Build agents worth texting.
Bring your own agent or start from a Plow-ready runtime. Plow handles hosting, chat, and connectors.
Looking for inspiration? Try one of the community's templates
Sam
Tiago Bello
Runs on Plow's agent infrastructure.
The tools that give you everything you need to create textable agents that stay secure and in your users' control. Reach for connectors, chat, and Mac access as your agent needs them.
Connectors. Create agents that run on Gmail, Google Calendar, and Slack. No developer account needed.
iMessage Chat API. Connect your agent through Plow's chat plugin. Receive messages over iMessage, send replies and scheduled updates, and support group threads.
Latch Password Manager. Let your agent act on a user's Mac, from the browser to files to commands, through per-request approvals and permission controls.
Deploy agents, fast and simple.
Go from a base image to a running agent that texts back in minutes. You write behavior, Plow runs it.
Start from a base image. Pick a preconfigured Hermes or OpenClaw base image and skip the scaffolding.
Base imagesWrite behavior in Markdown. Describe skills and persona in plain Markdown, dropping a SKILL.md in a folder, and the model routes on each skill's description.
Deploy with one command. plow-agents deploy puts your image on a Plow-hosted VM and gives it a line to answer on.
Develop locally. Edit a skill, run docker compose up, and watch it think before you push a new image.
Bring your own agent, or start Plow-ready.
Any container that follows the Plow image contract can run: build it for linux/amd64, push it to a registry anyone can pull from, and deploy it by digest. Or start from a Plow-ready runtime that has the Plow pieces wired in already.
from hermes import Agent, skill @skill("reservations") def book_table(when, party): return reserve(when, party) agent = Agent(skills=[book_table])
Hermes
A Python runtime rooted in Nous Research's Hermes work, at home with the models and libraries most AI tooling is built on.
View on GitHubimport { Agent, skill } from "openclaw"; const bookTable = skill("reservations", ({ when, party }) => reserve(when, party)); export const agent = new Agent({ skills: [bookTable] });
OpenClaw
A Node and TypeScript runtime for web and full-stack builders who live in the npm ecosystem.
Coming soonAct on a Mac,
with permission.
How Latch works
The vault is encrypted on the person's Mac, with the key protected by macOS Keychain. Plow keeps no cloud copy.
- macOS Keychain
- No cloud copy
- Encrypted relay
Latch enters an approved credential in the browser itself. The agent never sees it, and the log never records it.
- Never in agent context
- Never in the log
Allow it once, or always for this agent and these sites. Every request and every decision is recorded.
- Allow Once
- Always Allow
- Full activity record
Claude, Codex, OpenClaw, Hermes, and other agents that support MCP can act through Latch.
- MCP-compatible
- Open-source model
Publish to the community index.
Submit your agent to the AI Worth Using Agent Index so people can find it. Add the reporter-equipped image to publish install and usage to the leaderboard, so you see real adoption.