Personas

A persona is a reusable identity for your AI agent — combining personality, capabilities, and constraints into a single package.

Think of it as a personality + toolbox. Instead of configuring the same system prompt, model, and tool permissions every time you start a conversation, you bundle them into a persona and reuse it everywhere — in chat, in bots, and in automated workflows.

What's Inside a Persona?

Every persona is a configuration object with these building blocks:

Property	What it controls
Name, description, avatar, color	Identity — how the persona appears in the UI
System prompt	The core instructions that shape the agent's personality and behaviour
Preferred models	Which AI models to use (primary + fallback list, supports glob patterns like claude-*)
Secondary models	Lighter models used for background tasks like context-map generation and compaction
Allowed tools	Scoped tool access — restrict exactly what the agent can do. Use * for full access or list specific tools
MCP servers	Which external integrations (databases, APIs, services) this persona can connect to
Loop strategy	How the agent reasons — react (think → act → observe), sequential, or plan_then_execute
Context map strategy	How workspace context is gathered — general, code, or advanced (LLM-powered semantic analysis)
Prompt templates	Reusable Handlebars prompt snippets with optional input schemas, invokable from chat or workflows

Why this matters

Personas turn one-off configuration into reusable, shareable agent profiles. A "Code Reviewer" persona always reviews for security issues. A "Technical Writer" persona always outputs clean Markdown. You configure once and use everywhere — no drift, no forgotten instructions.

Built-in vs Custom Personas

HiveMind OS uses a namespace convention to separate system and user personas:

• system/ — Built-in personas that ship with the app (e.g. system/general). These are bundled into the binary, cannot be deleted, but can be customised or archived.
• user/ — Personas you create (e.g. user/code-reviewer, user/team/ops/monitor). Full control — edit, archive, or delete at any time.

The default persona, system/general, is a general-purpose agent with access to all tools (*) and the ReAct loop strategy. It's the blank canvas you start with.

How Personas Connect to Everything

Personas are the common thread across the three main ways you interact with HiveMind OS:

• Regular chat — Select a persona from the sidebar before (or during) a conversation. The agent adopts that persona's prompt, tools, and model preferences for the entire session.
• Bots — Every bot wraps a persona with additional triggers and schedules. The persona defines what the bot can do; the bot defines when it does it.
• Workflows — The invoke_agent and invoke_prompt steps accept a persona ID, so automated pipelines can call different specialist agents at each stage.

Skills

Skills are portable knowledge packs that add domain expertise, procedures, and reference material to a persona. Skills are managed per-persona through the UI (not as a field in the persona configuration). From the persona editor:

• Click Manage Skills to browse, install, enable, or disable skills for that persona
• Skills are scoped — a "Kubernetes" skill installed on your DevOps persona won't clutter your Technical Writer persona
• Skills inherit data classification — a skill marked CONFIDENTIAL elevates the persona's effective classification level

Creating and Managing Personas

Open Settings → Personas to manage your collection:

1. Create from scratch — Click New Persona, fill in the fields, and save. Your persona appears under the user/ namespace.
2. Start from a template — Use an existing persona as a starting point and customise from there.
3. Archive / Restore — Don't need a persona right now? Archive it to hide it from listings. It stays resolvable so existing bots and workflows that reference it keep working. Restore it any time.
4. Edit built-ins — Customise any system/ persona. You can always reset it back to factory defaults later.

Example: A Security-Focused Code Reviewer

Say you want an agent that only reviews code and always checks for security issues. Here's what that persona looks like:

id: user/code-reviewer
name: Code Reviewer
description: Security-focused code review specialist
systemPrompt: |
  You are a meticulous code reviewer focused on security.
  Always check for: SQL injection, XSS, auth bypasses, secrets in code.
  Be constructive but thorough.
preferredModels:
  primary: claude-sonnet
allowedTools:
  - filesystem.read
  - filesystem.search
  - web.search
loopStrategy: plan_then_execute

Notice what's not in the allowed tools list — filesystem.write, shell.execute. This persona can read and search, but it can never modify your codebase. That's the power of scoped tool access: you get a specialist agent that is capable but contained.

You could then:

• Start a chat with this persona to review a PR interactively
• Wire it into a bot that triggers on new pull requests
• Call it from a workflow step after your CI build passes

Learn More

• Agentic Loops — Deep dive into ReAct, Sequential, and Plan-then-Execute (plan_then_execute) strategies
• Bots — How bots wrap personas with triggers and schedules
• Workflows — Automating multi-step pipelines that invoke personas
• Tools & MCP — How tool access and MCP servers work