Hivelock Get early access

Open-source, self-hosted AI agent platform, in early access

AI agents that only open what you allow

Run your company's agents and workflows on your own server. People approve the steps that change things, and every action leaves a record you can check.

Get early access Try a sample run

Connects with the tools your team already uses

Put AI agents to work, without losing control

Read the docs

Start from real automations for sales, support, finance, HR, quality, legal and more. Each one installs as a draft for you to check.

74+ Ready-made templates

Connect mail, files, databases, chat, CRM and developer tools. Add anything else through a remote MCP server.

26+ Connectors

Give each team the access it needs, with roles for owners, admins, builders, operators and viewers.

11+ Business areas covered

Platform

Everything you need to run AI agents with confidence

Build, connect, approve and check your automations in one place.

Visual workflow builder

Drag agents, conditions, approvals and connector steps onto a canvas. Every save keeps a numbered version you can compare and restore.

People approve what matters

In chat, anything that changes data waits for a person. In a workflow, you place approval steps where you need them.

Agents that use your documents

Upload PDF, Excel, CSV and text files. Agents search them and name the file they used, so people can check the answer.

A record you can check

Follow live runs, model calls and tool calls. The audit log is hash-chained, so a changed record shows up.

  • Custom agents

    Describe an agent and Hivelock drafts it for your review. A person approves each memory note.

  • Triggers, including email

    Run by hand, on a schedule, from a signed webhook, or when mail reaches a watched folder.

  • Plug-and-play models

    Add a cloud provider in Settings, or a local model when you self-host, with automatic fallback.

  • Confidential mode

    Limit sensitive work to the model providers you have cleared for it.

How it works

From idea to safe automation in 4 simple steps

Start from a template or a blank canvas

Pick one of 74+ templates, describe an agent in plain language, or drag steps onto a canvas.

  • Templates install as drafts
  • Nothing runs until a person turns it on
  • No code needed

Sample run

See agents work, and read the log

Each run uses a different example agent. Decide what happens, then read its log. Nothing is sent, and no AI runs on this page.

Try a sample run

No AI running here
AgentInvoice checker RequestCheck invoice INV-2041 and pay the supplier
    Ready

    Security

    AI agent security, built as if one tenant will try to reach another

    Protection sits in the database and the network layer, not only in the screens people see.

    Row-level securityAES-256-GCMTwo-step sign-inOpenID ConnectAudit trail
    Read the security overview

    • Workspace isolation

      PostgreSQL row-level security limits workspace data to its own workspace.

    • Encryption per workspace

      Secrets, run data and chats use AES-256-GCM with a key for each workspace. Use disk encryption for uploaded files and the search index.

    • Network guard

      Connectors cannot reach private addresses unless a self-hosted operator allows them. In hosted mode they cannot at all.

    • Reviewed tools

      A changed MCP tool stays blocked until someone reviews it again.

    • Tamper-evident audit

      Each record is chained to the one before it, so a changed record shows up.

    • Mandatory two-step sign-in

      Password sign-ins always need it, and devices are not remembered. Single sign-on relies on your provider's second step.

    • Single sign-on

      OpenID Connect, with your domain proved by DNS. SAML goes through Keycloak.

    • Self-tested with Strix

      We test Hivelock ourselves with the open-source Strix tool, and we fixed what it found. Read the report.

    Why Hivelock

    Let agents handle the repetitive work. Keep people in charge of the decisions. Keep your data, keys and rules where you can see them.

    Use cases

    AI agents for real work in every industry

    See how teams could use agents for daily tasks, with people keeping the decisions.

    Pricing

    Simple plans. Bring your own model key

    You pay your model provider directly, so we never mark up your AI costs. Hosted plans are not live yet.

    Starter

    For a small team's first agents.

    $29/month

    Get early access
    • 3 people
    • 5 agents
    • 2,000 runs a month
    • 1 GB document storage
    • Email support

    Team

    For departments running real workflows.

    $99/month

    Get early access
    • 10 people
    • 25 agents
    • 10,000 runs a month
    • 10 GB document storage
    • WhatsApp channels
    • Email support

    Business

    For companies with security requirements.

    $299/month

    Get early access
    • 30 people
    • Unlimited agents
    • 50,000 runs a month
    • 50 GB document storage
    • WhatsApp channels
    • Single sign-on
    • Priority email support

    Paid plans are hosted by us, and a new hosted workspace starts with a 14-day free trial on Starter. Yearly prices are shown per month and billed once a year: $288, $984 or $2,988. Email support has no guaranteed response time.

    Connectors

    Connect AI agents to your business tools in one place

    Agents and workflows use the systems you already run, with credentials kept on the server.

    • MailSMTP, IMAP, Microsoft 365
    • FilesS3, Google Drive, OneDrive, SharePoint
    • DatabasesPostgreSQL and MySQL
    • Web and customHTTP APIs, remote MCP servers
    • ChatSlack, Teams, Discord, Telegram, WhatsApp
    • CustomersHubSpot, Salesforce, Zendesk
    • Developer toolsGitHub, GitLab, Jira Cloud, Linear
    • Productivity and commerceSheets, Notion, Airtable, Shopify

    See what each of the 26+ connectors does

    Questions

    Frequently asked questions

    Does my data leave my server?

    When you self-host, your database and files stay on your server. Agents send what they read to the model provider you choose, and connectors send data to the services you connect. A local model keeps the model step on your own network.

    Is Hivelock open source?

    Yes. The Community edition is open source under AGPL-3.0 and free to self-host, with 1 workspace, 5 people, 5 agents and 10 workflows. It keeps approvals, the audit log, two-step sign-in and any AI model. Single sign-on, WhatsApp, email triggers and the extra connectors need a paid edition. Get the code on GitHub or compare editions.

    Is there a hosted version?

    It is planned, and not live yet. We would run the servers and you would bring your own model key. Local models and local file tools are off in hosted mode. Compare the two.

    Can an agent act without permission?

    An agent can only use the actions you gave it. In a chat, anything that changes data waits until an authorised person decides. In a workflow nobody is watching, so an agent step cannot take those actions, and a connector step runs as you built it. Put an approval step before it.

    Which AI models can I use?

    Any service that speaks the OpenAI chat API. Settings has presets for Kimi, OpenAI, Anthropic, DeepSeek, Groq, OpenRouter and NVIDIA. When you self-host, you can also use local servers such as Ollama and LM Studio.

    Do I need to write code?

    No. Agents and workflows are built in the web app, and there are 74+ templates to start from. 59 work on the hosted version. Connecting a service needs credentials from that service, and self-hosting needs someone to run a server. Developers can add custom tools through a remote MCP server.

    How is Hivelock tested?

    We test Hivelock ourselves with automated tests and penetration testing with Strix. Some connections are still being validated with live services during early access. Read the security report.

    Is Hivelock certified?

    Not yet. An independent third-party penetration test is planned. The security overview explains what is in place today.