HalosHalosDocs

Workflows

This page contains step-by-step tutorials for common use cases. Each workflow shows you exactly how to set up your agents, training, and integrations to handle a specific part of your daily life.

These workflows will grow over time as the platform evolves. Think of them as recipes — follow them step by step, then customize to fit your needs.

Workflow 1: Email Triage & Management

Goal: Never manually sort through your inbox again. Your email agent reads incoming mail, prioritizes it, drafts replies, and keeps everything organized.

Time to set up: 10-15 minutes

Setup

  1. Deploy an Email Agent

    • Go to Agents > Deploy New Agent
    • Select the Email specialty
    • Name it something memorable (e.g., "Luna")

  2. Connect Gmail

    • Go to Settings > Connections
    • Click Connect next to Google
    • Authorize Gmail access

  3. Add Training Data

    Add these entries to your email agent:

    Knowledge — "Key Contacts"

    List your most important contacts with context: "Sarah Chen (CEO) — always respond same day. James K (client) — formal tone. Dev team — casual is fine."

    Instruction — "Triage Rules"

    Categorize emails as: URGENT (needs response within 2 hours — anything from my manager, clients, or with 'urgent' in subject), IMPORTANT (respond today — team requests, project updates), FYI (no response needed — newsletters, notifications, CC'd threads). Always show urgent items first.

    Instruction — "My Email Style"

    Keep replies concise — 2-3 sentences max for routine emails. Use first names. No "Dear" or "To whom it may concern." Sign off with "Best, [my name]" for external, just my name for internal.

  4. Enable Heartbeat

    • On Luna's profile, set heartbeat to Every hour
    • Luna will now check your inbox hourly and alert you to important emails

Daily Usage

Morning check-in:

@Luna what landed in my inbox overnight?

Draft a reply:

@Luna draft a reply to Sarah's email about the timeline — 
tell her we're on track for the March deadline

Batch triage:

@Luna go through my unread emails and categorize them. 
Archive anything that's purely FYI.

Leveling Up

Once your email agent reaches Level 3, it unlocks the Draft Replies permission. At Level 5, it can Apply Labels. As trust builds, promote it from Observer to Helper so it can start organizing your inbox autonomously.

Workflow 2: Daily Briefing

Goal: Start every day with a comprehensive briefing that covers your schedule, inbox, and anything else that matters — without checking five different apps.

Time to set up: 15 minutes (assumes you have 2+ agents deployed)

Setup

  1. Deploy at least two agents — An Email Agent and a Calendar Agent (and optionally a Research Agent)

  2. Connect integrations — Gmail and Google Calendar

  3. Set heartbeats

    • Email agent: Every hour (catches urgent items throughout the day)
    • Calendar agent: Once a day (morning schedule brief)
    • Research agent (optional): Once a day (industry news)

  4. Train your Calendar Agent

    Instruction — "Daily Brief Format"

    When giving a daily briefing: start with today's date and a one-line summary ("Busy day — 5 meetings" or "Light day — 2 meetings, mostly free"). Then list each event chronologically with time, title, and any prep notes. Flag conflicts or back-to-back meetings. End with "open blocks" where I have free time.

Daily Usage

Morning (in The Yard):

@everyone good morning — give me the rundown

Your calendar agent lists today's schedule. Your email agent summarizes overnight emails. Your research agent shares any relevant news. All in parallel.

Quick schedule check:

Am I free at 3pm tomorrow?

The calendar agent picks this up automatically (no mention needed).

Meeting prep:

@Atlas I have a meeting with Acme Corp at 2pm — 
what should I know going in?

Your research agent pulls context from your knowledge base about Acme Corp and any recent interactions.

Workflow 3: Research & Learning

Goal: Stay informed on topics that matter to you without spending hours reading. Your research agent tracks trends, summarizes articles, and compiles briefs on demand.

Time to set up: 10 minutes

Setup

  1. Deploy a Research Agent

    • Select the Research specialty
    • Name it something like "Atlas" or "Scout"

  2. Add Training Data

    Knowledge — "My Industry"

    I work in [your industry]. Key topics I track: [list 3-5 topics]. Key companies to watch: [list competitors or companies you follow]. Preferred sources: [publications you trust].

    Instruction — "Research Format"

    For quick research: bullet points, 3-5 key findings, sources linked. For deep dives: executive summary (2-3 sentences), detailed findings with headers, source list at the bottom. Always note when information is older than 6 months.

    Instruction — "Daily News"

    For daily briefings, focus on: [your industry] trends, [specific technology] developments, and anything from [key publications]. Skip generic tech news unless it's major.

  3. Set heartbeat to Once a Day for a daily industry digest

Daily Usage

Quick lookup:

What's the latest on [topic]?

Deep research:

@Atlas I need a deep dive on [topic] — focus on 
[specific angle]. Include data points and sources.

Competitor monitoring:

@Atlas any news from [competitor company] this week?

Meeting prep research:

I'm meeting with [company] tomorrow. What should I know about 
them? Recent news, funding, product launches, anything notable.

Learning:

@Atlas explain [complex topic] like I'm familiar with 
[related topic] but new to this specific area.

Workflow 4: Code Review & Development

Goal: Get instant code reviews, debugging help, and development support without context-switching between your editor and a chat tool.

