AnyModelBot is a free, MIT-licensed desktop AI agent that runs on your own machine rather than in the cloud. It installs on Windows, macOS, and Linux, or runs via a single Docker command on a NAS or server, and it connects to a wide catalog of mainstream models through a single Modelflare key. Instead of juggling subscriptions, you pay only for the tokens you actually use.
What it offers
The product centers on a set of plush "coworker" bots, each with its own persona, default model, and workspace. You can create assistants for general tasks, customer replies, daily reports, coding, writing, or translation and computer operation, then switch models mid-conversation whenever a different one fits better.
- Computer operation: The agent can read the screen, open apps, read and write files, and run commands, with a limited set of 24 exposed actions and workspace-scoped writes that require your approval when they go out of bounds.
- Long-term memory: Bot, group, and team memories are stored as plain Markdown files on your disk, with automatic rejection of passwords, keys, and ID numbers, plus a context manager that tidies long conversations using a cheaper model you choose.
- Chat channels: After a QR-code login, the same bot is reachable in WeChat or Feishu, so you can assign work from your phone while the desktop instance keeps running.
- Permissions and cost controls: Read-only, workspace, and full permission tiers are set per bot, and the app shows balance, low-balance alerts, daily usage, and per-bot spending caps.
Who it is for
AnyModelBot suits individuals and small teams who want a persistent personal agent that handles repetitive digital chores — organizing invoices, drafting meeting notes, running scripts, translating pages, or answering support questions — without sending their files and memory to a third-party SaaS. Because it is open source and self-hosted, it also appeals to users who want to inspect the code, choose their own model vendor, and keep data on their own hardware.
How it differs
Unlike subscription-bound assistants tied to a single vendor, AnyModelBot is model-agnostic and free to run, with usage billed per token. It also stands apart from purely conversational tools by acting on the desktop under visible, confirmable steps, and by keeping memory as editable Markdown rather than a locked database. A local web console, Docker deployment, and optional 24/7 hosting round out the setup.
