Runnable examples for the Agent37 Agents API: example apps, custom agent images, and full applications to fork. Each folder is self-contained: put your API key in .env, run it, and you have a working app.
- Mint an API key at agent37.com/dashboard/cloud/api-keys.
- Add at least $10 to your wallet at agent37.com/dashboard/cloud/billing. The smallest instance costs $4.76 per month, metered per minute, and the rest covers managed LLM usage.
Small, single-purpose apps that each teach one part of the API.
| Example | What it shows |
|---|---|
| hermes-chat | A streaming chat app on a Hermes agent instance: create and manage instances, stream replies token by token, sessions, cancel. |
| site-builder | A website builder: the user asks the agent for a site, the agent builds and serves it on the instance, and your server publishes it at a permanent public URL. The owner-consent pattern for public ports. |
| app-builder | An app builder with a database per app: the agent builds a web app, gives it its own Postgres with Claimable Neon (no Neon account or token), and your server publishes it. The user claims the database into their own Neon account to keep it. |
| fork-agent | Fork an agent together with its data: every fork gets its own Neon branch of the agent's database, so you can run a task several ways in parallel, compare the branches with one query, and keep the best attempt. |
| dots | Build your own Dots: every user gets a named agent with its own computer that owns ongoing jobs, keeps its own schedule, messages them first, and (optionally) lets them watch and take over its screen. |
| instinct | Build your own Instinct: a texting assistant per user with its own iMessage line, email address, and calls through Inkbox, reminders that text first, and a web workspace. |
| grok-bot | Build your own Grok Bot: a team of named Bots that share one computer per user, with parallel chats, per-Bot routines, a team chat with @mentions, and an optional live screen. |
| muse | Build your own Muse: a personal agent with daily ideas, goals it checks in on, a library of what it made, and an editable memory. |
| claude-code-byo-model | Claude Code on your own OpenRouter key: override ANTHROPIC_BASE_URL in instance env so model calls bill your OpenRouter account, with per-role model picks. |
| android-cloud-phone | An Android phone in the cloud: create a VM instance with the emulator already on it, drive it with ADB, and watch the screen from a browser tab. |
| claude-managed-agents | Claude Managed Agents with one instance per session: the agent's files persist between turns, the instance sleeps while you are away, and sessions run in parallel. |
| hermes-browser-use | Browser Use Cloud browsers for every user's Hermes agent without your Browser Use key in their sandbox: a small broker hands each agent a per-user link to its own browser. |
Dockerfiles that change what the agent itself is. Each folder builds into a workspace template with one command, npx agent37 templates build . --name <name>, and no local Docker needed.
| Image | What it shows |
|---|---|
| hermes-vnc-desktop | A live desktop view on the stock Hermes image: watch the agent open Chromium, click, and fill forms in real time, and take over when it needs you. Logins persist, and it embeds in your own app. |
Complete applications on the API, each in its own repo, ready to rebrand and deploy.
| Repo | What it is |
|---|---|
| starter-kit | A white-label multi-tenant agent dashboard: sign-in, workspaces, per-user agents, native chat/files/integrations tabs. The fastest way to put Agent37 in front of your users. |
| custom-agent-image | A working agent app whose agents run an image it builds itself: one Dockerfile, one skill folder, one command to publish. |
| pi-agent-image | The pi coding agent on Agent37: managed LLM + Composio integrations, with no API keys in the image. |
| hermes-byo-model | Agents running on your own OpenRouter key: per-agent tokens, a money proxy, and real per-agent spend. |
| hermes-openclaw-composio | Hermes and OpenClaw agents on your own Composio: your users' connected accounts (Gmail, Slack, Notion, 1,000+ apps), per-agent usage metering. |
The full API reference lives at agent37.com/docs. If you are using a coding agent (Claude Code, Codex, Cursor), point it at agent37.com/docs/llms-full.txt and it can build against the current spec directly.