The product
Your AI workforce builds the company brain you exit on.
28 named specialists run on your stack. They write what they learn into a versioned brain mirrored to your Notion. Documented agencies sell at 8–12× EBITDA — that’s the asset. Approval-gated by default, earned-autonomy as trust accumulates.
Pillar A · The Brain
A second brain for your business — written by your AI workforce, owned by you forever.
Every AI employee writes findings to a shared, versioned brain. Every entry is signed by the agent that wrote it, with a timestamp and a commit message. Mirrored to your Notion.
Every AI employee writes findings to a shared, versioned brain — Pattern Cards (what’s normal for client X), Decision Log (what we decided and why), Risk Register (what could hurt us next quarter), Real P&L, KPI Snapshots, Client Profiles. Every entry is signed by the agent that wrote it, with a timestamp and a commit message. The whole brain is mirrored to your Notion workspace.
Why the brain is the asset (not the workforce).
Documented agencies sell at 8–12× EBITDA. Undocumented agencies sell at 5–9×. For a $2M-EBITDA agency, that’s $6M of extra exit value. The agents commoditize — OpenAI Agents SDK lets any competent dev rebuild “28 agents” in a sprint. What they CAN’T rebuild is two years of your accumulated, signed-author, git-versioned brain. That’s the moat. That’s the artifact you sell when you exit.
Where the brain lives
At OpsRadar
3-table versioned Postgres (documents, immutable versions, commits). Curated paths get pgvector embeddings so the COO can semantic-search across them. Everything else stays markdown-only.
In your Notion
Every brain write is mirrored to a corresponding Notion page in your workspace. You can read it on Notion mobile at 3 AM with no OpsRadar login.
At /dashboard/brain/download
The full versioned brain, exportable as a markdown zip, any time, on any tier.
Pillar B · The Workforce
28 named specialists. Hire more in plain English.
Every customer starts with 28 named built-in AI employees the moment they sign in — 11 domain agents plus 17 toolkit specialists. Hire custom ones in plain English.
Every OpsRadar customer starts with 28 named built-in AI employees the moment they sign in — all shipped with senior-COO operator judgment baked into their prompts. Hire custom ones in plain English: “I want a Bookkeeper named Sarah.”
The 11 named domain agents
Riley
Revenue Analyst
Cole
Collections Lead
Scott
Scope Watchdog
Devon
Delivery Manager
Brian
Daily Briefer
Penny
Profitability Analyst
Tia
Team Lead
Grace
Growth Strategist
Renee
Renewals Lead
Sam
Stack Auditor
Evan
Evaluator
Plus 17 toolkit specialists (Stripe Specialist, HubSpot Specialist, Harvest Specialist, ClickUp Specialist, & more).
How a custom hire works
- 01
Open /dashboard/hire and describe the role. Plain English.
- 02
We compile the spec — name, role, schedule, scoped tools, system prompt, first task. You read it as a card.
- 03
Click Hire. Sarah starts running tomorrow at 8 AM.
- 04
Sarah has her own chat page at /dashboard/employees/sarah. Talk to her like a teammate. She remembers.
Built on the senior-COO playbook.
The 28 built-in agents don’t just call APIs. Their prompts contain operator judgment distilled from agency-operator publications — signals + thresholds, decision trees, verbatim conversation scripts. Cole (Collections Lead) doesn’t just say “this invoice is overdue.” He says: at $X days late, send Y; at $Z days late, queue a margin-alert action with this draft email. Penny (Profitability Analyst) flags clients trending toward unprofitable in week 2, not at quarterly review.
What the brain knows about your business after 30 days
- Every Stripe transaction Riley has looked at, with the margin context she derived
- Every overdue invoice and Cole's notes on the customer payment pattern
- Every scope risk Scott flagged, with the Harvest hours that triggered it
- Every decision YOU made in the COO chat — recorded as a Decision Log entry, signed by you
- Every rejection you gave an AI employee — recorded as a Pattern Card and applied permanently (see: Apprentice Mode)
Ask about your business
One COO. Seventeen specialist agents underneath. The one answer that matters.
When you ask the COO chat “which clients should I worry about?”, it doesn’t make up an answer. It reads what Riley (Revenue), Penny (Profitability), and Cole (Collections) have written to the brain in the last 30 days. Synthesizes across them. Cites the entries by author + date. Then answers — with a clickable trail back to the source brain entries.
Tool specialists (17)
Watch one app each — Stripe Specialist, HubSpot Specialist, Harvest Specialist, ClickUp Specialist, and so on. They go deep on their toolkit.
Domain agents (10)
Read what the tool agents wrote — Revenue, Collections, Scope, Delivery, Briefing — and synthesize across multiple tools. Riley reads what Stripe + HubSpot + QuickBooks specialists wrote and produces the actual revenue health summary.
