For operators
Start with one process. Add the next when it holds.
Human-in-the-loop for AI agents means a named person, not the agent, commits any change your rules mark as consequential. With elan1, the agent drafts the change, the rules decide whether it needs approval, and the approver sees exactly what will change before saying yes. Routine changes within limits can run alone, and every action is recorded.
01The first question, answered
Can an agent do real work here without anyone losing control of it?
Yes, when the work is a governed action. The agent compiles a typed write intent. elan1 Govern checks it against your rules, with no model in the check. Where your rules say so, a named approver who is not the requester sees it first. Then one writer makes the change. If anything is unclear, it stops and asks a person.
02What you see in the product
Open any card to see it at full size, with notes on what each part does. Functional colours appear only inside the product, always with a word.
Proposed changes and the record each would change.
The diff, gates passed or failed, rule version, and “Approve and commit” for the named approver only.
A typed code in plain words, what to fix, who to ask.
One proof record per change.
Illustrative specimen. Not customer data. *Kolam Components is a fictional company; all data is illustrative.
03How it starts
Month-end journals, vendor invoices, credit notes: work where a sign-off already exists.
Seven blocks, versioned. Who may act, who approves, what is never allowed.
Every refusal carries a typed code and what to fix. A refusal costs 40% of one governed action.
Above the autonomy ceiling, the approver commits. Every change leaves one record.
Paid per governed action. A refusal costs 40% of one. No per-seat fees.
List prices on the pricing page. All prices are exclusive of GST.
04Platforms and regulations that apply
Enterprise Intelligence Platform
Governed agents for the work that commits your business.
“What should change in the books?”
Meter: credit
Explore WorkDecision Intelligence Platform
Typed, calibrated decisions, routed by risk.
“What is the right answer?”
Meter: decision
Explore DecideGoverned Intelligence Platform
The governed write path for AI agents.
“May this record change?”
Meter: governed action
Explore GovernThe parts a first process touches
elan1 Work ERPAgents draft and apply changes in ERPNext, hosted by elan1 or attached to yours. Move from Tally with a migration.Status: Availableelan1 Decide RouterRoutes each answer by risk tier: auto, review, System-2 or hold.Status: Availableelan1 Decide Review ConsoleUnsure cases go to a person, who sees the case and the score side by side.Status: Availableelan1 Govern · L1 GuardedThe Right Path, approvals, a single writer and a tamper-evident record you can read.Status: AvailableRegulations your risk team will ask about
EU AI ActHuman oversight and logs at the moment of action.India DPDP Act 2023Residency, records and our role under the Act.05What we don't do for you
06FAQ
It means a named person, not the agent, commits any change your rules mark as consequential. With elan1, the agent drafts the change, the rules decide whether it needs approval, and the approver sees what will change before saying yes. Every outcome, approved or refused, leaves a proof record.
When approvers are asked to rubber-stamp requests they cannot read, or when the requester can approve their own change. elan1 shows the approver the difference, the gates that ran and the rule version, and the approver is never the requester. Changes within your limits run without approval, so approvals are kept for the changes that matter.
Those your declaration allows below the autonomy ceiling. Each process has an autonomy level from A0 to A5: below the ceiling, a routine change that passes the gates and rules commits without a person; above it, a named approver must commit. Every change, with or without approval, is recorded.
The difference the change would make, the gates and rules that held it, and the rule version in force. Only the named approver sees “Approve and commit”, and no answer in time is a refusal. The decision is written to the proof record.
No answer in time is a refusal. The write does not happen, the refusal is recorded, and the meter shows 0.4 of a governed action. Nothing is committed by default.
No. The approver is a named person who is not the requester, and only the approver sees “Approve and commit”. An agent never approves its own change.
No. You write the rules you already follow as a declaration: configuration, not a build. Start with one process where a person already signs off.
40% of one governed action. A refusal still runs every check and writes a proof record, but nothing is written to your books, so it costs less than an allowed change. Within your allowance, a refusal uses 0.4 of an included governed action. No per-seat fees. All prices are exclusive of GST.
An undo is itself a governed change. It goes through the same checks, the same approver rules, and leaves its own proof record.
// START WITH ONE PROCESS
Tell us the process, the record it changes and who signs off today. A person replies within 2 working days.