WHAT PEOPLE WANT ep-02: Own Your AI, Don't Rent It
By Juyel Rana · November 7, 2026 · Updated November 7, 2026

Somebody on Hacker News put it bluntly last month: they were tired of paying rent on their own assistant. It wasn't a complaint about one product. It was about the whole arrangement — your agent runs on somebody else's servers, your secrets sit in somebody else's vault, and the bill ticks up by the hour whether the agent is doing anything useful or not. October was the month a lot of people decided they'd rather own the thing.
The rent keeps going up
Anthropic's Claude Managed Agents launched in April 2026 at $0.08 an hour per running agent. OpenAI's Agents API went into public beta in September 2026. For a quick demo, either is fine. But an agent that stays on all day, every day, quietly turns into a line item — and the price isn't even the part that bothers people most.
What bothers them is the custody. The agent, its memory, and the keys it uses all live somewhere you can't see, under terms you don't control. The vendor can change the price, change the rules, or shut the service down, and your setup goes with it. People who build with these tools every day have watched that movie before with SaaS, and they're not keen on a repeat with the software that holds their passwords.
What "own it" actually means
When people in these threads say they want to own their agent, they usually mean three specific things.
First, secrets stay home. The API keys and credentials the agent uses should live on hardware you control, not in a vendor's secrets store.
Second, the model is swappable. If one provider gets expensive or drops in quality, you change a line in a config file — you don't rebuild the agent.
Third, the memory belongs to you. One of the newer tools, AnyModelBot, keeps the agent's long-term memory as plain Markdown files on your own disk. You can back it up, move it, or delete it. Nothing about your setup depends on staying with one vendor.
That's the demand in one sentence: the convenience of a managed agent, with the exit rights of software you installed yourself.
October's answer arrived all at once
The interesting part is how fast the open-source side moved to meet it. In the space of about two weeks, the front page of Hacker News and the tech press filled up with self-hosted alternatives to the managed agent services.
Omnara launched on October 21, 2026 — an open-source agent platform under Apache 2.0 that you can run yourself with Docker Compose. It handles the boring infrastructure agents need: tool access, secrets management, permissions, sandboxed execution, and state stored in Postgres so idle agents cost nothing and survive restarts. It hit the Hacker News front page the same day.
Brain v2 showed up the same day, from AEXHQ: a minimal MIT-licensed agent server built around persistent sessions, aimed at developers who want to run agents on their own infrastructure and pick their own models.
The personal-agent side was already big. A mid-October roundup that pulled the numbers straight from GitHub put OpenClaw, an MIT-licensed personal agent that lives on your own machine and works across chat apps, at roughly 392,000 stars. AnythingLLM, another MIT project, pairs document workspaces with agents for people whose "agent" is mostly chat over their own files. And on November 1, AnyModelBot arrived as a local-first assistant that stores its memory in Markdown on disk and lets you switch between 70-plus models without leaving your history behind.
Even the infrastructure crowd joined in. Oxide Computer published references on October 29 for running agents fully on-premises, one ephemeral virtual machine per session, deleted when the agent finishes. The message from all of them is the same: you don't have to rent anymore, and you don't have to be a large company to self-host.
The honest trade-off
Here's the part the launch posts skip: owning the agent means owning the chores too. Uptime, updates, security patches — that's all yours now. An agent holding your API keys and file access, running on a box you forgot to patch, is worse than a rented one. Self-hosting isn't free; it just moves the cost from a monthly bill to your time.
So don't migrate everything at once. The people doing this well start with one small job, give the agent read-only access, and watch it for a week before they hand it anything sharp. And managed services still make sense for throwaway tasks with nothing sensitive in them. Own the agents that touch your data. Rent the ones that don't matter.
If you want to try it this week
Pick one repetitive job first — watching a supplier page for a price change, triaging an inbox, drafting replies from your own documents. Then match the tool to the job: OpenClaw if you live in chat apps, AnythingLLM if the work is mostly your files, Omnara or Brain if you're building agents for a team. Run it on hardware you already have or a cheap VPS, keep your secrets in the tool's own store instead of pasting them into prompts, and start read-only. A week of good behaviour earns wider permissions. That's the whole test, and it costs you an afternoon.
People don't want a better landlord for their AI. They want the keys. October proved the keys are finally on the table — now it's just a question of which lock you put them in.
Keep reading on heyjuyel.com
- IAB Agentic AI Is Now the Core Ad-Buying Priority
- Pepper Launches Agent Atlas: AI That Executes GEO Work, Not Just Reports
- Connector.wtf: Free MCP Server for Google Ads, Meta and LinkedIn into AI Chat
- Kando Marketing Released 42 Free Browser Tools, No Signup
- WHAT PEOPLE WANT ep-01: Leads, Not Rankings
Sources
- Omnara Ships Self-Hostable Alternative to Claude Managed Agents and OpenAI Agents API
- Omnara Launches Open-Source Alternative to Claude Agents
- Brain V2 Launches as Minimal Open-Source Alternative to Managed Agent Services
- Best Personal AI Agents You Can Self-Host in 2026
- Can You Run AI Agents Fully On-Premises Without Handing Your Data To A Cloud Vendor?
- AnyModelBot: an open-source, local-first personal AI assistant that keeps its memory in Markdown on your disk
- Show HN: Open Responses – Drop-In OpenAI Responses API Alternative for Any LLM
