AbideBotAI agents for teams that need them to stay up.
AbideBot runs your agents on an always-on private node, with response-time guarantees that are measured, not assumed, and an alert to your on-call channel when one is missed. Built for platform, SRE and ops teams in Hong Kong and across Asia.
It abides. Joining the waitlist is free. We take no payments until paid plans launch, planned for around May 2027.
Designed to keep running, and to tell you when it isn't.
Most AI assistants are chat apps with a server bolted on. AbideBot runs on its own purpose-built runtime, designed so agents can do operational work without surprises.
Response-time guarantees you can measure
Declare how fast each agent must respond. Every run is measured against that guarantee, and a miss is detected and reported, not discovered in a postmortem.
Alerts where your on-call already lives
A missed guarantee, a failed step or a crashed routine goes to Slack or PagerDuty with its cause attached.
An always-on node of your own
Each customer gets a separate virtual machine. Customers never share a machine or its core. Your agents keep working when your laptop is closed.
Sealed and locked down
The node runs AbideBot and nothing else. There is no shell and no way to start other programs, and the kernel ships hardened. You can check the image fingerprint against the public release.
Supervised, with automatic rollback
Every component is supervised, and a crash stays inside its boundary. A new routine version runs on probation, and the previous one returns automatically if it fails.
An audit trail you can hand to compliance
Changes are signed and versioned, permissions carry a signed record of where they came from, and the audit log can be exported.
Your agent is called Dude. Rename it if you like.
Every new agent starts out as Dude. Call yours triage, payments-watch or anything else. Whatever its name, it writes real programs for your tasks and checks them before they go live.
- Compiled, not improvised. The agent writes a routine and compiles it on your node. Compile errors go straight back to the agent, and it fixes them.
- Calls the AI only when it needs judgement. The routine itself runs as checked code, so the AI is not consulted on every step.
- Approvals only a human can give. Risky steps pause until someone on your team approves them. The agent cannot answer for you.
- A factory reset that always works. A protected core can boot with only the essentials, so a bad change can never lock you out.
A node, a desktop app, and one shared record
The node does the work, the native desktop app is where your team talks to it, and both share one store of facts.
Choose where it runs
A hosted node with us, or the same sealed image on your own infrastructure. Hong Kong data residency is planned for Business SLA nodes.
Describe the job
Ask in plain language in a room or thread. Set triggers by time or by event, and set the response time you need.
It runs, measured
The agent builds the routine and runs it on probation. Every run is measured against its guarantee, and misses go to your alert channel.
Review and audit
Status shows what is working and what is waiting for you. Only changes sync between devices, and the audit log can be exported.
Plans for a laptop, a team, or a regulated business.
Paid plans are planned for around May 2027. Until then every button adds you to the waitlist or starts a conversation. Nothing is charged.
Uptime figures are targets, not commitments, unless a plan says it includes an SLA with service credits. The Business SLA node's SLA is planned and not yet available.
Prepaid, capped, and never billed after the fact.
Plans include no AI credits. AbideBot handles model access for you, so you never need your own AI account or key.
Top up a prepaid balance
Save a card, then top up your AI balance by hand or turn on auto top-up.
Set your own limits
You choose the spending limits. They live in a protected core that agents cannot change.
Provider cost plus a disclosed markup
Usage is charged at the model provider's cost plus a small markup, which we disclose. When the balance reaches zero the AI stops. You never get a surprise bill.
There are no free AI credits and no free AI trial.
Questions, answered calmly
Who is AbideBot for?
Platform, SRE and ops teams, mainly in Hong Kong and Asia, who want AI agents that run continuously and behave predictably: triage, checks, reports and other routine operational work.
What does "response-time guarantee" mean here?
You declare how quickly an agent must respond. AbideBot measures every run against that, detects a miss and alerts you. It is a guarantee you set and monitor for your own agents, not a contractual service level. Contractual SLAs are planned only for the Business SLA node.
Is there an uptime SLA?
Not yet. The Team node is designed for a 99.5% uptime target, shown on a public status page, with no service credits. A 99.9% monthly SLA with service credits is planned for the Business SLA node, which is coming later.
Is my node shared with other customers?
No. Each customer gets a separate virtual machine. Customers never share a machine or its core.
Where is my data stored?
Hong Kong data residency is planned for the Business SLA node. If you need full control, you can run the same sealed image on your own infrastructure, with annual support available.
Can agents see all our files?
No. They can read and write only inside the folders you grant, one at a time, and you can revoke a grant at any time. The app tells you when granted files go to a hosted node and to the AI model provider. Code written by the AI never runs on your own computer with your computer's permissions.
How is AI usage billed?
Through a prepaid balance that you top up by hand or automatically. It is charged at the model provider's cost plus a small disclosed markup. When the balance reaches zero, the AI stops. Nothing is billed after the fact.
Why is my agent called Dude?
It is the default name. Rename it whenever you like.
When can we start?
A private beta is planned for around February 2027 and paid plans for around May 2027. Dates may move. Join the waitlist and we will email you.
Get on the list.
Leave your work email and the plan you are interested in. We will contact you when the beta opens. No payment is taken.