Short answer: no. Here is the longer answer, and it is not a knock on Eve.

In June 2026, Vercel released Eve, an open-source framework where you define an agent as a folder of files. Your myPKA is also a folder. So the question writes itself, and Cole Medin's walkthrough (youtu.be/m8VC2SV2igM) put it in front of a lot of people at once. Members have been asking ever since.

The answer turns on one distinction. Once you see it, the rest falls into place.

Eve compiles a folder into an agent. Your myPKA folder is not an agent's definition. It is your knowledge base. It is the thing the agent reads and writes.

In Eve, the folder is the source code of the assistant. You write the instructions, the skills, the tools, and the subagents as files, then a build step compiles all of it into a deployed service. Once deployed, that folder is a finished artifact. The running agent does not keep editing it. Anything the agent needs to remember lives in a database-backed session store, not in the files.

Your myPKA is the opposite arrangement. The whole point is that your notes are writable plain markdown, sitting on your own machine, owned by you. Your journal entries, your project notes, your tasks: the assistant writes those, every day, and they are still there tomorrow.

Put your myPKA on Eve and the folder freezes at deploy time. The assistant could no longer write your journal. It could no longer add a note. It could no longer update a task. Everything that makes the folder yours would need rebuilding as a database that the agent talks to through a tool.

They share the word "folder" and almost nothing else.

This is not a hit piece, and being accurate here matters more than scoring points.

Eve is Apache-2.0 licensed, which is real open source with no strings attached. You can self-host it with no Vercel account at all, so there is no lock-in. That claim is proven, not just marketed: Vercel publishes a reference implementation running on a plain server with zero Vercel infrastructure.

The structure is elegant. Instructions become the system prompt. Skills load on demand instead of sitting in every prompt. Subagents handle focused subtasks with their own fresh context. Channels connect the agent to Slack, Discord, Teams, or plain HTTP. Connections wire in external services and MCP servers. Schedules run it on a timer. Evals test it. A compile step discovers all of it and wires it together automatically, so you never register anything by hand.

The engineering underneath is serious. Sessions are durable, meaning a conversation survives a crash and picks up where it stopped. Code runs in a sandbox, isolated from everything else. Risky steps can pause for a human to approve. Evals can gate a deploy, so a change that breaks behavior does not ship. And it scales to thousands of concurrent users.

Here is the honest frame. Eve is built to serve many people at once. Your myPKA serves one person: you. Neither of those is a flaw. They are answers to different questions.

Look at that list again: instructions as the system prompt, skills that load on demand, subagents as specialists, plain markdown and simple code sitting in a folder.

That is the same architecture your myPKA uses.

Vercel arrived at it independently. A major engineering company, running agents in production for real customers, reached the same conclusions about how to structure one. That is strong validation that the folder-as-agent idea is right.

It is also exactly why the confusion is so understandable. The structures genuinely rhyme. The difference is not the shape of the folder. It is what the folder is for.

1. Your Claude subscription will not work. Eve authenticates with an Anthropic API key. There is no subscription path, and this is not an oversight. Anthropic's own Claude Code legal documentation states it plainly: "Anthropic does not permit third-party developers to offer Claude.ai login or to route requests through Free, Pro, or Max plan credentials on behalf of their users."

In practice: today your subscription covers your myPKA sessions at a flat monthly rate. You know what you pay. On Eve, every message bills per token, and there is no ceiling.

2. It costs real money. Budget roughly $35 to $70 per month for one always-on agent. That splits into about $20 to $30 for Vercel plus $15 to $45 in metered API tokens. Most of the Vercel side is the Pro seat, and it is not optional: Vercel restricts the free Hobby tier to personal, non-commercial use, and all commercial usage requires a paid Pro or Enterprise plan. They read "commercial" broadly, so even a donate button counts. Compare that honestly with today, where your myPKA runs on the Claude subscription you already pay for.

3. It is a public preview. Eve launched on 17 June 2026 at Vercel Ship London. "Public preview" is Vercel's own term for it, which is their way of saying not yet final. They state directly that the framework, the APIs, the documentation, and the behavior may all change before general availability. Do not put something you depend on every day onto a preview release.

4. Your local tools stop working. A deployed agent runs on a server, not on your Mac or PC. Anything that touches your own machine or the apps installed on it cannot come along. That is not a bug in Eve. It is what deployment means.

Eve is the right tool if you want to build something your myPKA is not:

That is a real, worthwhile project. It is simply a different project from your myPKA.

Many people will correctly want both. Your myPKA as your private knowledge base, and an Eve agent as your team's shared assistant. This is an "and", not an "or". If you have been framing it as a migration, try framing it as a second, separate build instead. The decision gets a lot easier.

Three ideas from Eve are worth borrowing into how you think about your own folder, no migration required.

Evals. Test that your assistant still behaves correctly after you change its instructions. Most people never do this. They edit the instructions, the behavior quietly drifts, and they find out weeks later. Even a short list of "ask it this, it should do that" is more than most setups have.

Typed tools. When a tool's inputs are strictly defined, the assistant makes fewer mistakes. Vague instructions produce vague behavior. The more precisely you describe what something takes and what it returns, the less room there is to get it wrong.

Approval gates. Decide in advance which actions need a human yes. Not everything does. But the ones that touch money, send messages to other people, or delete things probably should, and deciding that ahead of time beats discovering it afterward.

If you want to deploy an agent for other people, look at Eve. If you want a private knowledge base you own, you already have the right thing. Keep your folder.

---

Source: Cole Medin's Eve walkthrough, youtu.be/m8VC2SV2igM, which raised this question for many members. Eve's own documentation and Vercel's published pricing and terms informed the cost and public-preview notes. Anthropic's Claude Code legal and compliance documentation is quoted directly on the subscription question.