Every decision remembered. On your side.
Every task and decision by every agent and every person, with the reasons of both, kept unchanged in your own database. It is your enterprise memory, and it stays when models and vendors change.
The problem
A log is not a record.
When an auditor, a customer or your own board asks how a decision was made, the answer is usually spread across five systems and someone's memory. The contract covers uptime, not judgement.
What you get
Can it show me how, and why?
Every step is traced
Every agent, model, tool and hand-off is recorded from the first run. Nothing to set up and no tracking code to add.
Both reasons, side by side
Every approval and rejection is logged with who made it, when and the reason they gave, right next to the agent's reason.
What it cost, call by call
Tokens and cost for every model call, by agent, model and run.
Records you hold
Run records sit in your own Redis or Postgres, kept for as long as you set.
Export on demand
Rules and records export in open formats, any time, without raising a project.
Written once
Template flows end with an audit agent that writes to a log that can't be changed after it's written.
Why it compounds
Without the record you cannot leave. With it, change is easy.
Every task, every decision, both sides' reasoning, every handoff and every time authority widened, and why. Held in one place on your side, it is the only part of your AI stack that compounds. Your agents improve on your history, not on everyone else's average.
How it plays out
An auditor asks how a loan was approved.
- You search your runs by date and open that one.
- You see what the document check agent was given, which model and tools it used and what it handed back.
- You see the credit officer's approval, when they gave it and the reason they wrote.
- You see what the whole run cost, call by call, and walk the auditor through it step by step.
Checked 7 documents against the loan policy.
Passed to the credit officer.
Approved.
Checked the decision against the rules. Passed.
Privacy by design
A record of the work. Not surveillance of people.
Customer text can stay out of traces
One setting masks message content in traces. Timings, costs and agent names stay visible.
Personal details aren't kept as lessons
When an agent saves a lesson, emails, phone numbers, dates and reference numbers are masked first.
Erase on request
One call deletes everything stored for a session, and the deletion itself is recorded with who asked for it.
Questions teams ask
Where is the record stored?
In your own Redis or Postgres, in your cloud, your VPC or on-prem, for as long as you set.
Can we take it with us if we change vendors?
Yes. Rules and records export in open formats on demand. That is the point: your memory survives every change of model or vendor.
Do we need to add tracking code to our agents?
No. Every agent, model, tool and hand-off is recorded from the first run, including agents built on other platforms.
Is this monitoring our employees?
No. It records the work and the reasons for decisions, so anyone can see how an outcome was reached. It is not surveillance of people.
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.
Explore the rest of the platform
How it works
From work arriving to the record, step by step
ExploreDiscover every agent
Find every agent you run, then own, govern or retire it
ExploreOne team
AI agents and people in the same roles
ExploreYour rules
Roles, access, checks, authority, escalation
ExploreYour budget
Cost known and capped before work runs
ExploreBring any agent
40+ agent platforms, any model, your stack
ExploreSecurity and deployment
Your cloud, VPC or on-prem
ExploreBring the decision you'd struggle to explain today.
We'll show you the full trail behind it, with both reasons, in your own database.