Moving from Verboo to NotHoly: what changes, what stays

If your code already talks to Verboo, Venice AI, Drael or another OpenAI-compatible service, moving it to NotHoly is mostly a base URL and a key. The rest of this guide covers the parts around those two values: model IDs, output settings, billing, licence terms and any tools that ran on the other side.

Data collected on 11 November 2026 from Verboo's API page, documentation and Terms of Use, and from NotHoly pricing. The Venice AI and Drael figures come from our comparisons, Venice AI vs NotHoly and Drael vs NotHoly, collected on 9 and 10 November 2026. Prices and terms change, so check the sources before you switch.

#What carries over

Verboo's API is OpenAI-compatible, and so is NotHoly's. An integration written against the OpenAI SDK keeps its request shape, streaming and tool calling, and changes two values. NotHoly also accepts the Anthropic format, so a client built for that format has a target too. Drael's API is OpenAI-compatible as well, but only on its paid four-week or annual plans.

#Change the base URL and key

Point the client at api.notholy.com/v1 and swap the key. With the OpenAI Python SDK, the whole change looks like this:

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.notholy.com/v1",
    api_key=os.environ["NOTHOLY_API_KEY"],
)

response = client.chat.completions.create(
    model="<model ID from the models page>",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Summarize this function."}],
)
print(response.choices[0].message.content)

Create the key in the console and keep it in an environment variable, outside the repository. Keys carry per-key rate limits and show up in the audit log, so give each tool or CI job its own key. Nothing else in the call changes for an OpenAI client.

#Pick model IDs

Model IDs change with the provider, because each catalogue is different. NotHoly's models are for offensive security work, from reverse engineering and pentest to malware analysis and OSINT. They are abliterated, which means the refusal direction was removed from the weights and the rest was kept. They can still err or refuse. Choose your replacement on models by base and context window, and match the base you need rather than the display name.

Venice AI and NotHoly share some base models, GLM 5.3 and DeepSeek V4.1 Flash among them. Compare the base before you compare the price, then check each rate on pricing on the day you buy.

#Check the output settings

Services differ in how they treat output limits and reasoning, so read your client's settings before you assume they carry over. Drael's public documentation says its API discards max_tokens on keyed calls, forces reasoning on, and bills the output with reasoning included. On NotHoly you send your own max_tokens and reasoning effort, and the gateway passes them through. It trims max_tokens only when your prepaid balance cannot cover the worst case. Set both values explicitly, then compare a week of real usage with what you expected.

#What changes in billing

Verboo's paid monthly plans, from US$ 20 to US$ 269, become prepaid credit. You top up US$ 5, 10 or 50, or 500 (5% off) or 1,250 (10% off), by card or crypto, and pay per token. The credit does not expire, nothing renews, and an idle month costs nothing. NotHoly takes no Pix. The rate for each model is on pricing.

Venice's API is metered on top of a monthly plan, and its monthly credit expires. Pro keeps none, and the larger plans keep unused credit for two or three months. Drael's API needs a four-week block of at least US$ 70, and the API stops when the block lapses. NotHoly starts at the US$ 5 top-up.

#What changes in the licence

Verboo's individual plans are for personal use, and its terms (clause 10) say that serving third-party users, clients or products may need a pay-per-use or enterprise plan. Section 7.1 of Venice's terms grants a licence for personal, non-commercial use, as Venice AI vs NotHoly explains. NotHoly's terms allow commercial use and name the holder by CNPJ, with governing law and venue. If your project serves clients or other users, check the licence of every service in the chain. This is a summary of the terms, not legal advice.

#What moves to another place

Some tools ran on the other side of the API. Drael's run_code and web_fetch, for example, run on its server. NotHoly runs no code on its server, and no server tool sends active traffic to a third-party target. Its page reader goes through an egress protected against SSRF. The CLI runs commands on your own machine, from your own address, and asks for approval first. If your workflow relied on a hosted agent that executed code, run that step locally with the CLI, or keep your own agent loop and send its model calls to NotHoly.

#What you give up

NotHoly does not execute your code on its servers: commands run in the CLI on your machine, with your approval for each one. It has no image generation, no video or music, and no Pix, and it holds no SOC 2 or ISO 27001 certification. If a hosted agent that runs code for you was the reason you used another service, weigh that before you move.

#Frequently asked questions

#Is NotHoly a drop-in for the OpenAI SDK?

For the request, yes. Change the base URL and key and keep the OpenAI client. You still pick a NotHoly model ID.

#Do I need to change my model IDs?

Yes. Model IDs belong to each provider, and NotHoly's catalogue is not Verboo's. Choose by base and context window on models.

#Can I keep Anthropic-style clients?

Yes. NotHoly accepts both the OpenAI and the Anthropic formats at api.notholy.com/v1. Verboo's documented API is OpenAI-compatible.

#Does my remaining credit transfer?

No. Balances do not move between providers. Cancel the old plan from its own dashboard before the paid period ends, and top up NotHoly on pricing.

#Verdict

For an OpenAI-compatible integration the move is small: a new base URL, a new key, a new model ID, and explicit output settings. The plan becomes prepaid credit that does not expire, and NotHoly's terms allow commercial use. Read the catalogue on models, add US$ 5 on pricing, and run your existing requests against both before you switch.