Skip to content

AI Agent Setup · for small & medium business

Start with one workflow. Let an agent do the legwork.

We set up a secure AI agent on one repetitive workflow, such as lead follow-up, client handoffs, weekly reporting, admin prep or inbox triage. It works in the tools you already use. The agent prepares the work; your team approves anything sensitive.

Request a workflow review See what a first agent does

Scoped around your tools, access and approval rules, then expanded once the first workflow earns its place.

Prepared for review · Monday 8:30

Weekly client handoffs

Checked

  • Project tracker
  • Shared inbox
  • CRM notes

Found and prepared

  1. Overdue
    Client update promised for FridayDraft note written from the latest tasks
  2. Missing
    No owner on a new onboarding taskSuggested owner from last month's handoff
  3. At risk
    Renewal date close, no follow-up bookedFlagged for your decision, nothing sent

Waiting on you

1 client note to approve and 1 renewal to decide before anything leaves the business.

Approved actions are logged. Exceptions become follow-up tasks.

Sample review packet: weekly client handoffs

Where to start

Pick the work your team repeats every week

A good first agent sits beside your existing tools, prepares the next action and waits for a person where it matters. These are common starting points. Your own workflow can be scoped the same way.

  1. Lead triage

    You'll recognise it
    New enquiries sit unanswered while someone works out who should reply.
    The agent prepares
    A priority, a suggested reply and a CRM update, for example in Zoho or HubSpot, and alerts the owner.
    You approve
    The reply and any record change.

  2. Client handoffs

    You'll recognise it
    Tasks go overdue and client updates slip between people.
    The agent prepares
    A list of overdue tasks and missing updates, with the next client note drafted.
    You approve
    Every client-facing message.

  3. Weekly reporting

    You'll recognise it
    Someone spends hours stitching updates together from several tools.
    The agent prepares
    A delivery pack with risks and decisions, pulled from Jira, Teams, documents or dashboards.
    You approve
    What gets shared and with whom.

  4. Invoice and admin prep

    You'll recognise it
    Billing waits on someone finding what was actually completed.
    The agent prepares
    Billing detail from resolved work, ready to check.
    You approve
    Anything that touches money, before it is sent.

  5. Owner inbox digest

    You'll recognise it
    Urgent email and calendar items hide among everything else.
    The agent prepares
    A daily list of urgent items, overdue replies and next actions.
    You approve
    Which replies go and which items become tasks.

Want to see more patterns? Browse the workflow patterns

Your first release

From one named workflow to an agent your team runs

  1. Choose and scope

    Pick the workflow, agree what good looks like, and name the person who owns it. Check the tools have an accessible API, webhook, export or approved integration path.

  2. Connect with least privilege

    The agent gets its own identity and only the access this workflow needs. Approval rules are set for sends, record changes and anything touching money.

  3. Run alongside your team

    The agent prepares real work while your team reviews it. Exceptions are routed to a person, not guessed.

  4. Hand over, then expand

    Your team takes over the operating rhythm. The next workflow is chosen from what the first one shows.

Engagement paths

Three ways to start, depending on how clear the work is

Timing and budget below are typical ranges, not quotes. The real scope is confirmed after a short intake, once tools, access and the approval owner are clear.

  1. Clarify scope

    Discovery before build

    When the pain is real but the workflow, data, approvals or tools aren't clear enough to build safely yet.

    Includes

    • Workflow selection and readiness check
    • Integration, access, risk and security map
    • A recommended Starter or Professional path, or honest advice to wait

    You provide

    Process walkthroughs, a tool list, a ranking of pain points, and any data or access constraints.

    Typically1–2 weeks

    Budget signalunder $5k

    Clarify the first workflow

  2. Starter

    One secure workflow

    When you can already name the repetitive workflow and want a contained, approval-gated agent before expanding.

    Includes

    • First-workflow mapping and success criteria
    • One agent for lead triage, handoff checks, reporting, inbox digest or admin prep
    • Least-privilege access, approval rules and handover

    You provide

    Tool access, a workflow owner, sample inputs and a success measure.

