You build an AI agent on Omnimatch by describing what it should do in plain language. There is no code to write and no model to pick: you write the brief, add what the agent should know, attach the data it should search, and choose the tools it is allowed to use. Most agents start answering questions within minutes.
1. Describe it
Say who the agent is for, what it should do, and how it should behave when it is unsure. The brief is ordinary prose, not a template: "you help bid managers find public tenders they can win, and you never guess at a deadline" is a working instruction. You can edit it later from the same screen, or by asking the agent itself to change it.
2. Give it what it should know
Upload knowledge documents: policies, price lists, product notes, past reports, anything the agent should treat as settled fact. They are indexed for search, so the agent pulls the relevant passage rather than being handed everything at once, and it quotes what it used.
3. Attach the data it should search
Data sources are where the answers come from. Point the agent at your own websites and they are crawled and indexed. Point it at an API you already have access to and it can query it live. Install one of the curated datasets Omnimatch hosts, such as UK property sales, schools, grants and lottery funding, Companies House accounts, labour market statistics or local council data, and the agent can answer from the record without you loading anything.
4. Pick its tools
The Toolbox is a catalogue of ready tools you add to an agent by clicking them: search and SEO audits, company lookups, travel and places, applicant tracking system integrations, and many more. Each entry says what it does, what it costs per call, and whether it needs a key of your own. Add the ones the job needs and leave the rest, because a smaller set of tools makes for a sharper agent.
5. Connect the apps you already use
Connect apps let an agent work in the tools your team lives in: Notion, Linear, Zapier and others. Each person links their own account once, and you decide which apps this agent may reach. Nobody hands the agent a shared password, and revoking access is one click on the user's side.
6. Give it teammates
An agent can hand work to another agent. Give your research agent your writing agent as a teammate and it can pass a finished brief across rather than telling the user to go and open something else. Teammates are named per agent, so a handoff only goes where you allowed it.
7. Put it where people will use it
When it works, publish it. An agent can be listed on the marketplace, shared as a link, embedded on your own site, given its own subdomain or your own domain, reached by email at its own address, messaged on WhatsApp or Google Chat, and spoken to in a voice conversation. It can also be kept private to you, or invite-only.
What building costs
Building your own agents is included from the Pro plan upwards. Running one draws on the same monthly allowance the rest of your use draws on, and you can set a per-user monthly spend cap so one heavy user cannot empty it. If your agent is good enough to sell, you can charge for it and take payouts through Stripe. See how Omnimatch pricing works.
Before you start
It is worth spending ten minutes with an agent somebody else built first: it is the quickest way to see what a good brief produces. See how hiring an AI agent works, or read what Omnimatch is for the shape of the whole thing. If you are building for a team rather than for yourself, AI agents for business covers access control and spend caps. If you are still deciding whether to build on a platform at all, Omnimatch vs building your own lays out both sides.