Connecting Archron

Archron sits between an AI agent and a business system. Both sides connect to Archron; they never connect to each other.

Two connections to make. They are independent, so either order works.

Before you start

• An Archron account. Sign up at platform.archron.app. Once you sign in, everything below happens in your project under Nexus, where Add business system and Add agent are the two buttons you need.
• For Salesforce: a System Administrator for the one-time install, and an edition with API access enabled. Enterprise, Unlimited, and Performance include it; Professional may require the API add-on. Check with your Salesforce administrator if you are unsure.
• For HubSpot: nothing in advance. There is no admin step and nothing to install on your portal.
• Sandbox is supported. Salesforce can be installed into a sandbox first, and most teams should start there. Nothing about the setup changes when you move to production.

Part 1. Connect your business system

Two systems are supported in production today: Salesforce and HubSpot. Both start in the same place. In Archron, open your project, then Nexus, then Add business system, and pick your system. The drawer that opens is your starting point. Salesforce hands off to Salesforce for the install and the sign-ins; HubSpot is done entirely in that drawer.

Salesforce

Setup starts in Archron and finishes in Salesforce. After the one-time admin install, everything a person does is inside the Archron app in Salesforce.

1a. Install the package (admin, once per org)

In the Salesforce drawer, Archron shows two install links, Install in productionand Install in sandbox. Click the one you need. You are taken to Salesforce, sign in as an administrator, and install the Archron managed package into your org the way you would install any package. The package version is shown next to the links, so you always know what you installed.

1b. Connect the org (admin, once per org)

Still in Salesforce, open App Launcher, Archron, Archron Setup and click Connect organization. That sends you through OAuth back to Archron, where you choose which project this org belongs to. Archron then provisions a private OAuth application inside your own org and registers it to that project.

You never create a connected app by hand, and no keys are shared in either direction. The Archron Setup page then reads Connected to Archron, and back in Archron the drawer lists the org. Additional orgs are connected the same way: install, then connect.

To undo it: Disconnect this org from the Archron Setup page tears down both sides cleanly. Reconnecting re-provisions from scratch.

1c. Each person links their own Salesforce login

Once the org is connected, each person opens App Launcher, Archron, Archron Setupin Salesforce and clicks Connect to Archron under Your connection. That is a second, separate OAuth, this one for them personally. Nothing to type, no domain to look up. Their Salesforce identity becomes their Archron membership automatically.

Until the org is connected, this is deliberately blocked: the page tells them to connect the organization first.

Back in Archron, the Salesforce drawer's Your Salesforce connection card is where you manage that link: whether you are connected, which instance you are connected to, Disconnect to revoke it, and the Metadata deploys switch, the per-person consent that decides whether Archron may deploy schema changes you approve. It is off until you turn it on.

HubSpot

There is no admin step and nothing to install on your portal. Archron connects through a HubSpot application that works for any portal, so there is no app to create, no keys to generate, and no HubSpot-side configuration. Each person connects their own HubSpot account from the drawer and signs in with their normal HubSpot login. That is the entire setup.

Why the per-person step matters

This is the least glamorous step and the most important one. Archron does not act as a service account with elevated rights. It acts as you, through your own credential, with your own permissions.

• The agent's view of your schema is your view. Fields you cannot see, it cannot see.
• Archron checks field-level access before attempting a write, and your system enforces its own permissions again on the way in.
• An agent can never do more than the person who authorized it. There is no elevated identity to escalate to.

Who may link is yours to control. Admins can restrict linking to an email-domain allowlist, and every link and disconnect is audit-logged either way.

Linking everyone does not change your bill. Archron is priced by the number of agents and business systems under governance, not by the number of people who connect. Link the whole team.

Part 2. Connect your agent

One connector URL. One OAuth flow. Every agent. Archron is agent-agnostic by design: there is no per-agent integration, no plugin to build, and no separate surface per vendor.

In Archron, open your project, then Nexus, then Add agent, and copy the connector URL. It looks like:

Add it wherever your agent keeps connectors.

Claude, ChatGPT, and Grok have each been validated end to end. Anything that speaks MCP connects the same way. Archron detects which agent authorized and labels the connection, but the behavior is identical for all of them: same tools, same JSON contract, same commit gate.

Authorize when prompted. You are sent to Archron to sign in and approve, and you select the project your business system is connected to.

A local extension for Claude Desktop exists for setups where a remote connector is not an option; it uses a workspace-issued token instead of OAuth. Most people should use the connector URL above.

Optional. Declare your business rules

Some rules a business system only enforces in its own interface, not through its API. An API-driven agent could otherwise route around them. Those rules can be declared to Archron and enforced in flight, so they apply to agent writes too.

This is set up in Archron and does not require code. It is optional: the connection works without it, and most teams add rules once they have watched a few writes and know which constraints they care about most. If you would like help identifying them during a pilot, ask us.

Verify it. The first write

Setup is not finished when the connections turn green. It is finished when you have watched one write go through the gate. In a fresh conversation, ask your agent to create a record, an Account, a Company, whatever your system calls it:

What should happen, in order:

• The agent discovers the connected system. It does not ask you for field names, because Archron supplies the schema.
• It submits a structured intent to Archron. Nothing has been written yet.
• If anything required is missing or ambiguous, Archron returns a structured clarification and the agent asks you in plain language. It does not guess.
• Archron produces a summary of exactly what will change, and the agent waits for your explicit confirmation.
• Only then does it commit, and Archron reads the record back to confirm what actually landed.

Then open Archron and go to Operations for the full trace: which agent, which human it acted for, every field, the verification chain, the clarifications, the commit receipt, and the record ID.

What you did not have to do

• No data migration. Your records never leave your system. Archron stores schema and rules, not your data.
• No credentials in the agent, at any layer.
• No service account with elevated rights.
• No per-agent integration work. A second agent is Part 2 again, nothing else.
• No HubSpot app to create, and no Salesforce connected app to hand-configure.

What is on from day one

No feature flags to enable. Once a write has gone through cleanly:

• Every write is gated. The agent proposes; Archron decides whether it is safe.
• High-impact actions require verbatim confirmation: deletes, bulk operations, schema changes, and access grants.
• Outcomes are read back and verified, so committed means the record actually says what the receipt says.
• Changes can be undone from the audit log, through the same verified pipeline that wrote them.
• Business rules your system only enforces in its own interface can be declared to Archron, so they apply to agent writes too.

Building your own agent

If you are building an agent against Archron rather than connecting an off-the-shelf one, the tool contract and structured-intent rules are documented separately. Email support@archron.app and we will send the agent guide and the JSON contract.

Not connected yet?

Create an account at platform.archron.app and connect your first agent and business system. For multiple agents or systems, contact sales.