    Typically2–4 weeks once inputs, access and approval owner are clear

    Budget signal$5k–$15k

    Ask about Starter

  3. Professional

    Agent plus tool ecosystem

    When several tools and recurring workflows connect, and approvals, reporting and escalation matter.

    Includes

    • An operating model adapted to your environment
    • Accessible APIs and approved integrations as scoped, for example Jira, Confluence, Microsoft 365, CRM, accounting, GitHub, Slack or Teams
    • Reporting loop, escalation rules, monitoring cadence and an optimisation backlog

    You provide

    System owners, a credentials path, representative workflow data and a named operating owner.

    Typically4–8+ weeks, staged

    Budget signal$15k+

    Scope Professional

Good fit

Worth a conversation if…

  • Your team repeats the same triage, reporting, drafting, follow-up or admin work every week.
  • The work lives in real systems with an accessible API, webhook, export, database or approved integration path. Permissions and feasibility are confirmed during scoping.
  • You want an agent that works under approval rules and scoped access, not an unsupervised chatbot.
  • Someone in the business can own the workflow and judge whether the agent is useful.

Not yet

Probably too early if…

  • There is no repeatable workflow or clear business outcome yet.
  • You need an agent making external commitments with no human review.
  • Your tools and data are locked away and nobody can approve access.
  • You're after general AI training rather than a working system.

Buyer questions

AI Agent Setup FAQ

Straight answers to what owners usually ask first.

Have a question that isn't here? Ask it in the enquiry form. A few sentences are enough. Ask in the enquiry

What tools can an AI agent connect to?

Most tools with an accessible API, webhook, export, database, inbox or approved integration path. Permissions, licences and technical feasibility are checked during scoping. Microsoft 365, Teams, Slack, Jira, Confluence, Zoho CRM, HubSpot, Xero, QuickBooks, GitHub and Google Workspace are familiar examples, not the limit.

How is the agent kept secure?

Security is part of the design, not added afterwards. Each agent gets its own identity and least-privilege access scoped to its workflow. Network paths, approval points, action logging and recovery steps are agreed for your environment. Sensitive actions, such as external sends, record changes and anything touching money, are set to wait for your approval.

How long does setup take?

A contained first workflow typically takes 2–4 weeks once the process owner, tools, approval rules and success measure are clear. Integration-heavy setups become a staged Professional path over 4–8+ weeks. Timing is confirmed after scoping, not promised up front.

How does pricing work?

There are three paths: Clarify scope, Starter and Professional. Effort is confirmed after a short intake, because tool access, workflow complexity and security requirements change the real scope.

Will the agent act without my approval?

Only where you have explicitly allowed it. Drafting, monitoring, summarising and preparing work can run automatically. External sends, CRM updates, spend and sensitive commitments wait for approval until you say go.

What happens when something goes wrong?

Exceptions are routed to a person rather than guessed. The agent flags what it could not complete, and monitoring and exception handling are part of the handover, so your team knows where failures go.

What are OpenClaw and Hermes?

OpenClaw is the agent platform Nebulas builds on. Hermes is the orchestration layer that routes work between agents and your tools. They are used to give each agent its own identity, scoped permissions and logged activity. You don't need to learn either to benefit from them.

What happens after I enquire?

You get a reply by email with a practical next-step recommendation. The form keeps your service context, so the response is specific, not generic.

Private intake

What would you like to connect or improve?

A few sentences are enough to start. Request a workflow review and we will reply by email with a practical next step. No booking is made by submitting this form.

What happens next:
  1. Nebulas reviews your context.
  2. We check fit and urgency.
  3. We reply by email with the most practical next step.

We use your details to review and reply to this enquiry, and keep a record of it. Privacy details

Call us directlyCall the team

Name, email, a short description and consent are all we need. Please do not include passwords or sensitive records.

Add details (optional)