Finn: automation for Hermes Agent

AI companion that lives in iMessage, with background workers, Patterns automations and a web dashboard

What Finn does

Finn is built as a companion you text rather than an app you open. A dedicated hot-path agent owns the live thread and replies immediately, while research, files and multi-step tasks run on background workers that report back. Workers use a code mode: a sandboxed JavaScript runtime exposed through workspace_search and workspace_execute over a typed finn object, instead of dozens of native tools. Tools and docs load on demand, history is compacted as it grows, and large tool output spills into artifacts.

Long-term memory goes through a provider, with Honcho recommended, and recall is sanitized before it reaches the chat. Patterns are scheduled or connector-triggered automations with run history. Connectors come through Composio, MCP and the Puter macOS app, and a web dashboard covers profile, connectors, Patterns and My Day. The recommended models are Opus 4.6 or 4.8 on the hot path and Kimi 2.6 for background work. Setup uses Docker Compose with bundled Postgres and needs a public HTTPS URL.

Key features

Hot-path agent for instant replies plus background workers for slow tasks
Code mode with a sandboxed JavaScript runtime over a typed finn object
Memory through a provider, with Honcho recommended
Patterns: scheduled and connector-triggered automations with run history
Connectors through Composio, MCP and the Puter macOS app
Tenant and user-scoped runtimes with isolated workspaces and allowed-number access

When to use it

Running a personal assistant you reach by iMessage
Setting up an assistant for friends and family without one shared context
Scheduling recurring automations that report back in the chat

Who it is for: People who want a self-hosted, iMessage-based AI companion and are comfortable running Docker Compose behind a public HTTPS URL.

How it fits with Hermes Agent

A standalone companion app whose repository is tagged hermes-agent, hermes and openclaw. The visible README does not describe how Hermes Agent is used.

How to install Finn

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

cp .env.example .env
docker compose -f docker-compose.no-cloudflared.yml up -d --build

Requirements: Docker Compose, a public HTTPS URL set as PUBLIC_URL, and a HONCHO_API_KEY for the recommended memory provider

Note: It needs a public HTTPS URL, and a localhost-only setup will not fully work.

FAQ

What is Finn?

Finn is an AI companion that lives in iMessage. It answers quickly in the thread, runs slower work on background workers, and keeps memory behind a boundary so it does not announce that it is checking memory.

Does Finn work with Hermes Agent?

The repository is tagged hermes-agent and hermes, but the opening part of the README does not explain a Hermes integration. Check the repository docs before relying on it with Hermes.

What do I need to run Finn?

You need Docker Compose, a public HTTPS URL because iMessage, the dashboard and webhooks depend on it, and a Honcho API key if you use the recommended memory setup. A tunnel can provide the URL.