Skip to main content

Developers

Your agents propose. One path allows.

  • Route your agent's tool calls through elan1 Govern, by MCP proxy or the REST API. Each call gets one of 3 answers: allow, ask a person, refuse. Unmatched calls are refused.
  • elan1 Decide returns typed answers (Boolean, Choice, Score) with calibrated confidence.
  • elan1 Run puts your AI stack behind versioned contracts, each with at least 2 open-source implementations.
  • elan1 Work applies ERP and CRM changes through the same path.
  • Works with Claude, the OpenAI Agents SDK, LangChain and Cursor.
  • Every decision leaves a hash-chained record your auditor can verify offline.

govern · session

tool_call → govern  agent    ap-agent-02  tool     erpnext.purchase_invoice.submitdecision: ask_person  approver finance-head · approvedgovern verify →Chain intact
Illustrative. Kolam Components is a fictional company.

What is the elan1 API and SDK for AI agents?

The elan1 API and SDKs are designed to send each AI agent tool call through elan1 Govern before it runs. A REST API described in OpenAPI 3.1, TypeScript and Python SDKs and an MCP server return one of 3 answers, allow, ask a person or refuse, and every decision leaves a hash-chained record. Sandbox access is on request.

Get started

Three ways in.

Documentation

Everything in the docs.

Full documentation opens at launch, with its own search. Until then, start with the quickstart below or ask for a sandbox.

  • Quickstart

    Route one tool call and read its record.

  • Concepts: the governed path

    Ask, check, approve, act, record.

  • API reference

    Every endpoint, in OpenAPI 3.1.

  • SDKs: TypeScript and Python

    Install, configure, propose.

  • MCP server

    Govern tool calls without changing agent code.

  • govern CLI and offline verifier

    Check a record on your own machine.

  • Contracts for the AI stack

    The 10 elan1 Run capability contracts.

  • Changelog and status

    What changed, and what is running.

Quickstart

Propose, don’t act.

Your agent proposes the action. elan1 Govern answers allow, ask a person or refuse, and only what was decided runs.

  1. 1.Install the SDK: npm install @elan1/govern
  2. 2.Set ELAN1_API_KEY from your sandbox.
  3. 3.Wrap the tool call, as shown.

The full quickstart opens with the documentation at launch.

propose.ts

import { Govern } from "@elan1/govern"; const govern = new Govern({ apiKey: process.env.ELAN1_API_KEY }); // 1. Propose the action instead of taking it.const decision = await govern.propose({  agent: "ap-agent-02",  tool: "erpnext.purchase_invoice.submit",  target: "PINV-2026-00410",  params: { supplier: "SUP-0193", amount: 184500, currency: "INR" },}); // 2. One of 3 answers. Unmatched calls are refused.switch (decision.outcome) {  case "allow":    await decision.execute(); // runs with credentials the agent never sees    break;  case "ask_person":    console.log("Waiting for", decision.approver);    break;  case "refuse":    console.log("Refused:", decision.reason);    break;} // 3. Every outcome leaves a record your auditor can verify.console.log(decision.recordId);

Or route through the MCP proxy

Point your agent’s MCP configuration at the proxy. The agent’s code does not change.

mcp.json

{  "mcpServers": {    "erpnext": {      "command": "govern",      "args": ["mcp", "proxy", "--upstream", "erpnext", "--rules", "kolam-finance"]    }  }}
Illustrative. Package, command and field names will follow the SDK reference, published with the documentation at launch.

Integrations

Integrations, by layer.

Applications on top. Govern in the middle. Run underneath.

Applications

elan1 Work

Agentic enterprise automation for ERP and CRM

Systems it changes

  • Frappe/ERPNext
  • Twenty CRM
  • Zoho CRM
  • SAP and Salesforce

Applications

elan1 Decide

Decision intelligence

Where work arrives

  • Email, APIs and queues

Where actions go, through elan1 Govern

  • Guidewire, ServiceNow, Zendesk, Finacle, SAP and Salesforce

Governance

elan1 Govern

Agent governance and assurance

Agents and frameworks

  • Claude, the OpenAI Agents SDK, LangChain, Cursor and your own code

How calls arrive

  • MCP proxy
  • REST API (OpenAPI 3.1)

Built-in connectors

  • Gmail and Zoho Books

Runtime

elan1 Run

Sovereign agent infrastructure

Capability contracts

  • 10 contracts, such as chat, search, models and memory, each with at least 2 open-source implementations

Where it runs

  • Your cloud account in India, or an operator's data centre
  • On-premises and air-gapped

Product names belong to their owners and are listed to describe compatibility, not partnership. None of the four platforms is generally available yet; see the claims register.

Reference

Read the words. Read the records.

Glossary

42 terms in plain English, from allow and ask a person to hash chain and shadow mode. Each has its own link.

Open the glossary

Proof record specimens

Four specimen records, one per outcome, with all 16 fields. Every value is illustrative.

All specimens

Status and changelog

What is running, and what changed.

Service status

Live status and incident history open at launch. Until then, see what is built, designed and planned.

Status at a glance

Changelog

Every API, SDK and contract change, versioned, with what to do about it.

Opens with the documentation at launch.

A developer works at his desk in a daylit office, eyes on the task

Sandbox

Already have access?

Sign in to the console to manage keys, write rules and export records. No access yet? Ask for a sandbox and we reply within 2 working days.

Book a technical session.

An hour with the engineers who build the platform, on your agent, your tools and your rules.

Intelligence you can answer for.