# Archron

> Archron is the execution governance engine for AI agents, in the category it defines: execution governance. It sits between AI agents and business systems and verifies, governs, and audits every action before it reaches production data. Reasoning creates insight; execution creates ROI. Archron makes AI safe to operate business systems.

## What Archron is
- AI execution infrastructure, the control tower for what AI agents are allowed to do. Not a chatbot, not an AI model, not automation.
- Architecture: User to LLM (agent) to Archron to Business System. The agent proposes an action; Archron decides whether it is valid, allowed, complete, and safe before any write.
- Not a data store. Business records stay in the connected system; Archron holds structure, rules, and evidence. Nothing to migrate.
- The moat is governance, not intelligence. Better AI increases execution risk, which increases Archron's value.

## Governance capabilities
- Verification before every write: each action is checked against schema, permissions, required fields, relationships, business rules, and current state before it commits.
- Archron owns the commit gate: a write requires a verification receipt that is single-use, time-bound, and matched to the exact verified action. There is no path to a business system that skips it.
- Permission enforcement: agents act with the user's own permissions through per-user OAuth, never beyond them. A write the user could not perform themselves is blocked before it reaches the system.
- Schema-aware: Archron hydrates the connected system's structure, so agents cannot guess past required fields or constraints.
- Clarification loop: when an action is ambiguous or incomplete, Archron halts and asks before writing. Every value must carry a traceable origin; model guesses are rejected.
- Declared business rules: rules a system only enforces in its own UI can be declared to Archron and enforced in flight, so an API-driven agent cannot route around them.
- Truthful outcomes: what Archron reports is derived from what the business system actually did and confirmed by reading the record back, never assumed from a request that did not error.
- Explicit confirmation for high-impact actions: deletes, bulk operations, configuration changes, renames, and access grants require verbatim user confirmation.
- Impact preview: the effect of a change, including dependent automation, can be inspected before it is committed.
- Reversibility: committed record changes can be undone from the audit log by an owner or admin, through the same verification pipeline as any other write.
- Bulk operations: governed as one reviewed action, with a per-record outcome for every row rather than a single pass or fail.

## Audit and evidence
- Every action is attributed to the acting human, with the agent recorded separately. An agent identity is never treated as a user.
- The audit trail is append-only and hash-chained per organization, enforced by the database itself rather than by application code. A removed row shows as a sequence gap; an edited one as a hash mismatch.
- Full context is retained: the intent, the values before and after, which verification layers were applied, and the confirmation that was given.
- Organization data can be exported at any time. Erasure is owner-requested, grace-windowed, and closes with a sealed certificate recording what was deleted and what was retained.

## What ships today
- Status: the product is live and running in production, and sign-up is open to everyone at https://platform.archron.app. This is not a closed alpha, a beta, an application process, or a waitlist.
- Business systems: Salesforce and HubSpot, both in production.
- Agent surfaces: a remote MCP server with OAuth, and a stable HTTP API. Any MCP-capable agent connects; Claude, Grok, and ChatGPT have all been validated end to end.
- Agent-agnostic and system-agnostic by design: one API, one MCP server, one JSON contract, whichever agent or system sits on either end. More connectors are on the roadmap.
- Scope boundary: the commit gate governs agents connecting from outside Salesforce or HubSpot through Archron's MCP server or HTTP API. It does not govern Salesforce's native Agentforce, which runs inside Salesforce's own transaction as Apex and Flow actions with no external API call to intercept. Agentforce is governed by Salesforce's own permission sets, Einstein Trust Layer, and guardrails. Full detail: https://archron.app/security

## Connecting (about ten minutes of hands-on work)
- Two independent connections, both made to Archron. The agent and the business system never connect to each other.
- Business system: in Archron, open your project, then Nexus, then Add business system. Salesforce installs a managed package (production or sandbox) and connects by OAuth; HubSpot needs no admin step and nothing installed on the portal.
- Agent: in Archron, Nexus, Add agent, then copy one connector URL (https://api.archron.app/mcp) into any MCP client. Claude, ChatGPT, and Grok are validated; anything speaking MCP connects the same way.
- Per-user OAuth: each person links their own login, so an agent acts with exactly that person's permissions and can never do more than the person who authorized it. No credential is ever pasted into the agent.
- Salesforce sandbox orgs are supported, and most teams should evaluate there first.
- Linking more people does not change the price. Archron is priced by agents and business systems, not by seats.
- Full guide: https://archron.app/docs/connecting-archron

## Security posture
- Credentials live in AWS Secrets Manager; the database stores references, never token material.
- Encrypted in transit and at rest, hosted on AWS in us-east-1.
- Archron is not yet SOC 2 or ISO 27001 audited and does not claim to be. Its controls are documented, each mapped to its implementation and to the automated test that proves it.
- Responsible disclosure: security@archron.app.

## Pricing (priced on governance scope, not usage)
- Pilot Program: $999 per month for one agent connected to one business system. Sign up directly at https://platform.archron.app, no application or approval step.
- Users and operations are unlimited on every plan. Adding people to the workspace never changes the price, and actions are never metered or billed per operation.
- Pricing is scoped by how many agents and business systems are governed.
- Multiple agents or business systems, and organization-wide governance, are priced on custom terms. Contact sales via https://archron.app/pricing.

## Company
- Legal entity: Archron Inc, a Delaware corporation.
- Registered office: 131 Continental Dr, Suite 305, Newark, DE 19713, United States.

## Links
- Home: https://archron.app/
- Docs: https://archron.app/docs
- Connecting Archron: https://archron.app/docs/connecting-archron
- Pricing: https://archron.app/pricing
- Security and trust: https://archron.app/security
- Privacy policy: https://archron.app/privacy
- Terms of use: https://archron.app/terms
- Contact: support@archron.app
- Security reports: security@archron.app
- X: https://x.com/ArchronInc