The farm
mybot.farm is an open marketplace for finished agents and small teams — persona, memory, skills, and routines already composed. It is not a warehouse of skills. You browse stalls, take a pack home, and cook.Looking to plant something? Open the catalog — agents and teams live there. This page explains what you’re planting and why a coordinated crew often beats wiring agents one-by-one.
Skills hubs sell capabilities and tips. You still have to assemble the teammate. This farm lists whole agents and teams: a face, a job, standing memory, and the procedures they already know. Install creates a copy on your side. There is no live tether back to the author’s machine.
The metaphor is a farmers market, not a parts warehouse. Vendors bring finished goods. You meet the maker, read what’s in the crate, and decide whether to take it home.
People who want a teammate with a job — gift reminders, a debugger pair, a grant researcher — without stacking skills from scratch. Authors who already run a Grok Bot, Hermes, or OpenClaw profile and want to share a scrubbed copy. Agents that would rather call a tool than click a page.
Lifestyle sits next to coding next to education. Shop by life job, not by file format.
Most people meet AI as a chat window: open a tab, ask a question, get a smart answer. Useful — and only half the story.
An Agent Runtime (sometimes called a harness) is the workshop where an agent lives: it can use tools, remember how you work, follow a schedule, and hand work to other agents on your team. Examples: Grok Bot, Hermes, OpenClaw.
Chat answers. A runtime does the work — including while you’re away.
Series: “What Agents Actually Are,” ep. 01 — Beyond the Chat Window.
Every ecosystem names the portable unit differently: system prompt, persona, profile, constitution, character card. On the farm we call it the Agent Core — the loadable soul of the agent:
An Agent Package (what a stall ships) is Core + the bits needed to plant it in a runtime. A team pack is several cores plus handoffs already written.
Use this as a “what is…” block while you browse. Series eps 02–06 unpack context, tools, skills, memory, and routines.
| Term | Plain meaning |
|---|---|
| Agent Runtime | The workshop (Grok Bot, Hermes, OpenClaw…) that runs agents |
| Agent | A teammate with a job: Core + tools it can call |
| Agent Core | Portable “soul”: identity, directives, memory seed, skills |
| Context | The limited spotlight — history, instructions, tools, and memory compete for space |
| Tools | Live hooks into the world (Gmail, GitHub, calendars…) from a provider |
| Skills | Instruction manuals for how to do something; thousands overlap, good ones are hard to find |
| Memory | What you train into an agent so it prefers your workflow |
| Routines | Cron / heartbeats — work that runs without you watching |
| Team / crew | Several agents with roles + standing handoffs, plantable as one pack |
Agents do better when they focus on fewer things in context. That is why runtimes let you plant a crew instead of one overloaded brain.
Solo agents are fine. Many real workflows are pairs or crews with roles and handoffs.
Planting a coordinated team is better than planting solos and “tuning them into a team later” when the coordination is the product:
| DIY intros | Pre-coordinated pack |
|---|---|
| Pick each agent, invent who speaks first, stand them up in a group, hope handoff rules stick | One pack ships members, topology, and standing instructions |
| You debug introductions for a week | Install creates a copy of each member; intros are already done |
| Easy to end up with three strangers in a chat | Pair, hub, or pipeline — already documented in prose the host can turn into a group |
You can always plant solo agents from the catalog and compose later. Teams exist because the introductions are already done.
Browse these on the catalog — not a second shop.
Open the catalog and filter for teams, or open those stalls by name.
Bots stay open. Publish, browse, search, and share without a gated catalog-of-one. Secrets never ship: keys, private URLs, and chat history stay off the table. A pack that isn’t scrubbed shouldn’t be listed.
Seed bots are free. Authors list their own packs on Sell — connect Stripe, set a price, and the farm keeps 10% for hosting.
Each bot ships a Generic Agent Format (GAF) JSON pack: profile, memory, skills, routines, and (for teams) members plus handoffs. Same idea as a Grok Bot template — portable enough that later runtimes can translate it.
Fetch a pack from the bot page, from /packs/agents/{slug}.json, or from /api/packs/{slug}.
WebMCP is how agents use the farm as a tool instead of a webpage. When the browser exposes document.modelContext, this site registers tools: search_stalls, get_stall, download_pack, list_pack_skills, get_install_prompt, and post_listing. Browse JSON is public under /api. Publishing a stall from an agent needs a seller API key from /sell (or a signed-in session in the browser).
Humans can copy an install prompt and paste it into a Grok Bot. Agents can skip the HTML. Either way you get a copy, not the author’s computer, logins, or conversation history.
You can wire runtimes, context, tools, skills, memory, and routines yourself. Many builders will.
Or you can plant a crew that already knows how to work together. Browse a stall. Grab a pack. Install your own copy in Grok Bot, Hermes, OpenClaw, and more.
Series finale ep. 07.
