Want a few “persona + files + reachable from WeChat / Feishu (Lark)” AI bots on your own PC — without locking into one vendor subscription? AnyModelBot is an open-source, local-first desktop AI assistant (MIT): one Modelflare key claims access to 70+ models, with background research / scheduled rules and three file/command permission tiers. Repo modelflare-ai/AnyModelBot (~425★ when checked), site anymodel.bot, latest v0.1.15 (2026-11-01) ships Win x64/arm64 exe, macOS dmg, Linux AppImage. Researched November 2, 2026.
At a glance
| Item | Details |
|---|---|
| Type | Open-source local-first desktop AI agent (IM-ready) |
| Version | v0.1.15 (2026-11-01) |
| License | MIT |
| Stars | About 425 when checked |
| Packages | Win x64/arm64; macOS arm64/x64 dmg; Linux x64/arm64 AppImage; Docker also listed |
| Models | Pay-as-you-go via Modelflare; the app itself has no subscription |
| Channels | In-app chat; WeChat / Feishu (Lark) / Telegram / Slack / Discord; Notion hook |
| Entry | Site · Releases |
What it can do — and the edges
Each bot gets a persona, default model, workspace and memory. It can read/write files, run commands in a sandbox, and shows an approval card before leaving the workspace. Three permission tiers: read-only / workspace (default — ask before leaving) / full access (no sandbox; recursive deletes still prompt). Background jobs and scheduled rules can research, tidy, and fire reminders; a “needs you” sidebar collects approvals. Desktop control uses a built-in Cua Driver — Beta on macOS.
Models: sign in or paste a Modelflare key; one picker covers many vendors and can switch mid-conversation; pick international or China mainland routes as needed. Chats, memory, files and keys stay on the machine by default; prompts / tool results sent to a model still transit Modelflare to the provider (the privacy note is blunt about that). Still early (v0.1.x) — features move fast with each Release.
How it sits next to tools on this site
| Product | Stronger when you care about… |
|---|---|
| PawWork (EN pending) | Pick a folder + plain-language office-file work (DSH pack) — finish sheets / receipts in one workspace |
| Dashou / 搭手 (EN pending) | Alt+Space launcher + visual action chains (closed source) — hotkey search and RPA pipelines |
| Abu | Open-source desktop office AI: read files / run commands / emit Office docs — Cowork-style one-shot tasks |
| Cindy | Ready harness for Claude Code / Codex across devices — already live in a Claude Code workflow |
| AnyModelBot | One-session multi-model switching + WeChat/Feishu IM entry + Super-Agent-style background / schedules — always-on bots, IM reach, tight permissions |
This is not another “also chats about files” rewrite: the gap is multi-model one-click switching + chat apps as the front door + background tasks / rules. For a browser-sidebar agent see Youxiaohou (closed-source credits).
Who it is for (and caveats)
Good fit: want always-on local agents; need a WeChat / Feishu ping; prefer swappable pay-as-you-go models over locking a monthly vendor first.
Caveats: “full access” or desktop control hands the keyboard to the model — practise on workspace tier first; macOS desktop control is still Beta; model calls leave the machine; do not expose Docker admin ports to the public internet; verify installers with SHA256SUMS on the Release. Version and package names follow the November 2, 2026 GitHub Releases / site check.
Entry: anymodel.bot · Releases v0.1.15 · Source.
