AiDog

One line: AiDog (aidog.im) is a personal AI companion that runs entirely on your own machine — it talks to you over Telegram, Feishu, DingTalk and other channels, ships with a skill system, natural-language cron jobs and a self-evolving memory; your credentials never leave your computer and there is no cloud account.

What you get

• A resident local assistant: loopback web console (127.0.0.1:19167) plus messenger channels
• Skills with a deterministic router, subagents, natural-language scheduling (/cron), memory evolution (/learn, /reflect)
• Completely free, no telemetry (optional anonymous usage counting is off by default) — you only bring a model endpoint

Before you start (5 minutes)

⚠️ Windows trap: the repository must be cloned onto WSL's own filesystem (e.g. ~/aidog), never onto a mounted C:\ drive — 9p/CIFS mounts reject chmod and break tmux; copying a Windows-side clone in also breaks the .sh scripts with CRLF line endings.

Install

Route A: install with AiBOX (recommended)

1. Open AiBOX → app directory → AiDog → Install
2. AiBOX downloads the hash-verified official source archive (accelerated automatically in CN), unpacks it to ~/aidog inside WSL and runs the official one-command installer deploy/install.sh (provision + boot + verify; safe to re-run)
3. When the installer reports, continue with configuration below

Route B: manual install (official commands)

WSL / Linux (same on macOS):

# Install tmux first (requires admin, do it manually)
sudo apt-get install tmux          # Ubuntu/WSL; on macOS use brew install tmux

git clone https://github.com/AiDog-im/AiDog.git aidog && cd aidog
bash deploy/install.sh             # one command: provision + boot + verify
aidog setup                        # guided wizard: fills only the missing slots

Configuration (make it talk)

aidog setup only prompts for slots that are still empty (hidden input, values are never echoed):

1. WEB_TOKEN: a console write token — any long random string you invent
2. Model endpoint: in the console Settings or the wizard, set an OpenAI-compatible base_url + model (default is the local gateway http://127.0.0.1:3460/v1, model mimo-v2.6-flash; fill MODEL_API_KEY only if your provider needs a key)
3. Telegram (optional): a token from @BotFather + your numeric user id (allowlist — strangers are rejected)

Verification: what "success" looks like

1. aidog setup --check exits 0 (fully ready) or 2 (only credentials pending) — never 1
2. Open http://127.0.0.1:19167, send a message in the chat card — a streamed reply means the model chain is live
3. Message your bot /help from an allowlisted Telegram account and get a reply
4. python3 ~/aidog/deploy/aidog_doctor.py: zero FAILs

FAQ

• tmux not found: install tmux manually (admin required, see above), then re-run bash deploy/install.sh
• port 19167 is used: boot on another port: AIDOG_WEB_PORT=19200 bash deploy/boot_services.sh start
• Stays PENDING / model endpoint unreachable: the brain isn't reachable yet. Make sure mycore.base_url in web/settings.json responds; exit code 2 with nothing configured is an honest "not configured yet", not a crash
• Telegram messages not arriving: often blocked egress — the runtime routes bot traffic through a local proxy (HTTPS_PROXY); set it in the shell that starts the tmux sessions
• aidog update apply refuses with a dirty tree: by design — commit local changes before upgrading

Uninstall

bash ~/aidog/deploy/boot_services.sh stop   # stop the runtime first
rm -rf ~/aidog                              # machine-local state lives beside the code — back up secrets.env and data/ first
rm -f ~/.local/bin/aidog                    # remove the CLI wrapper

More: official installation docs · aidog.im