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.

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.
Industries
Applications
Underneath
Your systems of record, underneath, not ours
ERP · CRM · helpdesk · commerce platforms · core banking · Tally · models from Anthropic, OpenAI, Google and open source
Industries
Applications
- elan1 Act, Agentic ERP
- elan1 Propel, Agentic go-to-market
- elan1 Resolve, Agentic service
- elan1 Pulse, Governed agentic analytics
Foundation
- elan1 Ontology, Relationship intelligence
- elan1 Decide, Decision intelligence
- elan1 Govern, AI governance. Only Govern allows.
- elan1 Run, Agentic AI runtime
Your systems of record, underneath, not ours
ERP · CRM · helpdesk · commerce platforms · core banking · Tally · models from Anthropic, OpenAI, Google and open source
The layers
What each layer does.
01
Industries
The same platform, set up for your sector's rules, systems and approvers.
Setup, not a build. Start with one process; add the next when it holds.
02
Applications
elan1 Act, Propel and Resolve do it. elan1 Pulse analyses it.
- elan1 ActAgentic ERP
- elan1 PropelAgentic go-to-market
- elan1 ResolveAgentic service
- elan1 PulseGoverned agentic analytics
03
Foundation
elan1 Run carries it. elan1 Ontology relates it. elan1 Decide judges it. elan1 Govern decides it.
- elan1 OntologyRelationship intelligence
- elan1 DecideDecision intelligence
- elan1 GovernAI governance
- elan1 RunAgentic AI runtime
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.
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.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.03
Finance head
Approved by Asha R.
The finance head, Asha R., approves. Illustrative data. Approve.
The finance head decides.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.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
- 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 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.
Yes, through the MCP proxy or the SDKs.
Models from Anthropic, OpenAI, Google and open source, behind one runtime. A model may propose an action. It may never approve one.
- Allowed · Approved by Asha R.
- Held for CFO
- Refused · rule 4.2