Time to set up: 10 minutes

Setup

  1. Deploy a Code Agent

    • Select the Code specialty
    • Name it (e.g., "Nova" or "Byte")

  2. Connect GitHub (optional but recommended)

  3. Add Training Data

    Knowledge — "Tech Stack"

    Our stack: Next.js 14, TypeScript, Tailwind CSS, Supabase (Postgres + Auth), Vercel for hosting. Testing: Vitest + React Testing Library. Linting: ESLint + Prettier. Package manager: yarn.

    Knowledge — "Code Conventions"

    Naming: camelCase for variables/functions, PascalCase for components/types, SCREAMING_SNAKE for constants. File structure: feature-based (not type-based). Prefer named exports. Use 'use client' directive only when needed. Error handling: always use try/catch for async operations, log errors with context.

    Instruction — "Review Priorities"

    When reviewing code, prioritize in this order: 1) Security issues, 2) Logic bugs, 3) Performance concerns, 4) Type safety, 5) Code readability, 6) Style/formatting. Always mention what's done well, not just problems.

Daily Usage

Code review:

@Nova review this function:
[paste code]

Debugging:

@Nova I'm getting this error: [error message]
Here's the relevant code: [paste code]
What's wrong?

Architecture questions:

@Nova I need to add real-time notifications to our app. 
We use Supabase. What's the best approach?

PR summaries (with GitHub connected):

@Nova summarize the open PRs on our main repo

Workflow 5: Personal Productivity System

Goal: Use a custom agent as a personal productivity coach and task manager that knows your goals, projects, and priorities.

Time to set up: 15 minutes

Setup

  1. Deploy a Custom Agent

    • Select the Custom specialty
    • Name it something like "Chief" or "Compass"

  2. Add Training Data

    Knowledge — "Current Projects"

    Active projects: 1) Website redesign (deadline: April 15, lead designer: Maria), 2) API migration (deadline: March 30, blocker: legacy auth system), 3) Hiring — looking for 2 senior engineers (posted on LinkedIn and Lever).

    Knowledge — "Goals"

    Q1 goals: Ship redesign, complete API migration, hire 2 engineers. Personal goals: Read 2 books/month, exercise 4x/week, learn Rust basics.

    Instruction — "Productivity Coach"

    Act as my productivity partner. When I share a task, help me break it down into actionable steps. When I share a problem, help me think through it before jumping to solutions. For daily check-ins, ask about progress on current priorities and suggest focus areas. Be direct — don't pad responses with filler.

    Instruction — "Weekly Review Format"

    For weekly reviews: 1) What I accomplished this week (from our conversations), 2) What's still in progress, 3) What's blocked, 4) Suggested focus areas for next week. Keep it to one screen.

  3. Set heartbeat to Once a Day

Daily Usage

Morning planning:

@Chief what should I focus on today? I have about 4 hours 
of deep work time.

Task breakdown:

@Chief I need to write the technical spec for the API 
migration. Help me outline it.

End of day:

@Chief here's what I got done today: [list]. 
Update my project tracker.

Weekly review:

@Chief let's do the weekly review

Workflow 6: Multi-Agent Team Coordination

Goal: Use 3+ agents together as a coordinated team, where each agent handles its domain and they work in parallel.

Time to set up: 20-30 minutes (deploying multiple agents)

The Ideal Starting Team

| Agent | Specialty | Role | |-------|-----------|------| | Luna | Email | Inbox management, communication | | Sol | Calendar | Schedule, meetings, time management | | Atlas | Research | Information gathering, trend tracking | | Nova | Code | Development support (if you code) |

Team Coordination in Practice

The power shows up when agents work together:

Cross-domain requests:

I have a meeting with the Acme Corp team on Thursday. 
Can someone prep a briefing and check if there are any 
outstanding emails from them?

Atlas (research) pulls together an Acme Corp briefing. Luna (email) checks for recent Acme Corp correspondence. Both respond in parallel.

Morning standup:

@everyone morning standup — what do I need to know?

Each agent reports on its domain: Luna on emails, Sol on schedule, Atlas on news, Nova on code/PRs. You get a comprehensive picture in 30 seconds.

Complex planning:

I need to plan a product launch for April. I need to: 
schedule team meetings, draft announcement emails, research 
competitor launches, and review the release code.

The router might send this to multiple agents, or you can follow up with each one individually for their piece.

Building Your Own Workflows

These recipes are starting points. The best workflows are the ones you build for your specific needs. Here's how to think about it:

1. Identify the Friction

What do you spend time on that's repetitive, tedious, or involves a lot of context-switching? That's your first workflow candidate.

2. Pick the Right Agent Type

Match the task to a specialty. If it doesn't fit neatly, use Custom and train heavily.

3. Train Specifically

Generic agents give generic results. The more specific your training data, the more useful the agent becomes. Include real examples from your actual work.

4. Start with Observer

Let the agent prove itself at read-only before giving it write access. Watch how it handles tasks and correct it via training before promoting trust.

5. Iterate

Your first version won't be perfect. Chat with the agent, see where it falls short, add training data to address the gaps, and repeat. Most agents hit their stride after 3-5 rounds of refinement.

Next Steps