OpenClaw™ framework
OPEN
CLAW.
The personal agent that works for you.
Open. Sovereign. Yours.
A custom AI agent built around your workflow. Reads your inbox, owns your calendar, researches in your voice, drafts on your behalf, and stops when you tell it to. Built on open weights and open-source runtimes — your data, your hardware, no vendor lock-in.
The window is open.
Personal agents went from research demo to production tooling in 18 months. Four things shifted at once — and they don’t close again.
Meet Hermes Agent.
Not every business needs a fully bespoke build. Sometimes you want the best open-source agent on the planet — deployed on your own infrastructure, wired to your tools, and managed so it never falls over. That’s Hermes Agent, built by Nous Research: a self-improving, open-weight agent that gets more capable the longer it runs. We deploy it, configure it, and keep it alive for you.
A built-in learning loop. Hermes creates its own skills from experience and improves them while it runs. It gets more capable the longer it works for you.
Agent-curated memory with cross-session recall. It remembers past work, builds a model of how you operate, and carries context forward instead of starting cold every time.
Local, Docker, SSH, Daytona, Singularity or Modal. A cheap VPS, a GPU box, or serverless with near-zero idle cost. Your hardware, your call — we deploy it there.
Reaches you across 20+ platforms — Telegram, Discord, Slack, WhatsApp, Signal, Teams and more. Direct the agent from your phone, get its work back in the same thread.
Web search, extract, browse, vision, image-gen, TTS, and programmatic tool-calling that collapses multi-step pipelines into one run. Plug in any MCP server. Add portable skills from agentskills.io.
MIT-licensed and open-source. Point it at Nous Portal, OpenRouter, OpenAI or your own endpoint — swap models per task. No vendor owns your agent. You do.
Not sure? That’s literally what the strategy call is for. We’ll tell you straight.
We deploy Hermes on your VPS / cloud / on-prem, wire it to your tools and channels, configure memory + governance, and keep it running. You own the box. Cancel the management anytime — the agent stays yours.
We code the tools.
We train the agent.
You own the stack.
OpenClaw isn’t a SaaS rebrand. Every engagement is a real engineering build — hand-coded tools, trained behaviour, governed runtime, branded dashboard, source code in your repo. Here’s the literal deliverable.
export const draftReply = openclaw.tool({
name: 'draft_reply',
desc: 'Draft a reply in Nikolaos\'s voice',
input: z.object({
threadId: z.string(),
intent: z.enum(['accept','defer','reject']),
}),
run: async ({ threadId, intent }) => {
const thread = await gmail.read(threadId);
return await llm.draft({
voice: VOICE_PROFILE,
tone: TONE_RULES[intent],
thread,
});
},
});actions:
draft_reply: { auto: true, log: true }
send_email: { auto: false, approval: human }
schedule_meeting: { auto: false, approval: human }
update_crm: { auto: true, log: true }
payment_action: { auto: false, approval: human }
models:
default: llama-4-70b-instruct
hard_reasoning: claude-sonnet-4-7
max_tokens_per_task: 8000
memory:
vector_store: pgvector@your-vps
retention: 90d
pii_redact: trueHow OpenClaw agents think.
Every action runs through the same four-step loop. Every step is logged, auditable, and reversible. No black-box reasoning.
The agent watches your sources of truth — inbox, calendar, sheets, drive, CRM, your own files. It builds memory.
On a trigger (new email, scheduled tick, your voice command), it plans. Multi-step, with tool selection and self-critique.
It uses tools — write a draft, book a slot, fire a webhook, query a database, send a Slack — under your policy and approval rules.
You see exactly what it did, why, and the evidence trail. Approve / reject / edit. The next loop learns from your call.
Six layers. All open.
No proprietary SDKs in the critical path. Every layer is swappable — today’s best model can be replaced by tomorrow’s without rewriting your stack.
From solo operator to enterprise fleet.
Same architecture. Different scope. Pick the one that matches where your business is right now — you can always upgrade the build later.
Your own agent. One operator. Built around your workflow.
A team of agents. Multiple roles. Wired into your stack.
On-prem. Sovereign. Built to your compliance posture.
Built for operators.
These are the shapes of business OpenClaw fits cleanly. If yours isn’t listed and you think it should be, that’s usually a sign you should book a call.
A chief-of-staff agent that drafts every reply, defends focus blocks, and keeps you out of the inbox.
After-hours buyer triage, open-home RSVPs, post-inspection follow-ups — all in your voice.
Client intake at EOFY, BAS reminders, document chasing. Tier-1 questions handled before staff touch them.
Conflict checks, intake summaries, deadline tracking. Read-only first, action-capable once trusted.
Brief intake, scope drafting, status updates to clients without a project manager.
One agent per business unit. They talk to each other. You see the dashboard, not the chaos.
The deliverable. No fluff.
Persona, voice, context — all yours. Talks like you when it drafts.
Direct command access. Scriptable. Pipe into anything.
See what the agent is doing, approve outbound actions, edit memory.
You own the repo. We hand it over with a runbook your engineer can use.
Pulse-hosted, your VPS, your cloud account, or your on-prem. You pick.
New tools, new prompts, new triggers — wired in for the first 30/90 days.
How we ship.
Five steps from kickoff to handover. No scope creep, no surprises.
Read this before you book.
Honest answers.
Wait — is "OpenClaw" a product I can buy off the shelf?
Why not just use Claude / ChatGPT / Copilot agents?
What models do you actually run?
How safe is it to let an agent reply to my emails?
What about hallucinations / wrong answers?
Can it handle voice?
What happens if I want to change models / providers later?
Australian-hosted option?
How does this compare to hiring a VA?
Will the agent learn my style?
Stop doing the work.
Build the agent.
30-minute strategy call. Free, no deck, no pitch. You walk away with a clear scope — or an honest “don’t build it yet”.