AI Agent Ops Install: Install a ready AI agent with limited access in 1–2 weeks
Install and harden an agent with RBAC, audit logs, MCP, A2A, and an operating procedure.
We reply within 24 working hours. No commitment.
Swipe the diagram sideways
AI Agent Ops Install: the agent works inside 3 layers with 1 approval gate
// problems we solve
Problems we often find.
You want an agent to support 1 workflow, but you worry it will read or change data outside its role.
We write role-based access limits before install: what data the agent can read, what actions it can run, and what actions need approval.
No one knows what the agent did last night.
Every agent action lands in an audit log that the operator and the owner can read.
An API key sits in a config file that anyone can read.
Hardening moves each secret to safe storage and limits server access to approved people.
You cannot decide between OpenClaw, Hermes, or another product.
Day 1–2 picks the product from your workflow, channel, and tool needs, not from a trend.
// how it works
The steps we run.
Swipe the diagram sideways
Pick product, set limits
We map your workflow, channel, tool, and user list, then pick OpenClaw, Hermes, or QwenPaw and write the role-based access matrix.
Output: An access-limit document and a user list per role.Install and harden the agent
We install the agent on your server, move each secret to safe storage, limit server access, and set up a configuration backup.
Output: A live agent with safe secrets and limited access.Connect tools and audit log
We connect your tools through MCP, set up the chosen channels, add an A2A baseline, and turn on the audit log.
Output: A list of active tools and channels with a log you can check.Test 1 workflow, with approval
We run the first workflow start to finish, with an approval gate on each action that carries impact.
Output: A test report: automatic actions, actions that asked for approval, and actions that were refused.Write the SOP and hand over
We write the operator SOP: how to assign a task, how to approve one, how to stop the agent, and how to read the log.
Output: An SOP document and a 60-minute handover session.// what you get
A clear scope. Useful handover.
Install and harden an agentAI agentAn AI program that performs work steps by itself, for example reading a message, drafting a reply, and recording the result.Open the glossary with RBAC, audit logs, MCPMCPA standard that lets an AI program use your tools and data under clear permissions.Open the glossary, A2A, and an operating procedure.
- AgentAI agentAn AI program that performs work steps by itself, for example reading a message, drafting a reply, and recording the result.Open the glossary installation and hardening
- RBAC and audit logs
- MCPMCPA standard that lets an AI program use your tools and data under clear permissions.Open the glossary and A2A baseline
- SOP and handover
Format: a live agent on your server, an access-limit document, a readable audit log, a 1-workflow test report, and a 4–6 page operator SOP.
Technical detail for the IT team
- Hardening
- We close the access that the server and the agent do not need.
- Access rights (RBAC)
- Each agent opens only the data you allow.
- Activity record (audit log)
- Every agent action is recorded with its time.
- MCP
- The protocol that connects an agent to your tools.
- A2A
- The protocol that lets 2 agents share a task.
- SOP
- A written guide so your team can run it without us.
// price and inclusions
What the price covers.
| Included | Not included | Billed apart |
|---|---|---|
|
|
|
Ground rules
- Revision: 1 round of changes to the access matrix and the SOP after the workflow test. A second workflow or extra tools get a separate quote.
- Schedule: Installation runs 1–2 weeks from the day server and model-provider access is complete. Cancel before install at no cost; after that, we bill the finished steps.
- Data and access: You keep the server, accounts, secrets, and data. We use limited access during the project and remove it after handover.
- Warranty limit: The agent passes the agreed workflow test with human approval. We do not guarantee an error-free agent; that is why the approval gate and the log exist.
What we guarantee
- We answer your message within 24 hours on a working day.
- Accounts, code, and data stay in your name from day 1.
- Every handover includes a test report or written notes.
- No renewal fee to Rama Digital. You pay domain, hosting, and third-party subscriptions to the vendor directly.
- A fixed price on this page is the price we invoice. A "From" price becomes fixed in the written quotation you approve.
// is this for you
Good fit for
- You already know the task the AI assistant must support.
- You want written access limits and an activity record.
- You have your own server or cloud account.
Not a fit for
- You do not know the task yet. Choose AI Diagnostic.
- You already chose Hermes Agent. Choose Hermes Agent Installation Service.
What you prepare
- Access to the server and the model provider account.
- A list of people allowed to use the AI assistant.
- 1 task to support first.
What the price excludes
- Server fees and AI model usage fees.
- New application development.
How it differs: Hermes Agent Installation Service installs 1 product only, while AI Agent Ops Install picks the product and sets the access limits.
// how this differs
Compare with other services.
| AI Agent Ops Install | Hermes Agent Installation Service | AI Workflow Audit | |
|---|---|---|---|
| Duration | 1–2 weeks | 2–5 working days | 1–2 weeks |
| Price | IDR 7.5–25M | IDR 6M | IDR 6M |
| Result | Agent live with access limits and a written SOP. | Hermes Agent live with a gateway and a basic SOP. | Use-case priority list with effort and risk in writing. |
| When to pick | You have not picked a product and you need an access matrix. | You already picked Hermes Agent. | The workflow you want to support is not clear yet. |
// evidence and sources
Evidence we checked.
