✦ Caytral Ops Guide
Your AI Ops Team
Is Ready for Instructions.
Learn exactly where the buttons are, which agent to pick for what job, and how to get real work done from the dashboard — start to finish.
7 agents
$350/mo
Mon–Fri 8 AM EST auto-run
Quick Start
Every agent lives in a row on the Agents page. The controls are on the right side of each row. Here's exactly what you're looking at:
This is one agent row — it looks like this on the dashboard:
Orchestrator
↑ The colored dot shows status. Buttons on the right are your controls. That's everything you need to operate an agent.
The core loop: Pick an agent → Assign Task → Run now → check the Runs tab to see what it did. That's it. Everything else in this guide is just variations on that loop.
Your First Task — Step by Step
Follow this once end-to-end and you'll understand the whole system. We'll send a research task to the Research agent.
Walkthrough
~2 minutes · no prior experience needed
1
Go to ops.cayesdesk.com → click "Agents" in the nav
You'll see a list of all 7 agents. Each row has a colored dot on the left and buttons on the right.
2
Find the Research agent row — click
+ (Assign Task)A dialog box opens. You need two things: a title (short label) and a description (the actual instructions).
Title: "Competitor research"
Description: "Find the top 5 AI phone answering services for medical clinics. List their pricing, key features, and any weaknesses compared to a small custom solution."
Description: "Find the top 5 AI phone answering services for medical clinics. List their pricing, key features, and any weaknesses compared to a small custom solution."
3
Click
Create — the task is now saved and queuedThe task won't run yet. It's just sitting in the queue waiting for the agent to pick it up.
4
Click
▶ Run now on the Research agent rowThis kicks the agent off immediately. The dot will turn blue and start pulsing — that means it's running.
5
Click the agent name to open its detail page → go to the
Runs tabYou'll see the active run listed. Click it to watch the live transcript — you can literally read what the AI is doing and thinking in real time.
6
Wait for it to finish — the dot turns green when done
The completed run will show the full output: research report, sources, token usage, everything.
Be specific in the description. The more context you give — who you are, what you already know, what format you want back — the better the output. Vague tasks get vague results.
Your 7 Agents
Each agent has a specialty and a model powering it. Pick the right one for the job — sending a writing task to the Engineer or a coding task to Comms wastes time.
Orchestrator
CEO
Opus 4.6 — most capable model
The manager agent. Breaks down complex projects, delegates subtasks to the right agents, and coordinates multi-step work across the team.
Best for: Complex multi-step projects, cross-agent work
Example ask: "Plan a 3-month outreach campaign for CayesDesk — research competitors, draft email sequences, and set up a content calendar."
Research
Researcher
Sonnet 4.6
Digs into topics, gathers data, and writes structured research reports. Pulls from the web and synthesizes findings.
Best for: Market research, competitive analysis, technical investigation
Example ask: "Find the top 10 medical clinics in Broward County that still use manual phone scheduling. Include their contact info and estimated patient volume."
Engineer
Engineer
Sonnet 4.6
Writes, reviews, and debugs code. Implements features, fixes bugs, and handles technical infrastructure tasks.
Best for: Code changes, bug fixes, feature implementation
Example ask: "Add a webhook endpoint to the CayesDesk backend that receives Twilio call logs and saves them to the database."
Comms
General
Azure OpenAI
Writes all external-facing communication: emails, client messages, follow-ups, proposals, and customer responses.
Best for: Email drafts, customer responses, external comms
Example ask: "Draft a cold outreach email to a dental clinic front desk explaining what CayesDesk does and why they should book a demo."
Analyst
CFO
Opus 4.6 — most capable model
Reviews numbers, analyzes business data, and produces clear reports. Good at financial modeling, metric summaries, and data-driven decisions.
Best for: Financial analysis, metrics review, data-driven decisions
Example ask: "Based on 7 agents at $350/mo, project the minimum number of CayesDesk clients needed to break even at $299/mo per client."
Content
CMO
Sonnet 4.6
Creates marketing copy, blog posts, landing page content, and social media posts. Focused on persuasion and brand voice.
Best for: Blog posts, marketing copy, landing pages
Example ask: "Write a landing page headline and 3 short benefit bullets for CayesDesk — targeting small medical clinic owners who are overwhelmed with phone calls."
QA
QA
Opus 4.6 — most capable model
Reviews work produced by other agents — catches errors, checks logic, verifies quality before anything goes out the door.
