OpenClaw, Hermes, and Orto: AI Agents in 2026

Curated by Matt Perry

CTO

8 May 2026

Three AI agents, three very different jobs

AI agents are software that read your brief, pick the right tools, and get the work done with little help. Three projects are getting a lot of attention in May 2026. OpenClaw is a personal AI assistant that runs on your own laptop. Hermes Agent is a self-improving open-source assistant that runs on your own server. Orto is a hosted agentic workforce that we are launching at Original Objective.

The simplest way to think about it: OpenClaw is your personal assistant. Hermes Agent runs on your server. Orto runs the agency.

Each one fits a different job. Picking the wrong tool for your business can cost weeks of work, leak data, or land you with a bill no one signed off. This guide explains what each tool does, who it is for, and where it falls short.

What is OpenClaw?

OpenClaw is an open-source personal AI assistant created by Austrian developer Peter Steinberger. It started life as ClawdBot, ran on Mac Minis, and exploded to 60,000 GitHub stars almost overnight. After an Anthropic trademark dispute and a chaotic rebrand from ClawdBot to MoltBot, Steinberger forked the project under a clean name: OpenClaw. Open source, community governed, free from the baggage of the old brand.

That decision paid off. By May 2026, OpenClaw has 180,000+ GitHub stars and millions of installs across macOS, Linux, and Windows. It connects to WhatsApp, Telegram, your calendar, your inbox, and over 50 other apps. The official marketplace, ClawHub, hosts more than 5,700 community-built skills you can plug in with a click. You bring your own LLM key (Claude, GPT, or a local model) and OpenClaw turns it into a personal assistant that lives on your laptop.

People use OpenClaw to triage email, book travel, manage their calendar, run home automations, summarise news, and a thousand other personal-productivity jobs. It is genuinely powerful for a single user with the time to set it up well.

Use OpenClaw when:

• You want a personal AI assistant that runs on your own machine.
• You like an open plugin ecosystem with thousands of community skills.
• You want full data control with no cloud vendor.

Skip OpenClaw when:

• Multiple staff need to share work or a shared library.
• You need enterprise backups, single sign-on, or approval workflows.
• You need an audit trail your finance and legal teams will accept.
• A laptop failure cannot be allowed to take the work with it.

What is Hermes Agent?

Hermes Agent is an open-source AI agent built by NousResearch. It runs on your own server, learns from each session, and gets more capable the longer it runs. The May 2026 release added a Kanban board for multi-agent jobs, full voice chat, and 20 messaging platforms including Telegram, Slack, and Google Chat.

Hermes is local-first. Your data stays on your own machine. There is no telemetry and no cloud lock-in. You can deploy it five ways: local, Docker, SSH, Singularity, or Modal. A background curator grades and prunes the agent's skills on its own schedule, so the system slowly tunes itself.

Use Hermes Agent when:

• You want a personal AI that learns over time.
• You can run your own server and Docker containers.
• You want full data control with no cloud vendor.

Skip Hermes Agent when:

• You need a hosted product that just works.
• Finance needs an audit trail that ticks compliance boxes.
• You need to roll out the same setup to fifty staff in one click.

What is Orto?

Orto is a production-ready AI agentic workforce launching soon at orto.run. You brief it in plain English. An org chart of AI agents figures out who should do the work, does it, and reports back. Marketing copy, customer outreach, data ingestion, calendar bookings, prospect research, social posts. Whatever a service business needs running.

You write the briefs. You sign off the work. The technical side, the model choices, the security, the integrations, the data pipeline, is handled by Original Objective. Easy to use. Hard to build. We do the hard part.

Where Orto is different

Orto is built for businesses that cannot afford to lose data, miss compliance checks, or run out of cover when a laptop fails. This is not a hobby version.

Built for non-technical owners, by people who know the stack

You run a service business. You did not start it so you could become an AI engineer. Orto is designed so the day-to-day user just briefs the work and approves the outputs. The technical depth underneath is real and serious. You just do not have to touch it.

Production-ready and secure by default

Orto runs in the UK and runs in highly available containers. The whole platform is defined as code, so it can be rebuilt from scratch in under an hour. Single sign-on for every user. Every API key and password sits in a hardware-backed vault. An exfiltration guard scans every outbound message for secrets before they leave the system. Multi-tenant scoping is baked into every record so client data never crosses lines.

Built-in backups, audit trail, and a shared library

Our database is geo-redundant by default. Point-in-time restore covers the last 35 days. Storage for artifacts is replicated across two regions. Every brief, every tool call, every cost, every deliverable is recorded with a full audit trail. And it all lives in one searchable workspace: artifacts, documents, proposals, images, transcripts, recorded outcomes. Agents pull from the same place they save to. Reuse what works. Stop reinventing the same draft.

