gawkbot

// Overview

gawkbot (published earlier as Wuphf) turns a described workflow into a running bot. You say what the job is in a sentence — or demo it once on a call — and it builds the bot that runs it: a real screen, its own schedule, its own tools, backed by more than 1,200 integrations. Everything runs on your own machine, on your own accounts, driven by whichever coding-agent runtime you already pay for.

Each bot ships with six parts rather than a chat box: an app with a live-built UI that reads and writes real workspace data; routines, which are versioned scheduled prompts with run history and a transcript per run; self-authored tools you teach it in chat; wiki-style knowledge pages where every claim is cited back to its source; typed data tables plus shared integrations; and an approval gate. Reads are free, but every write — every send, commit, purchase and delete — is held until you approve it, after which the bot runs continuously. If a workflow names a system that is not connected yet, gawkbot asks rather than silently re-scoping the job.

The runtime is yours to pick: Claude Code by default, or Codex, opencode, Ollama, Hermes Bot and OpenClaw gateways through `--provider`, including any OpenAI-compatible endpoint such as a LiteLLM or local proxy. Memory is built in with no backend to choose — bot knowledge, run transcripts and the shared company brain live as local files under `~/.wuphf/` that you can read directly. gawkbot is written in Go with a Bun-built web UI, and is published under the Sustainable Use License.

// What it does

• Describe a workflow in one sentence and watch the bot get built live, landing on a working microapp rather than a chat thread
• Routines: versioned scheduled prompts ("Every Monday 9:00") with run history and a per-run transcript
• Approval gate on every outbound action — reads run freely, writes wait for your tap, then the bot runs around the clock
• 1,200+ integrations plus per-bot typed tables; connect a system once and every bot shares it
• Cited knowledge pages so you can check the source behind anything a bot claims
• Runtime-agnostic: Claude Code, Codex, opencode, Ollama, MLX-LM, Hermes Bot or an OpenClaw gateway, with no cloud key required for the local options

// Getting started

gawkbot runs locally and opens a web UI. You need one signed-in bot CLI first — Claude Code by default, or Codex CLI / opencode; the first-run screen verifies the runtime before anything else happens.

npx gawkbot

npm install -g gawkbot && gawkbot

WUPHF_OLLAMA_BASE_URL="http://127.0.0.1:20128/v1" \
WUPHF_OLLAMA_MODEL="openai/gpt-5.4-mini" \
gawkbot --provider ollama --no-open

git clone https://github.com/najmuzzaman-mohammad/gawkbot.git
cd gawkbot
cd web && bun install && bun run build && cd ..
go build -o gawkbot ./cmd/wuphf
./gawkbot

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Turn a recurring manual workflow into a scheduled bot with its own screen, without writing the integration glue yourself
• Keep always-on automation on your own machine and your own accounts instead of a vendor's cloud
• Put a human approval gate in front of every write an agent makes to email, git or a payment system
• Reuse a coding-agent subscription you already pay for — or a purely local model — as the runtime behind business automation

// How gawkbot compares

gawkbot alongside other open-source workflow automation tools AI/TLDR tracks, ranked by GitHub stars.