Your data, fully under your control
Export everything. Delete everything. Inspect everything.
AI safety you can trust.
Command your own agents. Fast. Easy. Safe.
OpenClaw and Hermes are the best agent orchestration tools available — and they can't safely run on your laptop. They need their own server instance with full system access to do what makes them valuable. We provision it, install everything, harden the OS, and hand you a working agent. You bring your API keys.
What Dorne.ai gives you
Everything you need to run agents safely. Nothing you don't.
A single, opinionated platform for orchestrating AI agents — with safety and observability guarantees built into the core. Not bolted on.
Multi-agent orchestration
Run teams of agents. Hot-swap models. Full logs available for debugging if something goes sideways.
Your choice of AI
Anthropic Claude (Sonnet, Opus) and OpenAI GPT (4o, o1). Switch anytime. Use whichever works best for the task. No vendor lock-in.
Production observability
Token, latency, and cost telemetry per agent, per run, per tenant. Slice by anything. Alert on anything.
Tamper-evident audit log
Every prompt, tool call, and decision is hashed and logged. Hand your auditors a CSV and a clean conscience.
Security updates handled
OS patches, dependency updates, security fixes — we keep your server current so you don't have to think about it.
Pro
Small local model inference, on your own machine
The Pro tier comes with a larger server instance capable of running lightweight models directly on your hardware — no API call required. Heavy lifting still flows through your Claude / GPT keys. Pro gives your agent local capabilities for the tasks that don't need a frontier model.
-
Local embeddings for smarter semantic memory and document search.
-
Speech-to-text — your agent transcribes voice notes locally (whisper.cpp).
-
Small vision models for image understanding, no roundtrip required.
-
Faster retrieval. Memory and search run on your hardware. Zero latency.
Full local LLM support on GPU instances coming later this year.
Privacy, by default
Your data is yours. Full stop.
Most AI platforms treat your data as their training set. Dorne.ai treats it as a liability — minimized, encrypted, scoped, and deleted on your command. We built the platform we wanted to use.
- You own every artifact. Prompts, completions, tool outputs, embeddings — exportable anytime as portable JSON.
- Your API keys, your machine. Entered during setup, stored only on your instance, never transmitted to us.
- Nobody reads your messages. Your conversations never transit our systems. Our control plane handles billing and provisioning — zero access to agent data.
- Hardened by default. SSH key-only, disabled root, locked-down firewall, API keys at 600 permissions.
- Walk away with everything. Cancel anytime and export your data. Your agent configuration is fully transparent — you can maintain the server independently.
- Architecture, not a promise. "Trust us" isn't good enough for the data an AI agent handles. We built it so we couldn't peek even if we wanted to.
Pricing
Simple plans. No surprises.
Both tiers are single-tenant with no shared infrastructure and no data limits. You bring your own Anthropic and OpenAI keys — we never touch them. Larger Enterprise offerings coming later.
Basic
$20 / month
Or $18/mo billed annually ($216/yr)
A working agent orchestrator on its own hardened VM. Everything you need to get started.
- 2 GB RAM, 20 GB disk
- Single-tenant VM, dedicated filesystem
- Signal + Telegram, persistent memory
- All API-based AI models (Claude, GPT)
- SSH-hardened, firewalled, auto-restart
- OS updates & security patches handled
- Email support
Pro
$100 / month
Or $95/mo billed annually ($1,140/yr)
A larger instance plus local inference for the work that shouldn't leave your machine.
- 4 GB RAM, 60 GB disk
- Everything in Basic
- Local embeddings for semantic memory
- Speech-to-text via whisper.cpp
- Small vision models for image input
- Faster retrieval — zero-latency local search
- Priority email support
GPU instances for full local LLMs coming later this year.
FAQ
Answers to what you're already wondering.
Do you train on my prompts or data?
No. We do not train any models on your prompts, completions, tool calls, embeddings, or any other tenant data — ever. This is a contractual commitment in our DPA, not a setting you have to find and toggle.
Do I need to know anything about servers?
No. We handle everything: provisioning, configuration, hardening, updates, and monitoring. You interact with your agent exclusively through Signal and Telegram.
Where is my data stored?
On your own server that you control, provisioned by Dorne.ai.
Can I bring my own model API keys?
Yes. Dorne.ai supports BYO keys for OpenAI, Anthropic, AWS Bedrock, Azure OpenAI, and self-hosted endpoints. Your credentials are stored on your server.
What if I want to leave?
Cancel and keep everything. Your data, your agent configuration — all yours. Our deployment is fully transparent, so you have everything needed to maintain the server independently.
How does billing work?
Pro is billed monthly per active user via Stripe. You can upgrade, downgrade, or cancel anytime from the billing page.