Governance for agents that act

Smart enough.
Cleared to act.

Enhydra Lab is the control plane for enterprise AI agents — permissioned tool-use, a full audit trail, and an agent on your own machine. Deployable today.

Read is governed  ·  Write is permissioned  ·  Everything is logged

How it works

Three surfaces.
One control plane.

A customer-service agent on your website. Shared agents for every department. A personal Enhydra Agent on each employee's machine. All three run on one governed plane — the same knowledge, permissions, confirmations and audit trail.

Create your account

Already have an account? Sign in

Walkthrough. On the vendor's website, a customer asks the support widget whether their gateways are affected by an exploit CISA flagged today; it checks the security advisory against their installed firmware, finds three exposed devices and, once the customer confirms, runs a three-step workflow: find a patch window, open the change request, brief their admins. Inside the company, a support rep picks the Finance Copilot from the shared chatbots to check a disputed invoice against the customer's contract, scoped to their role, then runs the Finance-built billing dispute workflow. On the employee's own machine, Enhydra Agent reads the Q3 files, runs code, generates a cover image, asks for approval before writing files, and delivers a deck and a summary. The control plane reads the knowledge base, runs the workflows, checks permissions, waits for the human, and logs all three.

how it works · three surfaces · one plane

CS agent widget

Public · your website

Reasons over your docs and live data, then runs multi-step workflows: book the patch, open the change, brief admins.

Shared enterprise agents

Internal · every department

RAG chatbots and agents built once, shared across the org, scoped by SSO role.

Enhydra Agent

Personal · your machine

Decks, documents, spreadsheets, images, code — it works in your files and asks before it acts.

Same knowledge · same permissions · same audit trail — on every surface.

Local execution · data residency

The agent comes to your data.
Not the other way around.

A cloud-only agent lives in a vendor's container: you upload the project, it runs there, you download the result. Enhydra Agent works on your own machine — running commands, writing files, driving Git and the tools you already have — and it asks before it acts.

Cloud-only agent in a vendor's container Enhydra Agent on your machine127.0.0.1
Your project Uploaded to their container Stays on your disk — a model sees only the context it needs
Artifacts Downloaded from their cloud Written straight into your folder
History & memory Kept on their servers In a local database on your machine
Toolchain Whatever their sandbox provides Your own Git, CLIs and compilers
Model Their pick, on their retention terms ZDR endpoints, BYOK, or open weights you host

ZDR zero data retention — the provider keeps nothing you send. BYOK bring your own key — your own model account, billed to you.

Industry know-how · private data · workflows — held in your own hands.

Enhydra CRM · chat channels

A new CRM, built on AI.
From the first reply to the next purchase.

LINE, Messenger, Instagram, WhatsApp, Telegram, Slack and your website chat arrive in one inbox, where AI answers first and hands over when a person is needed. Broadcasts, journeys, coupons, games and member cards then turn LINE friends into repeat customers.

Inbox

AI keeps an eye on every conversation.

  • Social CRM

    Seven channels in one inbox, with each customer's history, tags and member data in one place.

  • AI customer service

    AI answers first, from your own knowledge base. A person can take over at any time.

  • 17 AI judgments

    AI tells you who is waiting, who needs attention first and who is ready to buy. The decision stays with you.

  • Go live on your own

    Guided setup takes you from your first channel to the first AI reply. Customers see your brand; a custom domain is an add-on.

Example inbox. Conversations from LINE, Messenger and Instagram sit in one list. The AI has flagged a question about a late order as Needs attention and a group order as High intent; in another conversation the AI assistant is replying. In the open conversation the AI assistant answers a shipping question, hands the conversation to the team when the customer asks for a person, and an agent writes the reply.

enhydra crm · inbox 7 channels · 1 inbox

Marketing

AI plans every campaign with you.

  • AI campaign planner

    Describe a campaign in one sentence: AI plans the timing, writes each broadcast and adds a coupon. You review it first.

  • Automated journeys

    Welcome new friends, then nudge them about unused coupons and unfinished checkouts. Set it up once; it runs itself.

  • Coupons and games

    Coupons for your store or your website. Customers spin a wheel or scratch a card, and the coupon they win lands in their wallet.

  • Member cards and referrals

    Members collect points and move up tiers in LINE; when a friend they invite joins, the reward arrives on its own.

Example AI campaign plan. The merchant writes one sentence: a VIP offer, 10% off from 11/1 to 11/30, with a coupon. The AI plans three broadcasts: a kick-off on 11/1 to the whole audience, a reminder on 11/15 to those who haven't claimed the coupon, and a last call on 11/28 to those who claimed it but haven't used it. The first broadcast is previewed as customers will see it in LINE, with a Claim coupon button. Nothing is sent until the merchant creates and schedules it.

enhydra crm · marketing 1 sentence · 3 broadcasts

AI answers · AI judges · AI plans · people decide.

FAQ

First questions.
Straight answers.

What security, IT and business teams ask first — answered from what ships today.

Another question? Book a walkthrough

What is Enhydra Lab?

Enhydra Lab is the control plane for enterprise AI agents. Three surfaces — a customer-service agent on your website, shared agents for every department, and Enhydra Agent on each employee's machine — run on one governed plane: the same knowledge, permissions, confirmations and audit trail.

What can the agents do?

They answer from your documents with cited sources, and they act: multi-step workflows across your systems — ERP such as SAP, CRM, billing, internal APIs. Enhydra Agent also works in your files: decks, documents, spreadsheets, images and code. Consequential steps wait for a human.

Is it turnkey? Can we launch it under our own brand?

Yes. The website widget, the team chat app, the admin console and the governance layer ship as one finished product — no framework to wire up, no front end to build. One line of embed code puts the widget on your site in your colors, logo and name; custom domains and "Powered by" removal are add-ons.

Does our data leave our machines?

With Enhydra Agent, the agent comes to your data: your project stays on your disk, a model sees only the context it needs, and artifacts and history stay on your machine. For the whole platform, self-host in your VPC or on-prem and prompts, knowledge and logs stay in your perimeter.

Which AI models does it work with?

Claude, GPT or open weights — your choice. Use zero-data-retention (ZDR) endpoints, where the provider keeps nothing you send; bring your own key (BYOK), billed to your own model account; or run open weights you host.

How do you stop an agent from doing something it shouldn't?

Every tool call runs with least-privilege permissions, consequential actions wait for an explicit human approver, and retrieval is read-only. Every call, approver and timestamp lands in an append-only audit trail. Read is governed · Write is permissioned · Everything is logged

Can departments share agents without oversharing?

Yes. Build an agent once and share it across the organization; each agent's access is scoped from your SSO groups — by organization, department, role or individual.

Where does it run?

In our cloud, in your VPC, or fully on-prem, managed from one admin console. Enhydra Agent runs on macOS and Windows.

Every "can we automate this?" becomes a governed, auditable workflow you can stand behind.

Read is governed · Write is permissioned · Everything is logged

Book a walkthrough