COO chat (1)
Reads the domain agents and writes the daily briefing. You read one thing, not seventeen.
Why OpsRadar is not a chatbot
Three things ChatGPT, Lindy, and 11x literally cannot do.
Differentiator 01
Apprentice Mode
Your AI workforce gets smarter every time you say "no."
Every time you reject an action, OpsRadar derives a generalizable rule from your reason and writes it to that employee's permanent learned-rules brain prefix. The next time Sarah is about to do the thing you didn't want, she reads the rule and doesn't.
Concrete example
Day 1: Sarah drafts a follow-up email to an international Stripe customer. You reject it: "We don't auto-follow-up international customers — they're on a different cadence."
Day 2 onwards: Sarah never drafts a follow-up to an international Stripe customer again. The rule is in her brain, signed by you, dated, reversible if you ever change your mind.
Differentiator 02
Approval-history personalization
After 30 approvals it drafts like you. After 100, your team can't tell the difference.
Every approve / reject / edit you make on an action feeds into every agent's next prompt as a "your past decisions" block. The system literally reads your decision history before drafting anything for you. The voice converges on yours.
Differentiator 03
Cross-agent synthesis
One COO that reads 17 specialist agents and tells you the one thing that matters today.
You don't get 10 separate AI reports to read. You get one daily briefing, synthesized across every agent's findings, ranked by what's actually at stake. The other 9 reports stay one click away if you want to drill in.
What’s under the hood
Three more durable moats. The receipts.
Receipt 01
Versioned commit-graph brain (git-style)
Every brain entry has an immutable version row. Every commit has a message and an author (the agent that wrote it). You can diff what changed between two versions. You can roll back any agent edit. It's the only AI-knowledge-base architecture in the category that gives you git-style operations on your company's memory.
Receipt 02
Agent-as-author attribution
Every fact in your brain is signed by the agent that wrote it, the timestamp, and the agent_run_id that produced it. When your COO chat tells you something, you can click through to see which agent learned it, when, and from what data.
Receipt 03
Earned autonomy (risk-tiered + confidence-graduated)
Three autonomy tiers, set per tool. High-risk actions (firing a client, sending invoices, mass email) stay Ask forever. Medium-risk actions (drafting replies, scheduling meetings) auto-fire with notification once they've earned 30+ approvals at >90% rate. Low-risk actions (Notion writes, internal logs) fire silently. The agents earn their autonomy. They don't get it on day one. You can always override the tier per tool: Ask, Auto, Off.
When X happens, do Y
Automations the way you’d describe them out loud.
“When a HubSpot deal moves to closed-won, post in #ops Slack and create a Notion kickoff page” — that’s an automation. You describe it to the COO chat in plain English. We compile it to a typed spec, show it to you as a card, you click Install. It runs forever.
- Built on Nango Triggers — 20+ webhook events across HubSpot, Stripe, ClickUp, Gmail, Calendly, and more
- Executed durably on Trigger.dev — retries 5 times with exponential backoff if something fails
- Every action still routes through the approval gate by default — automation doesn't mean autonomy
Your AI workforce has hierarchy
An org chart your AI workforce actually follows.
Custom AI employees can have managers and direct reports. When Marcus the Account Manager needs a Stripe lookup, he delegates to Sarah the Bookkeeper — pauses his run, waits for her to finish, reads her summary, and continues. Real hierarchy, real delegation, real durable wait.
- Built on Trigger.dev's tasks.triggerAndWait pattern — the manager pauses, the report runs, the manager resumes with the result
- Cycle-detection prevents infinite loops (managers can't report up the chain to themselves)
- Each employee appears in the YOUR TEAM sidebar with their own chat page
Stack
Built on what 2026 actually ships.
Frontend
Next.js 16 · React 19 · Tailwind v4 · shadcn/Radix · Framer Motion
Backend
Vercel Fluid Compute · Vercel AI SDK v6 · Vercel AI Gateway
Database
Supabase Postgres · pgvector · row-level security · multi-tenant
Durable workflows
Trigger.dev v3 (per-employee execution, delegation, automation runner)
Scheduled cron
Vercel Cron (28 built-in agents)
AI models
OpenAI GPT-5.5 / 5.4 / 4o / 4.1-nano · text-embedding-3-small · Anthropic Claude (scaffolded)
Tool brokerage
Nango (25 live integrations, growing to 500+)
Web search
Tavily
Knowledge layer
Anthropic Agent Skills (46 SKILL.md files across 9 domains)
Observability
Sentry · Statsig · Langfuse
Rate limiting + cache
Upstash Redis
Email
Resend
MCP server
OpsRadar's own brain is queryable as MCP by external agents
One workspace. Replaces 10 AI tools.
Starts at $99.
OpsRadar is in private beta. Join the waitlist for when the front door opens — or talk to the founder if you’d rather see it run on your stack first.
Private beta · Limited spots · Brain stays in your Notion