From API to a production-ready agent.

You probably already have the OpenAPI specs in-house, so why spend time rebuilding all your APIs in a way that works for agents? Just paste them into Agentify, connect MCP servers as you wish, and we'll do the rest! :)

Get started

Connect

Give your agent access to the tools you already have.

Upload an OpenAPI spec, get a tool per operation

After uploading your spec, we turn every endpoint into a tool your agent can call. You still keep control over which endpoints you enable.

Pick exactly which operations the agent may call
API keys, bearer tokens or OAuth sign-in per tool
Ready-made catalogue services like GitHub, HubSpot, Notion and Linear

Connect MCP servers in just a few clicks

Paste the URL of a remote MCP server and we'll handle the rest. If it needs a sign-in, we register Agentify with the server and run the OAuth flow for you, so there are no client IDs to copy around. Servers that use an API key just need a header.

OAuth is detected automatically and tokens refresh in the background
A manual client ID as a fallback when a server doesn't support registration
Tools are marked read or write, so you decide what's safe to enable

linearSigned in with OAuth

https://mcp.linear.app/mcp

linear_list_issuesREAD
linear_create_issueWRITE

internal-crmx-api-key: ••••••••

https://crm.acme.internal/mcp

internal-crm_find_customerREAD
internal-crm_update_contactWRITE

Build

Design the behaviour, not the prompt.

A visual flow builder for how your agent works

The visual flow builder lets you build your prompt through a diagram, which makes it easier to follow along as your prompt grows. It also forces you to think more about the structure, resulting in better overall prompts. Manual edits to a prompt are still possible if needed.

Steps, branches, actions, classifier and model-switch nodes
Context notes beside the flow for the details that matter
Prompt issues reported on every compile

Agentify's flow builder showing a lead-generation agent: a start node and steps wired to Browserbase tools, with context notes alongside.

Otto, the consultant that knows your agent

Otto is our built-in assistant and sits next to every page in the dashboard. Ask it how to structure a flow, whether your setup will hold up at scale, or why last night's runs failed. It can see your agent's tools, runs and usage, so its answers are about your agent, not agents in general. And if you like its suggestion, it can make the change for you.

Builds and edits nodes on the canvas, and fixes the issues it finds
Creates tools from a spec, or scrapes API docs into one if you don't have a spec
Nothing changes until you approve it

I'd put a decision in front of it. A classifier can check eligibility with a fixed answer shape, so the model never decides a refund in free text. Want me to add it before the payments step?

Proposed change

+ classifier node: refund-eligibility

ApproveDecline

Classifiers for decisions that must be reliable

A classifier can be used to keep a decision out of your prompt. This saves you some tokens but, most importantly, increases accuracy: the classifier only has one small task to do, and that task is taken off the main agent. The main agent just needs to know the outcome.

Use them as a node in your flow, or call them from your own code
Add instructions or pin a specific model per classifier
Edits apply instantly and identical requests are cached

Output shape

intent enumcomplaint | question | spam
order_number stringoptional
urgent booleanrequired

Test input

“Order #48213 arrived broken and I need it replaced before Friday.”

{
  "intent": "complaint",
  "order_number": "48213",
  "urgent": true
}

Test in the playground, then keep an eye on the numbers

Chat with your agent before your users do. For every turn you can see which model it was routed to, which tools it called, how long each step took and how many tokens it used.

A timeline of routing, text responses and tool calls per turn
Token usage per end user, plus a daily trend
The playground uses your draft, while your users keep talking to the live version

routinggpt-6-luna120 ms
Text response1 284 tokens610 ms
orders_get_ordertool340 ms
Text response412 tokens480 ms

Tokens, last 12 days1.2M

Compliance / Quality Assurance

Automated tests, Support for the EU AI Act and GDPR, built in.

Evaluations that gate every release

Write test cases once, from scratch or straight from a real conversation. When you publish a new prompt, every required suite runs against the draft, and it only goes live when the gate is met. For high-risk systems under the EU AI Act, that is the testing and release record your quality management needs.

Rule checks plus a reviewer model for the judgement calls
Every release recorded, including who published anyway and why
Run suites from CI with the SDK and fail the build on a miss

Live v13/Draft v14Check and publish

Refund policyRequired19/20
Order lookupsRequired12/12
Tone of voiceAdvisory7/9

Every required suite met its gate. v14 is live.

PII redaction with our own recognition model

Personal data is swapped for typed placeholders before anything reaches the language model, and put back in the answer your user sees. It runs on a PII model we fine-tuned ourselves, inside Agentify, so your model provider never receives the personal data, which supports GDPR's data minimisation.

One switch per agent, or the redact API for your own LLM calls
Names, emails, phone numbers, addresses, IBANs and ID numbers
Fails closed: no model, no unredacted text sent

Automated detection catches most personal data, not all of it. Agentify does not guarantee that every instance is redacted, so treat redaction as one safeguard among your other measures.

What your user wrote

Email Sarah Jenkins at sarah@northwind.io or call +44 20 7946 0958 about the refund to GB33 BUKB 2020 1555 5555 55.

What the model sees

Email [PERSON_1] at [EMAIL_1] or call [PHONE_1] about the refund to [IBAN_1].

4 entities in 38 ms, restored in the reply

Run

Plug it into your stack, wherever it lives.

Webhooks for everything your agent does

Pick the events you care about, add your endpoint and we'll send a signed JSON POST every time one happens. Think new conversations and messages, tool calls, routing decisions and workflow runs.

Signed with HMAC SHA-256, so you can verify it came from us
Every delivery is timestamped, so you never depend on the order they arrive in
Eval results are sent too, so your own tooling knows when a run passed

14:02:11.204chat.conversation.created200
14:02:11.390routing.finished200
14:02:12.018tool.invocation.completed200
14:02:12.733chat.message.streaming.complete200
14:03:40.511workflow.run.completed200

x-agentify-signature: 9f2c4e…b71a

Hosted in the EU, or on your own infrastructure

You can use our hosted version, which runs in the EU. If your data can't leave your network, you can run the whole platform yourself with a licence key and Docker Compose, using your own models and secrets provider.

The same product, licensed for your deployment
Conversations, run history and audit logs stay in your own database
Guides for running it locally and on Azure

Email us to set up self-hosting

$ export AGENTIFY_LICENSE=AG1.eyJ2IjoxLCJvcmci...$ docker compose up -d[ok] license verified[ok] agentify-api ready on :3002[ok] agentify-web ready on :3000

Estimate your plan

Every agent includes classifiers, model routing, webhooks, and custom model providers. Add seats, self-hosting, or Enterprise as you need them — startups may qualify for 25% off.

Agents

Seats

First 5 seats are included with your first agent.

Excludes LLM token usage, which is billed separately by your model provider based on actual usage.

Point Agentify at your spec. Watch it become an agent.

From spec to a working, testable agent — no manual schema writing.

Get started