your feedback
4. Use approvals - Keep human oversight for important actions
5. Monitor runtime - Keep an eye on runtime status and agent health

Related

• Memory System - Detailed memory configuration
• Research - Agent research capabilities
• GTD System - Productivity integration
• MCP Tools - Agent API tools