The MCP specification uses date-formatted versions (YYYY-MM-DD). The current version is 2026-07-28. Clients and servers may support several versions at once, so "supports MCP" alone does not mean 2 tools work together.
Model Context Protocol: VersioningRead 12 September 2026The OWASP Top 10 for LLM Applications 2025 lists Prompt Injection at LLM01 and Excessive Agency at LLM06. An agent with too much access is a recognised risk, so access limits and human approval are set on day 1.
OWASP GenAI Security Project: LLM Top 10Read 12 September 2026OpenClaw 2026.9.1 reached stable on 3 September 2026. The upgrade path rolls back when Doctor fails, and cross-session approvals are easier to trace.
Rama Digital: OpenClaw 2026.9.1 ReleaseRead 12 September 2026
- Lima Lapisan Engineering AI Agent: Prompt, Context, Harness, Loop, dan Graph (ID)
- Human Approval Matrix dan Governance Pilot Codex (ID)
- OpenClaw 2026.9.1 Rilis: Upgrade Bisa Rollback, Gateway Lebih Tahan Error, dan Approval Lebih Jelas (ID)
- Hermes Agent di Windows, Linux, macOS, dan VPS: Matriks Setup
Owner: Rama Digital team · Updated 12 September 2026
// definitions
Terms we use.
- AI agent
- An AI agent is an AI program that runs a work step on its own. For example, it reads a message, drafts a reply, then logs the result. See in the glossary
- MCP
- MCP is a shared standard that lets an AI program use your tools and data under clear permissions. See in the glossary
- Agentic
- Agentic work happens when an AI agent runs a routine task, then staff or the owner check and approve the result. See in the glossary
- Hardening
- Hardening tightens a server and its setup: it stores each secret in a safe place, limits access to approved people, and turns off unused services.
FAQ
What is AI Agent Ops Install and how much does it cost?
AI Agent Ops Install sets up an AI agent on your own server or cloud account for 1 workflow. We pick the product, write the access matrix, turn on the audit log, and test the workflow with human approval. Price runs IDR 7.5–25M based on scope. Installation takes 1–2 weeks once server access is ready.
How is this different from Hermes Agent Installation Service?
Hermes Agent Installation Service installs 1 fixed product, Hermes, at a fixed IDR 6M for 2–5 working days. AI Agent Ops Install starts with a product pick among OpenClaw, Hermes, or QwenPaw based on your workflow. This service then adds a role-based access matrix, MCP, and A2A. Pick Hermes when the product is set; pick this service when you still compare options or need finer access control.
Which tools does the service support?
We install OpenClaw, Hermes, or QwenPaw based on your workflow, channel, and tool needs. The pick happens on Day 1–2, before install, based on real need, not a trend. Once the product is set, we connect your tools through MCP and set up an A2A baseline for agent-to-agent messages. We then list every active tool in the handover document.
Does the agent get unrestricted access?
No. Each agent gets a role-based data limit and a list of actions that need human approval before they run. We write this matrix on Day 1–2, and turn on the audit log from day 1. We then test the first workflow with an approval gate on each action that carries impact. The operator and the owner can read the audit log at any time.
What moves the price from IDR 7.5M to IDR 25M?
Price follows scope: the number of workflows installed, and the channels and tools connected through MCP. Price also follows the number of users and roles in the access matrix, and how deep the SOP and workflow test go. A small scope with 1 workflow and few tools sits near IDR 7.5M. A wider scope moves toward IDR 25M.
Are server and AI model fees included?
No. Server or cloud fees, AI model usage fees through the API, and phone numbers or channel accounts stay separate from the IDR 7.5–25M price. Our price covers the product pick, install, hardening, access matrix, audit log, MCP, A2A, a 1-workflow test, and the SOP. Set up your own model-provider account before Day 1.
How long does the installation take?
Installation takes 1–2 weeks once server and model-provider access are ready. Day 1–2 picks the product and writes the access matrix. Day 3–5 installs and hardens the agent. Day 6–8 connects tools and channels. Day 8–10 tests 1 workflow with human approval. Day 10–14 writes the SOP and runs a 60-minute handover.
What happens if the agent acts wrong?
Any action that carries impact passes an approval gate before it runs, so an error gets caught before it plays out in full. Every action lands in an audit log the operator can read, so the cause is easy to trace. We do not guarantee an error-free agent. That is why the access matrix, the approval gate, and the log go in from day 1, not after a problem shows up.
Who takes care of the agent after handover?
Handover gives your team the agent, the access-limit document, and the SOP. It includes a 60-minute session on how to assign a task and how to stop the agent. Ongoing care after that, such as model updates, monthly log review, and prompt tuning, runs on the OS Care Retainer as a separate service. You may also run this care yourself with the SOP we write.
I do not know which workflow fits. Where do I start?
Start with AI Diagnostic when you do not know which task to support; 1 session maps the opportunity and the first step. Pick AI Workflow Audit when you want a data-based priority list before you choose. Once 1 workflow is clear and you hold your own server, come back to AI Agent Ops Install for a setup with written access limits.
// next step
Start with 1 workflow, with written limits.
Send the 1 workflow you want an agent to support and your server type. We reply with a product pick, a price estimate in range, and an access list within 24 working hours. No commitment.
