The platform

One runtime for rules, AI, and human judgment.

Most automation makes you choose — rigid rules or unpredictable AI. Hyphen runs both in one governed workflow, with your team holding the decisions that matter.

Read the docs
How it coordinates

One agent, orchestrating your workflows.

The agent runs work across multiple workflows, waits on results, decides what’s next, and escalates the judgment calls to your team — with every step logged.

The building blocks

Four parts. Each does what it’s best at.

WORKFLOWS

The work, defined once

Describe a process in plain English or define it precisely; Hyphen turns it into a workflow you can run by API, on a schedule, or from a form. Branch, loop, retry, and pause for people — defined once, run everywhere.

MATCHING ENGINE

Rules without AI cost

A built-in engine matches records by exact keys, number tolerances, date windows, and fuzzy text. It clears the cases that follow your rules instantly — no model call, no token cost, fully deterministic. The part most “AI agent” tools don’t have.

AI AGENTS

Bounded reasoning for the exceptions

When a case doesn’t fit the rules, an agent investigates — gathering data, calling only the tools you’ve granted, and building a recommendation. It works under stuck-detection and hard limits, so it can’t wander or escalate its own access.

APPROVALS & AUDIT

Humans hold the line

Low-confidence cases pause and route to your team with full context and the agent’s recommendation. Every step — rule, reasoning cycle, and human decision — is logged to an append-only trail, with sensitive data redacted before storage.

How it flows

Match what fits. Investigate what doesn’t. Approve what matters.

01 · INGEST

Data comes in, structure comes out

Send data from anywhere — webhooks, uploads, API calls. Hyphen normalizes and validates it before it touches your logic.

02 · PROCESS

Rules clear the obvious; AI reasons through the rest

The matching engine handles rule-based work with no AI cost. Ambiguous cases go to a bounded agent that investigates and recommends.

03 · APPROVE

When AI is unsure, it asks a human

The workflow pauses with full context and the recommendation. Your team approves, rejects, or redirects — then it resumes, logged.

Authoring

Describe it. Review it. Run it.

With Process Studio, an operator describes a process in their own words. Hyphen drafts the workflow, shows it back as a visual flow to confirm or edit, and — once approved — publishes it to the same runtime as everything else. No raw configuration required, and nothing runs that a human didn’t sign off on.

PROCESS STUDIO · BETA

Built to be embedded

Run it as a product, or run it inside yours.

Hyphen ships with a gateway for tenant-safe auth and rate limits, and an Ops Console SDK — drop-in components for approvals, data tables, and document feeds that connect straight to live workflows. Put governed automation inside your own product without building the operations layer yourself.

See the SDK

Why governed beats autonomous

The limit isn’t the model. It’s the guardrails.

Pointing a raw model at your operations is fast to demo and dangerous to run — no record of why it acted, no boundary on what it can touch, no human on the irreversible calls. Hyphen takes the opposite bet: the AI is powerful because it’s bounded.

Rules where rules are better. AI where reasoning is needed. A person on every decision that’s hard to undo. That’s what makes it safe to put in front of an auditor — and in production.

See the platform on your own work.

A 30-minute walkthrough on your own sample data — no slideware.