agents are markdown

Agents are just markdown.

Define an AI agent as plain .md files in a git repo - its role, its skills, its memory, its schedule. Push, and it runs. No code, no SDK, no YAML pipelines.

$ npx foldrun init my-agent

$ foldrun run my-agent

$ foldrun deploy# and it keeps running without you

Nothing above needs an account —start on your own machine.

---
name: competitor-watcher
description: Watches competitor sites, drafts a weekly digest.
schedule: "0 8 * * MON"
model: default
tools:
  - web
  - files
secrets:
  - SLACK_WEBHOOK_TOKEN
---

You watch our competitors and produce a weekly digest.

Check each site listed in memory/competitors.md. Note pricing
changes, new features, and new blog posts. Write the digest
to outputs/digest.md and post a summary to Slack.

No code. No SDK.

An agent is a folder of markdown files: agent.md for who it is, skills/ for what it can do, memory/ for what it learns. If you can write a README, you can build an agent.

Git-native by design

Behavior is reviewed in pull requests. Learning is committed to memory/. Rollback is git revert. Your agent has the same audit trail as your code.

Runs anywhere

The open-source CLI runs your agent locally against your own API key. The format is an open spec - no lock-in, ever.

Deploys like Vercel

git push and it is live: scheduled runs, vaulted secrets, logs, costs, and one-click rollback. You write markdown; we run the infrastructure.

Push the folder. It runs without you.

Schedules, webhooks, approvals, vaulted secrets, a full audit trail of every run and what it cost — for agents that are still just markdown in your git repo.