Quality checks on every job

A planner, generator, and evaluator triad reviews high-stakes work. The evaluator either accepts the deliverable or kicks it back with structured notes. Fourteen quality scorers grade every output across correctness, faithfulness, cost, latency, and tone. A canonical brief suite runs on every change to the platform itself, so a model swap or prompt edit that quietly degrades service gets caught early, not weeks later when a customer complains.

Learns what works, and what does not

Orto remembers good and bad results. Thumbs-down feedback, rejected drafts, and bad outcomes feed back into memory so the same mistake does not happen twice. Thumbs-up feedback, approved work, and good customer outcomes build up the skill library.

It goes further than that. You can replay any past run, fork off at any point, edit the inputs or the prompt, and capture what worked as a brand-new skill. The system turns one good outcome into a reusable playbook for the next time. Background scribes mine the results overnight and propose new memories and skills. Everything waits for your approval before it goes live.

Hire AI staff, load whole team packs

Need a new role? Open the dashboard, describe the job, and hire an AI specialist in a few clicks. The new agent appears in the org chart with the right tools and the right prompts already wired in. Replace one that is not pulling its weight. Promote a high performer to a more senior tier. Your AI workforce flexes with your business the same way a real team does.

You do not have to design the team yourself. Pick a pack: marketing agency, driving school, salon, sales operation, engineering team. Your workspace ships ready to run, with the agents already in place. Original Objective builds and maintains the packs. You hire from them.

Hundreds of tools, ready to use

Orto connects to hundreds of tools out of the box. Direct integrations for the everyday ones: Google Calendar, Gmail, Slack, GitHub, Telegram, social platforms, web search, image generation, video transcription. Add your own tools through any HTTP API. Anything that speaks the Model Context Protocol (MCP) plugs straight in. Each agent only sees the tool bundle their role needs.

Approval gates at every irreversible step

Inline approval cards live in the chat. Before a pull request merges, before a deploy goes out, before a payment is taken, before a customer email is sent, the agent waits for your yes or no. No surprises. No 3am bills. The agent does the work between the gates. You stay in control at the gates.

When something goes wrong, it picks up where it left off

Most agent frameworks are forward-only. One pass, fail, done. Orto records every step as a checkpoint, so you can fork from the last good state with a different model, a different agent, or an edited input. Failures become diagnostics, not lost work.

Any model you want, your own keys, your own cap

Orto is not locked to one model provider. Each agent can run on whichever model fits the job: Anthropic Claude (Opus included, for the heaviest reasoning), OpenAI GPT models, DeepSeek for cheap high-volume work, or local open-source models on your own hardware. In some setups you can even bring your own flat-rate ChatGPT subscription rather than metered API tokens, which keeps unit costs predictable for everyday work.

You can run Orto self-hosted with your own keys, on our hosted platform at orto.run, or mix the two. Specific jobs can route back to a hosted Orto instance when you need extra capacity or a different data-residency story. You set the cost ceiling on the key itself. Per-run and per-day guards catch runaways. Prompt caching is wired everywhere. Orto routes to the cheapest capable model for each job and only escalates when the work demands it.

Side-by-side comparison

When to choose which

The right tool depends on the job:

• Pick OpenClaw if you want a personal AI that lives on your laptop and you love the open plugin ecosystem. Skip it the day your work needs to be shared with a team.
• Pick Hermes Agent if you want a self-improving personal assistant on your own server and you do not mind the setup work.
• Pick Orto if you run a service business and need an agentic workforce that your finance and legal teams will sign off on, and you want a partner to build and maintain it for you.

This is not a hobby tool. Orto is built for the days when a missing backup, a leaked customer record, or a runaway bill is an actual problem.

What this means for your business

If you are a one-person shop tinkering on a side project, an open-source agent on your laptop is fine. The risk is small and the cost is zero. The day your AI agent starts running real customer work is the day all those missing controls become a problem. A leaked key, a deleted memory file, or a model that quietly burns £400 of credit overnight will cost you more than the price of a hosted platform.

Orto opens for early access in May 2026. Sign up at orto.run to be notified when it goes wider.

How Original Objective can help

At Original Objective, we have been building agentic AI systems for the kind of businesses that need real controls. We help you pick the right tool, set it up safely, and measure what it saves you.

Book an intro call or get in touch to talk through your setup.

More in AI Industry Trends

View all

12 May 2026

The Agentic Economy: A Simple Guide for UK Business

6 Apr 2026

WhatsApp Is Becoming a Sales Funnel: What Meta's Strategic Shift Means for Your Business

5 Apr 2026

Claude Code Source Code Exposed: What Happened and What It Revealed