Skip to main content

How it fits

The AI proposes. Code decides. A person approves.

elan1 Run carries it. elan1 Ontology relates it. elan1 Decide judges it. elan1 Govern decides it. elan1 Act, Propel and Resolve do it. elan1 Pulse analyses it. Every action, answerable.

A solutions architect maps layers on a glass wall as a colleague watches

What does an agentic AI architecture look like?

In elan1, three layers sit above your systems of record. Industries set the platform up for a sector's rules, systems and approvers. Applications do the work: elan1 Act, Propel and Resolve, with elan1 Pulse analysing it. The foundation relates, judges, decides and carries it: elan1 Ontology, Decide, Govern and Run. ERP, CRM and models stay underneath, and stay yours.

The platform

Applications do the work.

The foundation relates it, judges it, decides it and carries it. Your systems of record stay underneath, and stay yours.

Underneath

Your systems of record, underneath, not ours

ERP · CRM · helpdesk · commerce platforms · core banking · Tally · models from Anthropic, OpenAI, Google and open source

Only Govern allows.

The layers

What each layer does.

  1. 01

    Industries

  2. 02

    Applications

    elan1 Act, Propel and Resolve do it. elan1 Pulse analyses it.

  3. 03

    Foundation

    elan1 Run carries it. elan1 Ontology relates it. elan1 Decide judges it. elan1 Govern decides it.

  4. 04

    Your systems of record, underneath, not ours

    ERP · CRM · helpdesk · commerce platforms · core banking · Tally · models from Anthropic, OpenAI, Google and open source

How it works

Ask. Check. Approve. Act. Record.

Follow one supplier bill at Kolam Components, an illustrative distributor in Coimbatore, from a WhatsApp message to the books.

Every action an agent proposes runs on elan1 Run, reads elan1 Ontology, is judged by elan1 Decide where a decision is routine, and is allowed, sent to a person or refused by elan1 Govern.

  1. 01

    • WhatsApp · supplier bill

      supplier 0.99 · order match 0.97 · tax 0.41, not sure

    A supplier bill arrives on WhatsApp; elan1 Decide reads it: supplier 0.99, order match 0.97, tax 0.41 and not sure. Illustrative data.

    Ask.

    elan1 Decide reads the bill. elan1 Act drafts it on a copy.
  2. 02elan1 Govern

    • 6 checks pass · 1 fails

      12% GST billed, 18% due · ₹1,18,720 → ₹1,25,080

      Hold
    elan1 Govern runs the rules: 6 checks pass and 1 fails, 12% GST billed where 18% is due, so the bill moves from ₹1,18,720 to ₹1,25,080 and is held. Illustrative data.

    Check.

    elan1 Govern runs your rules. The GST rate is wrong.
  3. 03

    • Finance head

      Approved by Asha R.
    The finance head, Asha R., approves. Illustrative data.

    Approve.

    The finance head decides.
  4. 04

    • ERP bill posted

      ₹1,25,080

    Only what she approved is posted: the ERP bill at ₹1,25,080. Illustrative data.

    Act.

    Only what she approved is posted.
  5. 05

    • Receipt

      agent · rule · approver · outcome · time · hash

    The receipt card records the agent, the rule, the approver, the outcome, the time and the hash. Illustrative data.

    Record.

    A signed receipt: what happened, who said yes, when.

Illustrative. Kolam Components is a fictional company.

The rules

Only Govern allows.

The AI proposes. Code decides. A person approves. Every action, answerable.

  • It decides before the action.

    Allow, ask a person or refuse, before an agent's tool call runs. Anything not clearly allowed is refused; an error refuses.

  • A receipt anyone can verify.

    Every decision, refusals included, leaves a signed, hash-chained receipt that an auditor checks offline with one command.

  • One path.

    Every product's actions take the same path and leave the same receipt. No product builds its own gate.

Runs with

Your systems of record, underneath, not ours.

Agents work in Frappe, Zoho, Odoo, SAP, Microsoft Dynamics, Salesforce or Twenty.

  • Runs inside the systems you have

    • Frappe
    • Zoho
    • Odoo
    • SAP
    • Microsoft Dynamics
    • Salesforce
    • Twenty
    • Tally
  • Models

    • Anthropic
    • OpenAI
    • Google
    • Open source

    A model may propose an action. It may never approve one.

  • Runs where you say

    • The cloud
    • Your data centre
    • An appliance: Edge, Department, Rack

    AWS Mumbai by default, with no US region.

Questions

How it fits, answered plainly.

In elan1, three layers sit above your systems of record. Industries set the platform up for a sector's rules, systems and approvers. Applications do the work: elan1 Act, Propel and Resolve, with elan1 Pulse analysing it. The foundation relates, judges, decides and carries it: elan1 Ontology, Decide, Govern and Run. ERP, CRM and models stay underneath, and stay yours.

In the foundation, between every agent and your systems. elan1 Govern decides each action before an agent's tool call runs: allow, ask a person or refuse. elan1 Run carries the work, but only Govern allows. Every product's actions take the same path and leave the same receipt, and no product builds its own gate.

elan1 Govern, AI governance

No. Your systems of record stay underneath, and stay yours. Agents work in Frappe, Zoho, Odoo, SAP, Microsoft Dynamics, Salesforce or Twenty, alongside Tally, helpdesks, commerce platforms and core banking. elan1 adds what sits above them: agents that do the work, a decision before each action and a receipt after it.

No. Most start with one product and elan1 Govern. Every product shares one approval model and one receipt.

Yes, through the MCP proxy or the SDKs.

Developer docs for the MCP proxy and SDKs

An agent builder helps an agent act. elan1 decides whether it may. elan1 Govern sits between any agent, ours or yours, and your systems: allow, ask a person or refuse, with a receipt each time. Build with them, govern with us.

In India by default: AWS Mumbai, three availability zones, no US region. elan1 Run also runs in your data centre or on an appliance. Personal fields are masked before any model call.

Models from Anthropic, OpenAI, Google and open source, behind one runtime. A model may propose an action. It may never approve one.

elan1 Run, the agentic AI runtime

Start with one process.

Bring one process. We'll show you the path it takes.

  • Allowed · Approved by Asha R.
  • Held for CFO
  • Refused · rule 4.2
Illustrative data.

Add the next when it holds.

Approvals · Kolam ComponentsIllustrative
  • Allowed · Approved by Asha R.
  • Held for CFO
  • Refused · rule 4.2
RCPT-0042 · signed · chained
A product view with three decisions: allowed and approved by Asha R., held for the CFO, and refused under rule 4.2. Illustrative data.