Taskade BYOK: a new option on Pro

Imagine an inquiry arrives from a company you do not recognize. You want an agent to summarize what the company does and prepare useful context before you reply. Before choosing how to run that agent, separate access to the tool from the cost of its AI work.

BYOK means “bring your own key.” A key is a credential connecting your account to an AI provider. Taskade’s October 17 release notes describe personal OpenAI, Anthropic, and Google keys for custom agents and automation AI steps; a subsequent release adds xAI support. EVE app builds and edits do not use those personal keys. [1] [2]

The practical update is eligibility: Taskade’s pricing page now places BYOK on Pro and above, rather than the earlier Enterprise-only description. At the October 19 check, Pro displayed $10 per month with annual billing, 50,000 monthly AI credits, and 1,000 monthly automation runs. Credit packs do not add automation runs. [3]

What you are actually comparing

A personal key does not make every Taskade AI feature unlimited. It also does not mean your subscription includes the outside provider’s usage at no cost. Your comparison needs the Taskade subscription, any provider charges, and the time you spend correcting the output.

If you are considering Taskade, first check your account’s available providers and models, plus the current billing terms. Do that before buying a plan for a particular model or following a setup guide: this issue has not independently tested your account’s options.

Try a bounded lead-research comparison

Use a small set of fictional inquiries paired with real public company websites. Ask for the same output each time: a short company summary, the page supporting each important fact, and questions you should clarify before proposing a meeting. Avoid private customer information in the trial.

If your account supports both routes, compare the available Taskade-managed option with your personal-key option using those identical inputs. Note the relevant usage charges, whether the summary is usable, and how long corrections take. These are suggested tests, not measured Taskade results.

The decision is not “personal keys are cheaper.” It is whether that available provider, its charges, and its output fit your particular research task. A polished summary that invents company details is still a reply you cannot safely use.

Make: turn an inquiry into a researched reply draft

Once you know what a useful lead summary looks like, the next question is where it belongs in your day. Make’s official sales-outreach tutorial offers a concrete example that ends with something you can inspect: an email draft, not an automatically sent customer message. [4]

The example watches labeled Gmail leads, researches their companies, evaluates fit, checks calendar availability, creates an email draft, and sends the draft link to Slack. That is the tutorial’s described sequence—not a claim that we independently ran it or measured its performance. [4]

Adapt the reply to your business

If Make fits the tools you already use, read the official tutorial before connecting real inquiries. Begin with sample messages and confirm that the result stays a draft you review yourself.

Replace the example’s qualification criteria with plain business questions. For a local service company, that might mean whether the requested service is one you offer and whether the customer is in your service area. When a message leaves those details unclear, the useful reply asks—not guesses.

Check the meeting timezone, too. Before sending anything, verify the company facts and proposed availability yourself. A calendar suggestion can look reasonable while still being wrong for the customer or unsuitable for the work requested.

Look for a draft worth editing

For the sample inquiries, record minutes to a usable draft, correction time, factual errors, and whether meeting suggestions were relevant. Those are proposed pilot measures, not performance figures reported by Make.

You are looking for a useful starting point: a reply that reflects the inquiry and leaves uncertain details open for clarification. If you routinely rewrite the research or remove premature meeting offers, simplify the requested output before expanding the trial.

TSK-Bench: borrow the testing idea, not a ranking

A lead tracker can look finished before it works. A neat screen tells you little about whether entered records remain available or whether a small change breaks something you already need.

Taskade’s October 16 TSK-Bench announcement describes Build and Code arenas that test whether generated apps work, retain entered data, and survive a follow-up change. These are vendor-run tests inside Taskade Genesis—not independent proof of the best model or small-business ROI. [5]

Check the tracker like a user

For a proposed lead tracker, enter a few fictional records with different statuses. Reopen the app and check that the names, notes, and statuses remain correct. Then request one modest change, such as a filter for leads awaiting a reply.

After the change, repeat the original actions. Can you still add a lead, edit its details, and reopen its saved record? Does the new filter show the right entries without hiding information you need?

Treat that as your own practical test, not a result promised by the benchmark. It also reinforces the BYOK distinction: EVE builds and edits are outside personal-key usage. Choosing a key for lead research and checking a generated tracker are separate decisions. [1]

Quick Win

Before connecting a lead agent, write one fictional inquiry that deliberately leaves out the budget, location, and preferred meeting time. Ask for a reply draft using only that message and a public company page you provide.

Read the result for assumptions. Did it invent a budget, claim you serve the location, or suggest a time the customer never requested? Mark every unsupported statement and ask for a revised draft that turns missing details into questions.

This is a suggested exercise, not a product benchmark. It gives you a concrete example of what “ready for my review” should mean before you try the larger tutorial.