Skip to content
Velum

Custom AI agent development

Agents built around how your team already works.

We design, build and ship AI agents that do real work inside your tools: reading documents, updating systems, drafting replies and making routine decisions, with a person approving anything that matters. Tested on your real cases before launch.

  • 4 to 8 weeks per agent
4–8 wk
From scope to live
100%
Actions logged
0
Irreversible steps without approval

In short

Velum's custom AI agent development designs and builds production AI agents for a specific business workflow. Each engagement covers agent design, integrations with your systems, human approval steps, an evaluation suite built from your real cases, a supervised pilot and launch, typically in four to eight weeks.

A good fit for

  • Teams with a clear, repetitive workflow to hand off
  • Companies that need agents inside existing systems, not a new app
  • Enterprises that need approvals, audit and on-premise options

What you get

6 deliverables, all yours.

Everything below is handed over at the end, documented and owned by your team.

Scope this with us
  1. 01

    Agent design

    What the agent reads, decides and does, what it must never do, and where a person approves.

  2. 02

    Integrations

    Connections to your CRM, ERP, help desk, email, chat and documents through APIs, MCP servers or browser automation.

  3. 03

    Approval steps

    Review and approve in Slack, Teams or email, with the evidence attached to every request.

  4. 04

    Evaluation suite

    Hundreds of test cases drawn from your real work, run on every change so quality never quietly slips.

  5. 05

    Supervised pilot

    The agent runs alongside your team on live work, with every action reviewed until it earns trust.

  6. 06

    Handover

    Runbooks, dashboards and training so your team owns the agent after launch.

How it runs

From first call to done.

  1. Week 1

    Design

    Workflow mapped, agent behaviour and limits agreed, test cases collected.

  2. Weeks 2–4

    Build

    Agent, integrations and approval steps built and evaluated against your cases.

  3. Weeks 4–6

    Pilot

    Supervised run on live work with the team, tuning until acceptance criteria are met.

  4. Weeks 6–8

    Launch

    Released as the default path, with monitoring, documentation and training.

Timeline

4 to 8 weeks per agent

Pricing

Fixed price per build, quoted against an agreed scope and acceptance criteria.

Before you ask

01

Which models do you use?

Whichever fits the task and your data policy: Claude, GPT, Gemini or open models, often more than one. Model providers are configured with zero data retention.

02

Where does the agent run?

In our managed cloud, your cloud account or on-premise. The choice is made in the design week based on your data and security needs.

03

What happens if the agent gets something wrong?

Anything irreversible waits for approval, every action is logged and replayable, and the evaluation suite catches regressions before release.

Start with a 30-minute call.

We will tell you honestly whether this service fits, and what it would take.