Best for: Code review, testing, quality checks
Example ask: "Review this email draft from the Comms agent. Check for tone, clarity, any factual claims that seem off, and suggest improvements."
Opus vs Sonnet: Opus 4.6 is the smartest but slower and costs more tokens. Sonnet 4.6 is fast and efficient. The platform already assigned the right model to each agent — don't stress about this.
Status Indicators
The colored dot on each agent row tells you what's happening — and what you should do about it.
Status
What it means
Your move
Active / Idle
Healthy and ready. No tasks running right now.
Nothing needed. Assign a task anytime.
Running
Actively working on a task right now.
Just watch. Open the Runs tab to see live output.
Paused
Paused manually. Won't run on schedule or pick up tasks.
Click Resume when you're ready to re-activate it.
Error
Something went wrong on the last run.
Read the logs first, fix the cause, then click Clear error.
Queued
Task is assigned but agent hasn't started yet.
Click Run now to skip the wait, or let it run at 8 AM.
Common Workflows
Step-by-step playbooks for the situations you'll hit most often.
Send a task to an agent
1
Open
ops.cayesdesk.com → Agents page
If you're already on the dashboard, look for "Agents" in the left nav
2
Identify which agent fits your task — use the agent cards above if unsure
3
Click
+ (Assign Task) on that agent's row4
Fill in a short title and a detailed description of what you want
The description is your actual instructions — be specific, include context, say what format you want back
5
Click
Create — the task saves to the queue6
Click
▶ Run now to process immediately, or wait for the 8 AM EST auto-runMonitor an active run
1
Look for agents with a blue pulsing dot — those are running right now
2
Click the agent name to open its detail page
3
Go to the
Runs tab4
Click the active run to see the live transcript and output
You'll see tool calls, reasoning, intermediate steps, and final output — all in real time
Recover from an error
1
Filter the Agents list by the
Error tab — or look for red dots2
Click the agent →
Runs tab → click the failed run3
Read the error message carefully
Common causes: expired API key, model rate limit hit, network timeout, or a bad task description that confused the agent
4
Fix the underlying issue (rotate the key, wait for rate limit to reset, rewrite the task)
5
Click
Clear error — agent resets to idle and is ready againSet up a recurring routine
1
Go to the
Routines page in the nav2
Click
New Routine3
Set the title, description (what to do each time), which agent to assign it to, and the schedule
Example: assign the Analyst to run a weekly revenue summary every Monday at 8 AM
4
Save — it runs automatically from now on, no manual trigger needed
Automated Schedule
Your agents work while you sleep — here's what happens automatically.
8:00 AM
Monday – Friday · EST
Every weekday morning, each agent wakes up, checks its task queue, and processes everything queued up since the last run. You don't have to do anything — just assign tasks and they'll get handled.
Assign tasks the night before to have results waiting for you in the morning
Use Run now any time you don't want to wait
Weekends and holidays: agents don't run — use Run now if you need something done
Troubleshooting
Something not working? Click the question that matches your situation.
You must be on the Agents page — check that the URL contains
/agents. The buttons are on the right side of each row. On smaller screens or narrow windows, only the icons show (no labels). Try widening your browser window.Open the agent → Runs tab → click the failed run to read the full error. Do not click Clear error until you've read the logs — clearing it without fixing the cause will just cause it to error again. Common root causes: API key expired or missing, model rate limit hit (wait 1–5 minutes), or a task description the agent couldn't parse. Fix the cause, then click Clear error.
Check three things in order: (1) Is the agent paused? Look for a yellow badge — if paused, click Resume. (2) Is the budget exhausted? Go to the Budget tab on the agent — if it's at zero, the agent won't run. (3) Does it have any tasks queued? Agents only run when there's something to do. You can always click Run now to force an immediate check.
Tasks queue until the next run — they don't auto-start. Click Run now on that agent's row to process it immediately. Or wait for the 8 AM EST weekday heartbeat. To confirm the task saved, go to the Tasks page and check it's listed there.
Click the agent name to open its detail page → go to the Runs tab → click any completed run. You'll see the full conversation transcript, every tool call the agent made, the final output, and the token count for that run. This is also how you verify quality and catch mistakes.
This almost always comes down to task description quality. Be more specific: include what you already know, the format you want back (bullet list, report, email, etc.), who the audience is, and any constraints. Also make sure you're using the right agent for the job — sending a writing task to Engineer will get mediocre results compared to Comms or Content.