convez.ai
AgenteAutomaçãoIntermediário
yogasw

Wick

yogasw/wick

Agentes IA em Go — rode Claude/Gemini em Slack/Telegram ou crie tools internas

Ver no GitHub
BomPluginScore · de 0 a 100

Framework Go para construir agentes IA internos e ferramentas customizadas, ou executar Claude/Codex/Gemini como bot em Slack, Telegram e web. Oferece controle total do código com auditoria de comandos via Command Gate.

Instalação

$npx convez-ai yogasw/wick

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
4
Forks
1
Acompanhando
0
Saúde
57%
Último commit
há 1 dia
Versão
v1.18.1
Issues abertas
0
Contribuidores
7

Casos de uso

  • Hospedar agentes IA em Slack/Telegram sem dependências externas
  • Construir ferramentas internas com IA escrevendo código Go real
  • Criar jobs em background e conectores personalizados auditados
  • Executar agents locais com whitelist e aprovação interativa de comandos

Veredito

Wick é uma solução sólida para quem precisa de agentes IA auditados e controlados em infraestrutura própria. A capacidade de hospedar binárias localmente e o Command Gate oferecem segurança superior ao SaaS, ideal para ambientes corporativos ou sensíveis.

README

A tradução para o português deste README ainda está na fila.

Wick

Build internal tools and AI agents in Go — or just download and run Claude / Codex / Gemini as a Slack + Telegram + web agent host. No copy-pasting. You own the code.

Two ways to use wick:


1. Run AI Agents — no Go, no framework

Want Claude / Codex / Gemini as a Slack bot, Telegram bot, or web assistant? Just download the binary.

# Linux / macOS / Termux — auto-detects OS + arch
curl -fsSL https://yogasw.github.io/wick/install.sh | sh
wick-agent server   # web UI at http://localhost:9425 — creds auto-generated on first run
# Windows
iwr -useb https://yogasw.github.io/wick/install.ps1 | iex
# Docker — single-container: HTTP + cron in one process
docker run -d \
  -p 9425:9425 \
  -v wick-agent-data:/root/.wick-agent \
  ghcr.io/yogasw/wick-agent:latest all

The binary supports two modes — pick one:

ModeCommandBest for
System traywick-agent (no args)Desktop — right-click menu, icon shows state, auto-start on login
Headlesswick-agent serverRemote server / Docker — no GUI, logs to stdout

Then in the web UI (/tools/agents):

  1. Providers — point wick-agent at your Claude / Codex / Gemini binary and your PAT
  2. Channels — connect Slack (Socket Mode), Telegram bot, or just use the built-in Web UI
  3. Workspaces — pick a folder for the agent to work in (a default is created automatically)
  4. Send a message → wick-agent spawns the agent and routes the conversation

Every Bash command the agent runs goes through the Command Gate — whitelist globs or escalate to interactive 4-mode approval (Approve once / This session / Always / Block), audited to JSONL.

→ Agent host docs


2. Build Internal Tools & Jobs — AI writes real Go files

# Install wick CLI (needs Go installed)
go install github.com/yogasw/wick@latest

wick init my-app
cd my-app
wick dev   # http://localhost:9425

The install.sh / install.ps1 scripts above install wick-agent (the runtime binary), not the wick CLI used for scaffolding. For development you need go install.

Open my-app/ in Claude Code and prompt what you need:

add a tool called "base64" that encodes and decodes text
add a background job that syncs data from our API every 30 minutes
add a connector for GitHub with list_repos and create_issue operations

Claude writes real Go files in your repo — you own everything. git diff to review, git revert to undo.

What wick scaffolds

my-app/
├── main.go          # register tools, jobs, and connectors here
├── AGENTS.md        # Claude reads this — framework conventions
├── .claude/skills/  # bundled AI skills (tool-module, job-module, connector-module)
├── wick.yml         # task runner config
├── tools/
│   ├── convert-text/   # example tool (UI page)
│   └── external/       # external link cards
├── jobs/
│   └── auto-get-data/  # example background job
├── connectors/
│   └── crudcrud/       # example connector (LLM-facing via MCP)
└── tags/
    └── defaults.go     # shared tag catalog

What the framework handles

You write business logic. Wick handles everything else:

You writeWick provides
func Register(r tool.Router)Admin UI, tag-based access, runtime config editing
func Run(ctx) (string, error)Cron scheduler, job history, run/retry UI
func Operations() []connector.OpMCP endpoint, per-call audit log, OAuth + PAT auth

Module types

TypeAudienceEntry pointURL
ToolHumans (web UI)Register(r tool.Router)/tools/{key}
JobSchedulerRun(ctx) (string, error)/jobs/{key}
ConnectorLLMs via MCPOperations()/mcp
AgentsSlack / Telegram / Webbuilt-in/tools/agents

Common commands

CommandWhat it does
wick devGenerate templ + CSS, start server
wick serverStart HTTP server only
wick workerStart background job worker
wick allHTTP + cron in one process (single-node deploys)
wick buildCompile binary with version metadata
wick testRun go test ./... with coverage
wick skill syncRefresh bundled AI skills after upgrade

→ Framework docs · AI Quickstart


Contributing

See CONTRIBUTING.md or the Contributing guide.

License

MIT

Ver o original no GitHub

Bom para

Relacionados