Rent the capability. Own the judgement.

Models will keep changing. Vendors will keep changing. Your rules, your record and your budget should not. LeafCraft is the layer that keeps them on your side.

NeutralYoursOpenEqual

Book a demo Experience now

Your cloud accountAWS, Azure or Google Cloud

Your VPCInside your private network

On-premDocker or Kubernetes in your data centre

Short outagesLocal models, up to 7 days offline

Keys in your vault, or one of 14 you useRecords in your Redis or Postgres

Not buy versus build

Both, under one roof.

Buy the vendor's agent: it ships this quarter and it works. Build your own where the process is your edge. Most large companies will do both at once. Neither one arrives with governance attached.

The agent you bought

Ask it five questions. Watch it go quiet.

Who owns this decision? Human or agent, name it.
Walk me through every step. Then replay it.
Can I swap the model without the answers changing?
What does training it on my data cost? And break?
If the model gets costly or retired, how do you keep me running?

The agent may work very well. But the decision trail, the first thing your auditor will ask for, sits in the vendor's stack.

The agent you built

The agent was the easy part.

A team can demo a working agent in one sprint. Getting it ready for production means building a control layer around it:

Guardrails: what it may approve alone, what needs a checker, what it must never touch
Test sets, re-run on every model change and prompt change
People breaking it on purpose before it breaks in production
An escalation matrix with named people and clear thresholds
An audit trail your auditor will accept

That is 12 to 18 months before the first agent is safe in production, then a permanent team to keep it running. And you rebuild it for every agent.

What is missing is not another agent. It is a layer of your own underneath all of them. That layer is LeafCraft.

See the platform

Never let the counterparty hold the evidence

One vendor should never be the seller, the meter and the referee.

Your auditor does not report to the CFO they audit. Your custodian is not your fund manager. The layer that decides when to switch a model, when to stop an agent and what you carry into your next renewal should not belong to anyone who sells you the agent.

The sellerSells you the agent and the model behind it.
The meterCounts what you owe, by its own rules.
The refereeHolds the rules and the record of what happened.

The sellerSells the agent

The SaaS vendor, the cloud or the model provider. They gain when you use more.

The meterCounts what you owe

If the seller also runs the meter, you find out about cost from the invoice.

The refereeHolds the rules and the record

This must stay with you. LeafCraft sells no agents, no models and no compute, so it has nothing to protect.

What to look for

Neutral. Yours. Open. Equal.

Ask for these four in the contract, whichever layer you choose.

Neutral

LeafCraft sells no agents, no models and no compute. It gains nothing from which model, cloud or agent you pick.

Yours

It runs in your environment, so it keeps working even if a vendor does not. Any agent you build on LeafCraft or bring into it belongs to you, not to LeafCraft.

Open

Works with any model and any cloud, on open standards like MCP and A2A. Rules and records export in open formats on demand, without raising a project.

Equal

A bought agent, a built agent and a person follow the same rules and leave the same record, on whichever cloud they run.

The four things you never rent

They are not technology. They are your risk appetite.

Guardrails, cost limits, escalation paths and the record are your regulator's expectations and the promises made to your teams. Keep them with you.

GuardrailsWhat each agent may do alone, escalate or never touch

Cost ceilingPer task, decision and agent, set before it runs

Escalation pathNamed people, clear thresholds

The recordEvery task and decision, with both reasons

Claim C-20418 · the recordExample data

Claims agent matched the claim to scheme S-118Why: within the scheme, quantities match the dispatch

10:41

Claims agent passed it to Priya NairWhy: above my set limit, so not mine to decide

10:41

Priya Nair approvedWhy: scheme confirmed with the regional team

14:12

Audit agent wrote it to the audit logWhy: any later edit to this entry can be detected

14:12

In your database · unchanged · exportable

How the options compare

Where should your rules and record live?

A fair comparison of the usual choices, from the buyer's side.

What you need	Governance inside one agent vendor	A cloud provider's control plane	Build it in-house	LeafCraft
Covers agents from other vendors	Its own agents	Strongest on its own stack	Whatever you build for	40+ platforms plus your own
People in the same roles as agents	People approve	Admins manage agents	If you design it	Any role, AI or person
Where the record lives	Vendor's stack	Provider's tools	Yours	Your database, exportable
Sells agents, models or compute	Yes	Yes	No	No
Cost known before work runs	Usually after, on the invoice	Varies	If you build it	Per task, agent and type
Time to first governed process	Fast, for that vendor's agents	Depends on your stack	12 to 18 months	One process at a time, from a ready-made team

What you get instead

One layer. Your rules on every run.

Every hand-off between an agent, a person and a system goes through one layer that applies your rules, asks the right person, checks the cost and writes it down.

How it works, step by step

One run · example data

01Work arrives
WebhookKafkaSlack
Claim C-20418North region · 10:41:01
API, webhook, schedule, message or queue
02Facts first
- Scheme S-118
- Invoice INV-7731
- Dispatch quantities
Your code pulls what the task needs
03A role acts
Claims agent"Approve. Matches the scheme and the invoice."
Agent or person, within its limits
04Rules decide
Above the agent's limit
Next agentA person signs offStop
Next agent, sign-off, escalation or stop
05Recorded
Who
Priya Nair
Why
Scheme confirmed with the regional team.
How long
3 h 31 m
Cost
1.84 of 4.00
Who, what, why, how long, what it cost

One run · example data

Discover every agent

You can't govern what you can't see.

LeafCraft connects to your ecosystem and finds every agent you run: the ones vendors shipped inside your software and the ones your teams built. Then each one gets an owner, comes under your rules, or is removed.

Shadow AI endsEvery agent ownedUnwanted agents removedMemory starts from day one

How discovery works

Inside software you buySwitched on by vendor updates

CRMERPHR systemService deskOffice suite

Built by your teamsPilots, scripts and side projects

LangGraphCrewAINotebooksAutomationsCopilot studios

LeafCraft connects to your ecosystem and finds every agent

One registerexample counts

23found

9no owner yet

11governed

3retired

When a model is retired

Switch the model. Keep the judgement.

An agent built straight on one model is rebuilt and re-approved when that model is retired or repriced. On LeafCraft the model is a setting per agent. Switch it, test it against your own record, and keep the owner, rules, limits and memory.

Agent tied to one model

The model is retired or repriced
Prompts, tests and sign-offs rebuilt
Months of work, a new approval cycle
The history stays with the old tool

MonthsA rebuild budget

Agent on LeafCraft

Pick another model for that agent
Switching is a config change
Test it against your own record first
Rules, limits, owner and memory stay

HoursNo rebuild

Before you renew any AI contract

Ask four questions.

Save these for your next renewal. They are the questions your auditor and your board will ask you.

Can you replay the reasoning, step by step?
Which model and version, and can I swap it without changing my output?
Who absorbs it if token prices rise?
Can I export the full record: tasks, decisions, overrides, approvals and escalations?

Not how much AI. Who decides, and who checks.

Bring one process and your current agents. We'll show you how they come under your rules.

Book a demo Experience now