Flows
The workflow, written down: the goal, the order of steps, which agents are involved, and where a human must approve. Publish a new version any time — work already in flight is never disturbed.
Platform
One engine. Every boring workflow.
You describe what needs to happen, in plain language. Mundane works out how — with a team of AI agents that do the work through your systems, pause for your people, and keep a record of every step.
The building blocks
Three things. That's the whole setup.
No code, no deployments, no six-month integration project. Every workflow on Mundane is made of the same three parts — all configured, all versioned.
Agents
Specialist AI workers, each with a role, instructions and a set of tools. One manager agent runs the room — deciding who works next, when to ask a question, and when the job is done.
Tools
What agents are allowed to touch: your internal systems, third-party services, documents, knowledge bases. Credentials stay encrypted, and every outbound call is screened before it leaves.
How a run works
From “please handle this” to handled.
Every piece of work — a complaint, a claim, an onboarding packet — becomes a run. Here's what happens to it.
STEP 01
Work arrives
From your app, an API call, a schedule or a webhook. The engine loads the flow and a run begins — visible from second one.
STEP 02
The manager delegates
The manager agent reads the goal, breaks it into tasks and hands each one to the right specialist. It tracks what's done and decides what happens next.
STEP 03
Specialists deliver
Each specialist works autonomously — reading documents, calling your systems, checking its own results — then hands back structured output, streamed live.
STEP 04
People stay in charge
Any sensitive action — a refund, a payout, an account change — can require sign-off. The run pauses itself and waits. Nothing executes until a person decides.
STEP 05
Nothing gets lost
Every step is saved as it happens. Servers restart, deployments ship, approvals take a day — the run picks up exactly where it stopped. Hours or weeks, it holds.
STEP 06
Everything is on record
The result arrives as structured data — and behind it, a ledger: every decision, every action, every approval, and what each step cost. Provable, not anecdotal.
Built for production
The unglamorous parts, done properly.
Anyone can demo an agent. Running one on real work — sensitive systems, real money, real customers — takes the parts nobody shows on stage.
Durable by construction
Runs live in the database, not in memory. If anything dies mid-run, the work resumes from its last step — automatically.
Approvals built in
Flag any action as needing sign-off, and the engine enforces it — with the request, the decision and the decider on record.
Every penny visible
Each AI call is metered — what was asked, how long it took, what it cost. Spend is a report, not a surprise at month-end.
Change without downtime
Adjust instructions, add a tool, reorder steps — publish and new runs pick it up instantly. Everything versioned, nothing silent.
Fits into your product
Mundane is headless — it works behind your interface, over live connections or plain APIs. Your customers see your brand, not ours.
Grounded in your knowledge
Bring your documents, policies and data. Agents search them directly — answers come from your sources, and cite them.
Your environment
Runs where you run.
Deploy in your own cloud or on-prem. Swap AI providers without rewriting anything. Keep your data where your regulator expects it.
Talk to us about deployment01Your VPC, your on-prem, or air-gapped — your call.
02Works with the major AI providers — switch models without changing your workflows.
03Credentials encrypted and vaulted; agents never see raw keys.
04Every outbound connection screened against your allow-list before it leaves.
05We never train models on your data. Ever.
Next step
See it run your workflow.
Bring one process you're tired of. We'll walk you through how it looks as a flow — agents, approvals, audit trail and all.